Indicator-Based Methodology for Assessing EV Charging ... - MDPI

energies Article

Indicator-Based Methodology for Assessing EV Charging Infrastructure Using Exploratory Data Analysis Alexandre Lucas *, Giuseppe Prettico, Marco Giacomo Flammini, Evangelos Kotsakis, Gianluca Fulli and Marcelo Masera European Commission, JRC, Directorate C Energy, Transport and Climate, PO Box 2, 1755 ZG Petten, The Netherlands; [email protected] (G.P.); [email protected] (M.G.F.); [email protected] (E.K.); [email protected] (G.F.); [email protected] (M.M.) * Correspondence: [email protected]; Tel.: +351-961741327 Received: 27 June 2018; Accepted: 16 July 2018; Published: 18 July 2018







Abstract: Electric vehicle (EV) charging infrastructure rollout is well under way in several power systems, namely North America, Japan, Europe, and China. In order to support EV charging infrastructures design and operation, little attempt has been made to develop indicator-based methods characterising such networks across different regions. This study defines an assessment methodology, composed by eight indicators, allowing a comparison among EV public charging infrastructures. The proposed indicators capture the following: energy demand from EVs, energy use intensity, charger’s intensity distribution, the use time ratios, energy use ratios, the nearest neighbour distance between chargers and availability, the total service ratio, and the carbon intensity as an environmental impact indicator. We apply the methodology to a dataset from ElaadNL, a reference smart charging provider in The Netherlands, using open source geographic information system (GIS) and R software. The dataset reveals higher energy intensity in six urban areas and that 50% of energy supplied comes from 19.6% of chargers. Correlations of spatial density are strong and nearest neighbouring distances range from 1101 to 9462 m. Use time and energy use ratios are 11.21% and 3.56%. The average carbon intensity is 4.44 gCO2eq /MJ. Finally, the indicators are used to assess the impact of relevant public policies on the EV charging infrastructure use and roll-out. Keywords: charging infrastructure; electric vehicles; service ratio; network design; exploratory data analysis

1. Introduction The global stock of electric vehicle (EV) cars surpassed three million vehicles in 2017 after crossing the one million unit mark in 2015 and the two million mark in 2016. With a 39% sale share, Norway has undoubtedly achieved the most successful deployment of EVs in terms of market share globally, followed by Sweden and the Netherlands, at 6.3% and 2.7% respectively. This also corresponds to the highest car stock in Europe of over 178,000 in Norway, followed by the United Kingdom and the Netherlands with 133,000 and 119,000, respectively [1]. More than half of global sales of electric cars were in China, which corresponds to a market share of 2.2% in 2017. In 2017, China had an EV stock of 951,000, followed by the United States, with 401,000. Just 20 cities account for about 40% of the world’s EVs [2]. Strong regulatory and fiscal policy has driven the early electric vehicle market. The markets with the highest electric vehicle uptake are the following: China, Europe, Japan, and the United States, which have implemented a combination of vehicle CO2 or efficiency regulations, strong consumer

Energies 2018, 11, 1869; doi:10.3390/en11071869

www.mdpi.com/journal/energies

Energies 2018, 11, 1869

2 of 18

incentives, and direct electric vehicle requirements. Charging infrastructure support to consumers is also a common characteristic of these markets. The main concerns addressed in the literature regarding the charging infrastructure deployment are related to cost, charging effectiveness, and ability to satisfy dynamic demand, as well as its overall environmental impact. According to the EAFO (European alternative fuels observatory) [3], the ratio of cars per charger vary widely from country to country, from 66 to 3.7 cars per charger in Iceland and Spain, respectively. Even though at first sight, factors such as the driven distance per year, and population size and density, do not change proportionally. There are no commonly accepted goals or standards for charging infrastructure density, either on a per-capita or per-vehicle basis. Different countries seem to follow different sizing and options for the public charging infrastructure development. This depends on many factors. There is no clear way to achieve an efficient deployment of EV charging infrastructure and the associated policy that will need to be addressed to help pave the way for electrification. A study on the emerging best practices for EV charging infrastructure [4] provides insights into the differences between present infrastructure roll outs. Using a multivariable regression of 350 metropolitan areas, the authors find that both level 2 and direct current (DC) charging are linked to EV acceptance, as are consumer purchase incentives. Yet the significant charging variability across hundreds of cities points to major differences across the EV markets regarding the role of public charging. An economic-based approach is asserted by other authors. A study [5] assesses factors influencing EV adoption across 30 countries in 2012 at a national level, focusing primarily on financial incentives. A regression of several variables, with EV market share as the dependent variable, shows that a charging infrastructure is the best predictor of national EV market share. Nonetheless, there are exceptions to this trend, such as Israel and Ireland, with relatively extensive charging infrastructure and low EV sales shares. From a cost point of view, other authors [6] have modelled the impact of several factors, including charging infrastructure, on EV market share in Europe. This model calculated the utility and respective market share of different powertrain types, using feedback loops to capture realistic decision-making patterns by drivers, manufacturers, charging infrastructure providers, and policymakers. The model also assesses the profitability of charging stations under various scenarios and considers subsidies and government targets for charging infrastructure. The authors found that the private market can profitably support 95% of public charging stations, up to a ratio of 25 electric vehicles per charge point. They also found that EV market share increases as the electric vehicle/charge point ratio decreases from 25 to 5 electric vehicles per charge point. Charging infrastructure availability also appears to have the strongest impact on uptake once EV stock share exceeds 5%, which is currently the case only in Norway. An environmental approach assessing the materials and the overall impact of the infrastructure is also a topic that is receiving increasingly more attention [7]. Analysing the complete supply chain and infrastructure materials for the Portuguese case, the authors conclude that EV supply infrastructure is more carbon and energetic intensive than conventional supply infrastructure. Considering a ratio of 9.5 and 333 vehicles per normal and fast charger, respectively, the contribution in overall vehicle life cycle analysis (LCA) may reach 7.9% of gCO2eq /km, identifying the chargers as the main contributors. This clearly puts a threshold onto the limits of chargers to be installed per vehicle with a risk of attaining higher environmental impacts than desired. Other studies address charging point allocation phenomena, presenting models or solving optimization problems [8–14]. Particular attention is given to the road segment allocation of chargers [8], privileging the use of existing infrastructure such as fuel stations and rest areas or parking lots. There seems to somehow be a tendency to separate electrical studies from their geospatial distribution. Other optimization problems [15–19] focus on case studies trying to find the optimal ratio of chargers per car and distribution intensity. No common approach is followed; different objective functions and variables are considered, but tend to be the following: charging costs, securing a minimal distance from a charger to a given point on a map, distance between chargers, relation to population density, or relation with fuel stations. The literature [20] compares rollout strategies in the Netherlands, demand driven or strategically driven. Demand driven charger installations are considered the ones placed on demand by EV users for home charging. Strategic installations are largely based on

Energies 2018, 11, 1869

3 of 18

expected use at points of interest or highly populated areas. From the literature, one may conclude that there is neither a common benchmark for the number of electric vehicles per public charge point nor a common approach to the infrastructure roll-out. Given the wide variation of public charging availability across markets with higher EV adoption, and housing differences, as well as the population density characteristics, it seems clear that there is no ideal global ratio for the number of EVs per public charge point. Comparisons of similar markets still offer an instructive way to understand where and how charging is insufficient. For this reason, there is a need to formulate representative indicators providing evidential feedback that can be used to identify performance gaps and best practices. In this paper, we describe the methodology to obtain such indicators using exploratory data analysis. Given the Dutch advanced implementation regarding charging infrastructure and data availability, this study focuses on a dataset provided by one of the main organisations in the field of smart charging infrastructure, ElaadNL [21]. Despite the data format, the indicators developed should be simple, reliable and effective in quantifying the potential impact on given any similar dataset. Charging Infrastructure Programs In terms of public infrastructure, Belgium, Norway, and Sweden had a ratio of 1 publicly accessible charging outlet for just over 20 electric cars in 2017. Both of these values are significantly disproportional from the global average of one charger per seven electric cars observed in 2017. Another ratio guidance comes from the Alternative Fuel Infrastructure (AFI) Directive of the European Union (EU) [22], which recommends a ratio of one publicly accessible charger per ten electric cars. Countries with densely populated cities such as China and Japan already have a stronger publicly accessible EV infrastructure than others, with one outlet for six electric cars in China and one outlet for seven electric cars in Japan. About 43% of electric vehicle sales in 2016 were registered in only eight countries across Asia, Europe, and North America [4]. An overview of the public infrastructure programs of these main markets is provided below. There are two main programs in China: (i) city government-funded construction in pilot cities, and ii) regional investments by automakers and State Grid national charging corridors. This last one intends to construct fast charging plazas within cities and intercity corridors, aiming at 120,000 fast charging stations and 500,000 total public stations by 2020 [23]. The mechanisms of support come from mainly state owned utility programs, public–private partnerships, and grants to local governments. In Japan, the main program was the Next Generation Vehicle Charging Infrastructure Deployment Promotion Project to fund charging stations around cities and highway rest stations in 2013–14 [23]. Apart from this, the Development Bank of Japan also partnered with automakers and power company TEPCO to construct the Nippon Charge Service (NCS), a nationwide network of charging stations. Almost 7500 stations are now part of this network, with continued funding at least through 2018. Main mechanisms of support are grants to local governments and highway operators as well as public–private partnerships. In Europe, charging infrastructures have been constructed by a combination of private charge point providers, power companies, automakers, and governments, primarily at the national and city levels. The main policy document is the AFI Directive [22]. The EC has supported more than a dozen electric vehicle infrastructure projects through the TEN-T/CEF-T program, with a focus on creating trans-European corridors and linking the projects operated by Member States (MS) [24]. Moreover, the EU has promoted interoperability, open standards, and smart charging, as demonstrated for example by the Green eMotion research project [25]. At an MS level, in France, local governments may apply for grants with main programs involving the funding given to 3000 cities for deploying 12,000 charge points. In Germany, the program involves €300 million for 10,000 Level 2 and 5000 DC fast charging stations with subsidies for 60% of costs for all eligible businesses. In the Netherlands, much of the early construction of charging infrastructure was initiated by ElaadNL [21], which maintains and upgrades about 3000 stations. Support has been provided through contracts tendered to businesses on a project-by-project basis. The federal government also provided €16 million in incentives for charging

Energies 2018, 11, 1869

4 of 18

infrastructure through their “2011 Electric Mobility Gets Up to Speed” program. More recently, the federal government consolidated various programs and began to promote charging stations through its “Green Deal” [4]. The key sponsor of Norway’s charging infrastructure has been Enova [26], an agency funded through petroleum and natural gas. It has been working with grant schemes from 2009 onwards with quarterly calls for proposals for targeted projects. In Sweden, the government introduced two investment support schemes (Climate Leap and Urban Environment Agreements), with the goal to promote charging infrastructures. Public normal charging is developed by many actors, both public and private. The public network has reached approximately 3600 charging points distributed over 1060 charging stations [26]. In the United States, much of the initial investment in charging infrastructure came from the American Recovery and Reinvestment Act of 2009, which has provided federal funding through the EV Project and the U.S. Department of Transportation’s Transportation Investment Generating Economic Recovery program, among many infrastructure projects in the United States from 2010 to 2013. According to the Department of Energy (DoE) [4], there were over 18,000 public Level 1 and 2 and DC fast electric charge points installed. 2. ElaadNL Dataset The main tool for data processing and statistical analysis used for developing the evaluation framework is R studios Crystal Ball [27] was used for uncertainty estimation and statistical distribution fit. Regarding geospatial analysis the open source, Q-GIS 2.18 has been used and the Geographic Resources Analysis Support System GIS, GRASS 7.4, which is an open source geographic information system providing powerful raster, vector, and geospatial processing capabilities. The EV dataset was obtained by ElaadNL, a Dutch smart charging service provider, and contains the records of the charging stations’ utilisation. It includes historic transactions, from January 2012 until March 2016, of 1747 charging stations (2900 charging points) installed over the entire geographical area of the Netherlands. The charging stations are all three-phase with a maximum output power of 12 kW. In the year 2015, the dataset represented 16% of all the charging stations installed in the Netherlands [3]. The dataset has approximately one million charging events (transactions) over a four-year period, and for each transaction, the parameters are registered with the corresponding timestamp using Universal Time Coordinated (UTC) units and updated every 15 min (the dataset has around 32 million observations). From the transactions identifier (user’s card), it has been possible to estimate 53,849 EV drivers who have used the charging stations during the four-year observation period. The installation of the charging stations was, however, phased in time and only in the beginning of 2014 could the whole network be completed. For this reason, and to ensure higher volume of transactions and a higher number of the chargers, the years 2014 and 2015 were used. After removing the null lines with no data or error, a total of 1683 chargers remained, corresponding to over 96% of the initial dataset. The total number of transactions is 733,794, which represents approximately 70% of the initial total number of transactions in the whole dataset. The numbers of individual cards are assumed to be EV owners (42,400 EVs). The original dataset consists of a set of 19-tuple elements: “Index”, “Transaction Id”, “Charge Point”, “Connectors”, “UTC Transaction Start”, “UTC Transaction Stop”, “Meter Start”, “Meter Stop”, “Start Card”, “Stop Card”, “Connected Time”, “Charge Time”, “Idle Time”, “Total Energy”, “Max Power”, “Latitude”, “Longitude”, “Street”, and “Road Segment”. As the goal of the study is to focus on the charging columns, the data has been grouped by chargers (Charge Point), charging event data was discarded, such as UTC Transaction Start/Stop, Meter Start/Stop or Card Start Stop, and the remaining corresponding information is aggregated. This originated the following tuple: “Charge Points ID”, “Sum of Energy”, “Sum of Idle Time”, “Sum of Connected Time”, “Mean of Max. Power”, “Max. Power”, “Sum of Charge Time”, “Latitude”, “Longitude”, “Road Segment”, and “Number of Connectors”. From this stage of the dataset, other indicators useful to the indicators analysis were generated and given by Equations (1)–(3). Such parameters may be derived from other dataset sources depending on the dataset composition

Energies 2018, 11, 1869

5 of 18

under analysis. The first one is the energy use ratio of a given charger over the observed period. This corresponds to the energy actually provided in a given period by each charger divided by the energy that it could have been provided at its nominal power rate in the same period of time. T0 is the observation time in hours. In this analysis, T0 = 2 × 365 × 24 = 17,520 h for a two-year period. EnergyUse =

Sum_Energy (connectorsx T0 × meanMaxPower)

(1)

The use time ratio in Equation (2), is given by the total time (expressed in hours) during which a charger is occupied by a vehicle (connected time, that is, charge time plus idle time), divided by the number of hours of the observation period, multiplied by the number of connectors. Usetime_ratio =

Sum_Connectedtime (T0 × connectors)

(2)

Since the use time ratio indicator provides the percentage of use, where 1 corresponds to 100% of usage of the chargers, the availability factor will help analyse this indicator and can be obtained by Equation (3) as follows: Availability_Factor = 1 − Use _ratio (3) 3. Analysis of Indicators Across the world’s main economies, the targets for the electrification of the mobility sector are being set with a tendency to achieve ambitious goals by 2020 to 2030. In the EU, despite an agreement on time horizons, MS seem to have a different pace in implementation. The “trilogue” between the European Commission (EC), Parliament, and Council, often results in non-binding targets, where numbers are scarce as is the case of the AFI Directive [22]. According to the comments to the directive [28] “the completeness, coherence, and ambition of the National Policies Frameworks (NPFs) vary greatly”. The adoption of a reliable set of indicators will contribute to a more solid benchmarking procedure and to supporting future initiatives. A crucial factor that influences the infrastructure sizing and design is its energy demand. To estimate this factor, several aspects need to be taken into account, among them, the total amount of energy required by the fleet and the residential structure, such as urban densification and private parking/garage per car. The first one can be estimated by an average driven distance per car and an average consumption per km. The second one will provide an estimation of how much of the charging on average is done at home. The remaining energy to be fulfilled should provide a good estimate of the energy required from public chargers. However, to acquire and treat such information is not so straightforward if we consider the different realities across markets, urban, semi-urban, and rural areas. Hence, we follow a bottom up approach, starting from real implementation data in a developed market and using information from openly available external datasets. 3.1. Energy Demand from the Network The willingness to charge at home or on a public network depends on the residential characterization. EV owners in California, for example, more frequently have access to home and workplace charging, and one public charger per 25 to 30 electric vehicles is typical. On the other hand, in the Netherlands, private parking and charging are relatively rare, and one public charger per 2 to 7 electric vehicles is typical [4]. Hence, the levels of support by such infrastructure should derive from a functional unit of chargers/MJ or equivalent use and not simply chargers per car. It is in fact misleading to compare the number of cars per chargers (or connector) if the network is used differently. Consider two networks with the same number of vehicles and efficiency, with the first one having a 10 EV/charger ratio and the second a 5 EV/charger ratio. If, on average, a car charges 10 kWh per week on the first network and 20 kWh on the second one, even though the second one is larger in terms of number of charging stations

Energies 2018, 11, 1869

6 of 18

per EV, both networks are dimensioned to provide the same amount of energy (1 charger per 100 kWh). Based on this reasoning we thus pursue an energy focused analysis of the network. From the dataset under Energies 2018, 11, x 6 of 19 analysis, the two-year average energy demand is 6.25 × 103 MWh, which means 11.25 × 106 MJ/year. 3 MWh, Allocating these numbers infrastructure results in an× energy demand of 6685.72these MJyear /chargertoand energy demand is 6.25 × to 10the which means 11.25 106 MJ/year. Allocating numbers the 4040.23 MJ /connector. year results in an energy demand of 6685.72 MJyear/charger and 4040.23 MJyear/connector. infrastructure Yet Yetaamore moredetailed detailedpicture picturecan can be be extracted extracted from from the the EVs EVs dataset dataset regarding regarding charging charging behaviour behaviour and frequency. Considering only the vehicles charging more than 24 kWh in two years (20,069), and frequency. Considering only the vehicles charging more than 24 kWh in two years (20,069), we we focus on the number of chargers used by these As can be observed in Figure 1a, more focus on the number of chargers used by these EVs.EVs. As can be observed in Figure 1a, more thanthan 83% 83% of are EVscharging are charging less200 than 200onkWh on ElaadNL’s infrastructure. a reference EV of EVs less than kWh ElaadNL’s infrastructure. Taking a Taking reference EV efficiency efficiency of 0.15 kWh/km, this implies that on average each EV is driven less than 1333 km per year, of 0.15 kWh/km, this implies that on average each EV is driven less than 1333 km per year, only 10% only 10% compared to the average current average of 13,000 km annual distance the Netherlands. compared to the current of 13,000 km annual drivendriven distance in theinNetherlands. The The remaining charging needs are satisfied by other networks or home/work chargers. Further remaining charging needs are satisfied by other networks or home/work chargers. Further considerations energy demand patterns. Figure 1b show the energy (%) that considerationsmay maybe bedrawn drawnregarding regarding energy demand patterns. Figure 1b show the energy (%) drivers retrieve from the three chargers. Evidently, the majority drivers of have a preference that drivers retrieve from themost threeused most used chargers. Evidently, the of majority drivers have a to charge ontoa charge restricted of chargers. More than 83% of the is requested preference on number a restricted number of chargers. More thantotal 83%yearly of theenergy total yearly energyto is three chargers. requested to three chargers.

Figure 1.1. Total Totalyearly yearlyenergy energycharged chargedby byelectric electricvehicles vehicles(EVs) (EVs) (a) (a)and andpercentage percentage of of energy energy drawn drawn Figure from the thethree threemost mostused usedchargers chargers(b). (b). from

3.2. Energy Use Intensity 3.2. Energy Use Intensity Whenever designing the charging network, both distribution and density of chargers should be Whenever designing the charging network, both distribution and density of chargers should taken into account. The electricity grid is of utmost importance, hence costly upgrades, potential be taken into account. The electricity grid is of utmost importance, hence costly upgrades, potential overburden in transformers, and cables capacity should be avoided. The public infrastructure in overburden in transformers, and cables capacity should be avoided. The public infrastructure in urban areas is more likely to have higher energy demand from mobility than semi-urban or rural urban areas is more likely to have higher energy demand from mobility than semi-urban or rural areas. This means more connection points tend to be deployed in areas with higher population areas. This means more connection points tend to be deployed in areas with higher population density. density. However, the service ratio (chargers/vehicle) should be sufficiently high outside urban However, the service ratio (chargers/vehicle) should be sufficiently high outside urban areas so as areas so as to secure a reasonable distance from a charger to every point in the map at which an EV to secure a reasonable distance from a charger to every point in the map at which an EV user may user may be located. In this section, we take advantage of the fact that the chargers’ locations are be located. In this section, we take advantage of the fact that the chargers’ locations are often given often given in the datasets and perform a geospatial analysis in two steps. The first one in the datasets and perform a geospatial analysis in two steps. The first one geographically analyses geographically analyses the energy use intensity to see if there is a homogeneous use of the network, the energy use intensity to see if there is a homogeneous use of the network, and the seconds one and the seconds one evaluates correlations to support its density distribution (charger’s intensity evaluates correlations to support its density distribution (charger’s intensity distribution). The energy distribution). The energy use intensity analysis of the network provides insights on how well it is use intensity analysis of the network provides insights on how well it is distributed. The analysis refers distributed. The analysis refers not only to its geographical distribution, but also to the energy not only to its geographical distribution, but also to the energy demand balance across the network. demand balance across the network. Figure 2 shows the locations of the chargers by road segment Figure 2 shows the locations of the chargers by road segment (a) and a heat map of energy demand (b). (a) and a heat map of energy demand (b). High energy demand is mostly concentrated in six areas. High energy demand is mostly concentrated in six areas. This corresponds to the main urban centres This corresponds to the main urban centres where in fact higher density of chargers exists, hence a where in fact higher density of chargers exists, hence a higher aggregated value of energy is observed higher aggregated value of energy is observed in the serving areas. in the serving areas.

Energies 2018, 11, 1869

7 of 18

Energies 2018, 11, x

7 of 19

Energies 2018, 11, x

7 of 19

Figure 2. (a) Chargers’ location by road segment and (b) heat map of energy demand during the total

Figure 2. (a) Chargers’ location by road segment and (b) heat map of energy demand during the total timespan under analysis (radius 10 mm). timespan under analysis (radius 10 mm). For instance, despite its high aggregated energy demand, in the Den Haag area, charging points Figure 2. (a) Chargers’ location by road segment and (b) heat map of energy demand during the total have a demand lower than the middle value of all the chargers in the dataset. This may be observed timespandespite under analysis (radius 10 mm). For instance, its high aggregated energy demand, in the Den Haag area, charging points in Figure 2b. This indicates that there are in fact many chargers, but with a relatively low energy use have a demand lower than the middle value of allaggregated, the chargers in the dataset. This may be observed per charger (greendespite points).itsSuch when result evidently in an charging energy intense For instance, highlocations, aggregated energy demand, in the Den Haag area, points in Figure 2b. This indicates that there are in fact many chargers, but with a relatively low energy use area. welllower be then, that chargers that the mostinare located necessarily in urban have It a may demand than thethe middle value ofare all used the chargers thenot dataset. This may be observed per charger (green points). Such locations, aggregated, result evidently in energy regions. To understand intensity of behaviour, awith statistical analysis can help.use A intense in Figure 2b.better This indicates thatthe there are inwhen factcharging many chargers, but a relatively lowan energy with quantile breaks is presented in Figure 3a, together with the corresponding Q-GIS area. It histogram may well be then, that the chargers that are used the most are not located necessarily per charger (green points). Such locations, when aggregated, result evidently in an energy intensein urban quantile mapwell in Figure 3b.thatthe may be then, the intensity chargers that used the most are not located necessarily in urban regions.area. To Itbetter understand ofare charging behaviour, a statistical analysis can help. regions. To better understand intensity of charging a statistical analysis can help. A Q-GIS A histogram with quantile breaksthe is presented in Figurebehaviour, 3a, together with the corresponding histogram with quantile breaks is presented in Figure 3a, together with the corresponding Q-GIS quantile map in Figure 3b. quantile map in Figure 3b.

Figure 3. Histogram of energy demand per charger (a) and geospatial distribution of quantiles (b).

The histogram is presented with quantile breaks, which means that each colour divides the range of the probability distribution into contiguous intervals with equal probabilities. One can Figure 3. Histogram of energy demand per charger (a) and geospatial distribution of quantiles (b). observe example,of that 25% of the chargers provided 1000 kWh,distribution or that 75% of chargers Figure 3. for Histogram energy demand per charger (a)below and geospatial of the quantiles (b). provided below approximately 5000 kWh in two years. The fact that the last quantile (in red) ranges The histogram is presented with quantile breaks, which means that each colour divides the from approximately 5000–25,000 kWh with frequency iswith a good indicator of concentration range of the probability distribution into significant contiguous intervals equal probabilities. One can The histogram is presented with quantile breaks, which means that each colour divides the range of energyfor demand. The figure the energy demand with 1000 a long red or tail. This canofbe observe example, that 25% shows of the chargers provided below kWh, that 75% theexplained chargers of the probability distribution into contiguous intervals with equal probabilities. One can observe for by the factbelow that only 19.6% of the total charging columns is responsible for 50% theranges total provided approximately 5000 kWh in two years. The(330) fact that the last quantile (in of red) example, that 25% of the chargers provided below 1000 kWh, or that 75% of the chargers provided energy supplied by the infrastructure under study. from approximately 5000–25,000 kWh with significant frequency is a good indicator of concentration By observing Figure 3b, itin is two visible that the points theThis last can quantile (inranges red) from below approximately 5000 years. The factcorresponding thatathe (inbered) of energy demand. The kWh figure shows the energy demand with longlast redtoquantile tail. explained also show geographical density identical to the one shown in Figure 2b. Therefore, both approaches by the fact that only 19.6% of with the total charging columns (330) for 50% the total approximately 5000–25,000 kWh significant frequency is isa responsible good indicator of ofconcentration of demonstrate that by there high concentration energy use both geographically and in the number of energy supplied theisinfrastructure under of study. energy demand. The figure shows the energy demand with a long red tail. This can be explained by charger. By observing Figure 3b, it is visible that the points corresponding to the last quantile (in red)

the fact that only 19.6% of the total charging columns (330) is responsible for 50% of the total energy also show geographical density identical to the one shown in Figure 2b. Therefore, both approaches supplied byChargers the infrastructure under study. 3.3. Intensity Distribution demonstrate that there is high concentration of energy use both geographically and in the number of Bycharger. observing Figure it is visible that the points to the last quantile A suitable way to3b, evaluate a given dataset is to try to corresponding look for correlations. The literature tends (in red) also show geographical density identical one shown in Figure 2b. Therefore, both approaches to support the concept that the intensity to andthe location of chargers should follow existing structures to 3.3. Chargers Intensity demonstrate that there is Distribution high concentration of energy use both geographically and in the number of charger. A suitable way to evaluate a given dataset is to try to look for correlations. The literature tends to support the concept that the intensity and location of chargers should follow existing structures to

3.3. Chargers Intensity Distribution A suitable way to evaluate a given dataset is to try to look for correlations. The literature tends to support the concept that the intensity and location of chargers should follow existing structures

Energies 2018, 11, 1869

8 of 18

Energies 2018, 11, x

8 of 19

to minimize costs, social impacts, resources, make of existing stations minimize costs, social impacts, andand resources, andand make use use of existing datadata [8]. [8]. FuelFuel stations and and parking lots are hence regarded as a proxy to estimate the agglomeration of demands. Further, parking lots are hence regarded as a proxy to estimate the agglomeration of demands. Further, the the population density seems to be a comparison factor whichthe thesize sizeororintensity intensityof of the the network network is population density seems to be a comparison factor totowhich is often compared [1,3,4]. For this reason, and because of the availability of data when analysing other often compared [1,3,4]. For this reason, and because of the availability of data when analysing other datasets we use datasets from from other other infrastructures, infrastructures, we use these these three three parameters parameters and and check check their their correlation correlation with with the chargers distribution. The availability of high-resolution aerial imagery has led to many points Energies 2018, 11, x 8 of 19 the chargers distribution. The availability of high-resolution aerial imagery has led to many points of of interest asas areas (building or site outlines), not points, in open streetstreet map. interest (POI) (POI)features featuresbeing beingrecorded recorded areas (building or site outlines), not points, in open minimize costs, social and resources, make use of existing data [8]. Fuel stationsuse and One for example, oftenimpacts, find a restaurant or and hotel an area. In area. this study, map.may, One may, for example, often find a restaurant ordrawn hotel as drawn as an In thiswe study,the wedata use parking lots are hence regarded as a proxy to estimate the agglomeration of demands. Further, the for fuel stations available by points and parking lots from [29] using the “traffic” point and the area the datapopulation for fuel stations available points and parking lotsthe from using of thethe “traffic” and density seems to be by a comparison factor to which size [29] or intensity networkpoint is subset. The choice of [1,3,4]. using the area filethe for parking lots isavailability to acknowledge thatanalysing these facilities have the areaoften subset. The choiceFor ofthis using area file of for lots is towhen acknowledge that these compared reason, and because theparking of data other different areas and number of parking places, which may have an impact on the outcome. However, other infrastructures, we use of these three parameters and check their correlation with facilitiesdatasets have from different areas and number parking places, which may have an impact on the the chargers distribution. The availability of high-resolution aerial imagery has led to many points of the correlation was also done for point layer. For the population density, the Eurostat database outcome. However, the correlation was also done for point layer. For the population density, [30] the interest (POI) features Data being and recorded as areas (building or(SEDAC) site outlines), points, in open or NASA’s Socioeconomic Applications Centre are not useful centres of street information. Eurostat database [30] or NASA’s Socioeconomic Data and Applications Centre (SEDAC) are useful map. One may, for example, often find a restaurant or hotel drawn as an area. In this study, we use Such references present data the greatpresent majority of the enabling its use when applied to other centres of information. Suchfor references forworld, the from great majority of“traffic” the world, its the data for fuel stations available by points anddata parking lots [29] using the pointenabling and infrastructures’ datasets. Figure 4 presents the way the correlation was obtained between fuel stations use when to other infrastructures’ Figure 4 presents way the that correlation was the applied area subset. The choice of using the datasets. area file for parking lots is to the acknowledge these data and the chargers data. obtained between stations data the chargers data. facilities havefuel different areas andand number of parking places, which may have an impact on the outcome. However, the correlation was also done for point layer. For the population density, the Eurostat database [30] or NASA’s Socioeconomic Data and Applications Centre (SEDAC) are useful centres of information. Such references present data for the great majority of the world, enabling its use when applied to other infrastructures’ datasets. Figure 4 presents the way the correlation was obtained between fuel stations data and the chargers data.

Figure 4. Fuels stations (blue) and chargers point (brown) distribution on grid. Figure 4. Fuels stations (blue) and chargers point (brown) distribution on grid.

The goal of this exercise is not to infer on the ratio of fuel stations and chargers, but instead on 4. Fuels stations (blue)on andthe chargers (brown) on grid. The goal distribution. of thisFigure exercise is not to infer ratiopoint ofmay fuel stations and chargers, but instead its its intensity The ratios between inputs be distribution completely different, becauseonfuel intensity distribution. The ratiosfuel between inputshave may be completely different, because fuelhigher stations stations normally kindsand of chargers, fuel, and are now in The goalhave of thisseveral exercise is notdispensers, to infer on the ratiodifferent of fuel stations but instead on normally have several fuel dispensers, have different kinds of fuel, and are now higher in number. number.itsInintensity order todistribution. perform aThe correlation, data normalization is required. For thisbecause reason,fuel the use of ratios between inputs may be completely different, In orderstations to a correlation, data isdifferent required. Forofbe this reason, the of ain is have fuel normalization dispensers, kinds fuel, andby aregenerating nowuse higher a matrix is perform an normally efficient wayseveral to correlate the two have datasets. This can done amatrix grid on an efficient way correlate the atwo datasets. This cannumber be done generating a grid Q-GIS number. Into order toaperform correlation, normalization is required. Forin this reason, theon use of Q-GIS and by running counting script to data return the ofbypoints each grid cell. Fromand the a matrix is an efficient way to correlate the two datasets. This in caneach be done by generating a grid on by running a counting script to return the number of points grid cell. From the density density data series generated, the correlation can then be calculated. The correlation betweendata the Q-GIS and by a counting scriptbe to calculated. return the number of points in each grid cell. From the point series generated, therunning correlation can then Thewas correlation charging charging point density data series and fuel stations density found tobetween be 75.7%the strong. Likewise, density data series generated, the correlation can then be calculated. The correlation between the density data series and fuel stations densityfor was found to be 75.7% strong. Likewise, Figure 5 shows Figure 5charging shows the analysis of information the parking lots in three steps (a), (b), and (c) point density data series and fuel stations density was found to be 75.7% strong. Likewise, the analysis of information for the parking lots in three steps (a), (b), and (c) Figure 5 shows the analysis of information for the parking lots in three steps (a), (b), and (c)

Figure 5. Parking Lots density raster (a) with chargers points overlapping (b) and cell assignment by

Figure Parking Lots raster Figure 5. 5. Parking Lots density density raster (a) (a) with with chargers chargers points points overlapping overlapping (b) (b) and and cell cell assignment assignment by by matrix generation (c). matrix generation (c). matrix generation (c). First a raster layer and corresponding area/point was loaded in Q-GIS. Because of the large quantity of information, good way to perceive the areas of intensity maps or First a raster layer and acorresponding area/point was loaded is inthrough Q-GIS.heat Because ofby the large

quantity of information, a good way to perceive the areas of intensity is through heat maps or by

Energies 2018, 11, 1869

9 of 18

Energies 2018, 11, x

9 of 19

First a raster layer and corresponding area/point was loaded in Q-GIS. Because of the large quantity information, a good way can to perceive theinareas intensity through heat yellow, maps orand by creating of raster layers. This intensity be visible threeoflevels; red is (most intense), creating raster layers.The Thissecond intensity can three levels; red (most intense), yellow, bluea blue (least intense). step is be to visible overlayinthe charging infrastructure point layer.and Third, (least intense). The second step is to overlay the charging infrastructure point layer. Third, a grid is grid is generated and the Q-GIS counting script outputs the number of points on each cell. The generated and the Q-GIS counting script outputs the number of points on each cell. The correlation correlation between charging points density and parking lots density, is moderate to strong being between charging points density andparking parkinglots. lots Using density, is point moderate to however, strong being using 63.00% using the area layer for the the layer, the 63.00% correlation is the area layer for the parking lots. Using the point layer, however, the correlation is 73.19%. 73.19%. For [31], two raster layers layers were were used. used. For covariance/correlation covariance/correlation estimation estimation of of the the population population data data [31], two raster The population density grid refers to a 5 km scale for the year 2015, the latest year to which the dataset The population density grid refers to a 5 km scale for the year 2015, the latest year to which the refers. population density raster in raster Figurein6aFigure consists estimates of humanofpopulation (number datasetThe refers. The population density 6aof consists of estimates human population of personsofper squareper kilometre), which is consistent with national and population registers. (number persons square kilometre), which is consistent withcensus national census and population The charger density presented in Figure 6b was created using the heat map tool of the raster setthe in registers. The charger density presented in Figure 6b was created using the heat map tool of Q-GIS, following a 10,000 radius of the layer’s units. As the raster layers act as a matrix of pixels, raster set in Q-GIS, following a 10,000 radius of the layer’s units. As the raster layers act as a matrix no normalization is needed, is hence no grid generation as in the later cases. Using the Grass of pixels, no normalization needed, hence no grid generation as in the later cases. Usingcommand the Grass tool, “r.covar”, and a generated sample of size n, Q-GIS provided the output of the covariance/correlation command tool, “r.covar”, and a generated sample of size n, Q-GIS provided the output of the matrix. With n = 2759, amatrix. correlation obtained.ofWe can visually observeWe from raster covariance/correlation Withofn 62.51% = 2759, was a correlation 62.51% was obtained. canthe visually layers that there is in fact a moderate positive relation between the concentration of people and charging observe from the raster layers that there is in fact a moderate positive relation between the columns density byand the charging dataset. columns density given by the dataset. concentration ofgiven people There be many factors influencing the decision for the charging deployment, Theremay maywell well be many factors influencing the decision for theinfrastructure charging infrastructure however from the three factors analysed, it can be verified they are positively correlated (moderate to deployment, however from the three factors analysed, it can be verified they are positively strong), and(moderate hence a good explanation the chargers density. In addition this, the density. parameters used are correlated to strong), andfor hence a good explanation for theto chargers In addition easily available and may be applied in a straightforward to be other datasets study as described to this, the parameters used are easily available andway may applied in under a straightforward way in to the present study. other datasets under study as described in the present study.

Figure 6. Raster layers of (a) population density in the Netherlands, and (b) public chargers density Figure 6. Raster layers of (a) population density in the Netherlands, and (b) public chargers density distribution in the Netherlands. distribution in the Netherlands.

3.4. Nearest Neighbour Distance and Availability 3.4. Nearest Neighbour Distance and Availability The distance at which charging points are located from each other or from a point on a map is The distance at which charging points are located from each other or from a point on a map is yet another variable of study often found in the literature [6,9,11]. The authors find different ways of yet another variable of study often found in the literature [6,9,11]. The authors find different ways of approaching distances, such as using polygon overlay methods, line layer distance, and Delaunay approaching distances, such as using polygon overlay methods, line layer distance, and Delaunay triangulations, among others. In this study, we use a nearest neighbour (NN) distance analysis, triangulations, among others. In this study, we use a nearest neighbour (NN) distance analysis, which, which, despite its shortcomings, compensates for its simplicity of use. Furthermore, working with despite its shortcomings, compensates for its simplicity of use. Furthermore, working with mostly mostly meshed networks, averages, and large scale data contributes to mitigating the plainness of meshed networks, averages, and large scale data contributes to mitigating the plainness of the method. the method. It is important to notice that the indicator only refers to the dataset under analysis and It is important to notice that the indicator only refers to the dataset under analysis and does not include does not include the whole country’s network. The minimum, maximum, and average distance the whole country’s network. The minimum, maximum, and average distance between chargers and between chargers and a given point in the network must be secured to allow the vehicle to reach a a given point in the network must be secured to allow the vehicle to reach a charging point within charging point within “reasonable” distances regarding its state of charge (SOC). This should be “reasonable” distances regarding its state of charge (SOC). This should be done by road segment in done by road segment in order to avoid situations such as a charger on a highway being close to order to avoid situations such as a charger on a highway being close to another outside the highway, another outside the highway, but impossible to reach. The distances are important to consider with but impossible to reach. The distances are important to consider with the availability to charge factor. the availability to charge factor. On one hand, if high availability to charge is observed, then longer distances between connectors may exist, on the other hand, if low availability to charge exists, then it should be compensated by shortening the distances between connectors. Table 1 presents the

Energies 2018, 11, 1869

10 of 18

On one hand, if high availability to charge is observed, then longer distances between connectors may exist, on the other hand, if low availability to charge exists, then it should be compensated by shortening the distances between connectors. Table 1 presents the distances between connectors and columns by road segment and total values. Also, the availability is shown by road segment, which tends to be higher in primary roads and lower in residential. Table 1. Average nearest neighbor (NN) distances to connector and to column. Road Mean NN 2018, 11, x (m) Segment Energies Charger Dist.

Mean NN Connector Dist. (m)

Min. (m) Distance

Max (m) Distance

N. of Availability (%) Charger—Connectors * 10 of 19

Primary 15,565 9462 3.1 52,900 40–67 94.27 between connectors 1101 and columns by road total values. Also, the availability is 87.86 Residentialdistances 1838 1.78segment and 21,910 1203–1974 road segment, which tends to be higher in primary30,593 roads and lower in residential. Secondary shown by6359 3985 1.73 175–279 91.28 Tertiary 4872 2850 6.99 27,382 258–451 90.49 Total 1603 968 neighbor (NN) 1.73 distances 12,087 88.78 Table 1. Average nearest to connector and1683–2785 to column.

* seven columns not assigned under any road segment.

Mean NN Mean NN Min. (m) Max (m) N. of Availability Charger Dist. Connector Dist. Distance Distance Charger—Connectors * (%) (m) (m) In Q-GIS, layers are often assigned a geographic coordinate reference system (WGS 84 or similar), Primary 15,565 9462 3.1 52,900 40–67 94.27 1838 21,910 1203–1974 reference 87.86system). It is which willResidential provide degrees instead1101of metric 1.78 system distances (projected Secondary 6359 3985 1.73 30,593 175–279 91.28 important toTertiary define the 4872 same reference to the positioning 2850 system for 6.99 the layers 27,382 according 258–451 90.49 of the area Total 1603 968 1.73 12,087 1683–2785 88.78 online [32]. under study. A straightforward tool to identify the grid zones of the world is available Road Segment

* seven columns not assigned under any road segment. The values presented in Table 1 refer to the mean, nearest neighbour distance, maximum and minimum between charging and theaaverage nearest neighbour distance between In Q-GIS, layers columns, are often assigned geographic coordinate reference system (WGS 84 orconnectors. similar),65% which(1087 will provide metric system distances have (projected reference Approximately out ofdegrees 1683)instead of theof charging columns more than system). two connectors. It is important define thethe same reference system layers according the positioning the For these columns, wetoconsider NN distance to for be the zero, which is thetonext availableofoption a user area under study. A straightforward tool to identify the grid zones of the world is available online has for charging his/her car if the first connector is occupied. The maximum distance is the same [32]. The values presented in Table 1 refer to the mean, nearest neighbour distance, maximum and for both situations if we consider the columns have nearest more than one connector. It is important minimum even between charging columns, and the to average neighbour distance between connectors. 65% (1087 out for of 1683) of the charging have one moreconnector. than two to highlight that theApproximately minimum value is zero the chargers withcolumns more than Table 1 connectors. For these columns, we consider the NN distance to be zero, which is the next available also shows the minimum distances of the nearest neighbour column. Not only is the distance of the option a user has for charging his/her car if the first connector is occupied. The maximum distance is next available charging option per segment important to assess, but so is the case when the complete the same for both situations even if we consider the columns to have more than one connector. It is network isimportant considered. In thisthat case, distance road is calculated and, hence, to highlight the the minimum valuebetween is zero for the segments chargers with more than one Table 1 also shows the be minimum distances of the nearest neighbour column. Not only is the closestconnector. neighbour will frequently a charger belonging to another road segment, which is why the distance the nextwhen available charging option important to assess, but the case there is the total has a lowerofvalue compared with per thesegment individual segments. In so theis dataset, when the complete network is considered. In this case, the distance between road segments is one charger in the secondary road segment that has 3 connectors and another charger with 14 and calculated and, hence, the closest neighbour will frequently be a charger belonging to another road 3 connectors in the residential Figure 7 provides a comprehensive view of the In availability segment, which is why thesegment. total has a lower value when compared with the individual segments. the dataset, one charger the secondary (a) and relation withthere idle istimes (b) byinroad segment.road segment that has 3 connectors and another charger with and 3 connectors residential segment. Figure 7 provides a comprehensive Considering the 14 complete dataset in forthethe two years under analysis, the highest maximum value view of the availability (a) and relation with idle times (b) by road segment. of idle time is Considering observed the in the residential and tertiary road segments, which are approximately the complete dataset for the two years under analysis, the highest maximum value same, just ofover min. Ainhigher idle time impacts thewhich availability, which is idle 15,000 time is observed the residential and directly tertiary road segments, are approximately thewhy there just over min. A higher idle time impacts the availability, which is why On therethe is contrary, is a lower same, minimum of15,000 availability to charge ondirectly the road segment’s chargers (37%). a lower minimum of availability to charge the road segment’s chargers (37%). Onof theidle contrary, if the primary road segment is observed, theonmedian and maximum values timeifare lower, the primary road segment is observed, the median and maximum values of idle time are lower, which correspond to a higher general availability to charge. More than one connector per column which correspond to a higher general availability to charge. More than one connector per column mitigates the lack of in some locations. mitigates the availability lack of availability in some locations.

Figure 7. Boxplot of availability (a) and idle time (b) per road segment.

Figure 7. Boxplot of availability (a) and idle time (b) per road segment.

Energies 2018, 11, 1869 Energies 2018, 11, x

11 of 18 11 of 19

According According to to Figure Figure 8, 8, the the percentage percentage of of residential residential chargers chargers with with two two connectors connectors (64%) (64%) is is less less than than that that in in the the tertiary tertiary (70%) (70%) or or primary primary road roadsegment segment (67.5%). (67.5%). The The lower lower percentage percentage of of chargers chargers Energies 2018, 11, x 11 of 19 with is observed observedin inthe thesecondary secondarysegment segment (59.4%). improve availability in with two two connectors is (59.4%). ToTo improve thethe availability in the the residential segment, which is the lowest, either the number of connectors per charger should be residential segment, which is the lowest, either the number of connectors per charger be According to Figure 8, the percentage of residential chargers with two connectors (64%) is less increased, or idle time be decreased. increased, or the the idle timeshould should beprimary decreased. than that in the tertiary (70%) or road segment (67.5%). The lower percentage of chargers with two connectors is observed in the secondary segment (59.4%). To improve the availability in the residential segment, which is the lowest, either the number of connectors per charger should be increased, or the idle time should be decreased.

Figure 8. 8. Percentage Percentage of of charging charging columns columns with with more more than than one one connector connector by by road roadsegment. segment. Figure columns with more one connector road segment. The factFigure that 8.aPercentage vehicle ofischarging parked in idle time than prevents otherbyvehicles from using the The fact that a vehicle is parked in idle time prevents other vehicles from using the infrastructure infrastructure for charging. If the charger is occupied, the vehicle has to approach the nearest The If fact a vehicle is parked idle time prevents other from using charging the for charging. thethat charger is occupied, the in vehicle has to approach the vehicles nearest neighbour neighbour charging column/connector to charge, which again may be occupied. In principle, this infrastructure for charging. If theagain charger theInvehicle has this to approach the nearest column/connector to charge, which mayisbeoccupied, occupied. principle, would motivate an extra would motivate an extracolumn/connector installation, increasing thewhich number of chargers per vehicle. Idle time should neighbour chargingthe to charge, again occupied. In be principle, this in installation, increasing number of chargers per vehicle. Idlemay timebeshould hence discouraged hence be discouraged in order to reduce the use service ratio, allowing maximum availability. would motivate an extra installation, increasing the number of chargers per vehicle. Idle time should One order to reduce the use service ratio, allowing maximum availability. One would expect to see less would expect to see less by aratio, charger whenever high idle time hence be discouraged in energy order to being reduce provided the use service allowing maximum availability. Oneexists. energy being provided by a charger whenever high idle time exists. However, this cannot be observed However, directly frombythe data, because is directly would this expectcannot to see be lessobserved energy being provided a charger whenever the highidle idle time time exists. directly from the data, because the idledirectly time is directly both thetime charging time and However, cannot be observed fromprovided. theproportional data, However, becausetothe idle is dataset directly proportional tothis both the charging time and energy looking at the from energy provided. However, looking at the dataset from the vehicles perspective (only vehicles that proportional to both the charging time and energy provided. However, looking at the dataset from the vehicles perspective (only vehicles that charge more than 24 kWh), it is visible in Figure 9a that charge than 24 kWh), (only it is visible inthat Figure 9a more that vehicles often ittend to be in parked in idle themore vehicles perspective charge 24 kWh), is visible Figure thatmore vehicles often tend to be parkedvehicles in idle more than thethan actual charging time. Figure 9b9ashows the than vehicles the actual charging Figure 9b shows the the positive energy charged often tend to time. be parked in idle more than actual relation charging between time. Figure 9b shows the and positive relation between energy charged and idle time. positive relation between energy charged and idle time. idle time.

Figure 9. Relation between the sum of EV idle time and charge time (a) and total energy drawn with Figure Figure 9. 9. Relation Relation between between the the sum sum of of EV EV idle idle time time and and charge charge time time (a) (a) and and total total energy energydrawn drawn with with corresponding idle time sum (b) both per vehicle. corresponding idle time sum (b) both per vehicle. corresponding idle time sum (b) both per vehicle.

3.5. Use Time Ratio

3.5. 3.5. Use Use Time Time Ratio Ratio

The use time ratio, given by Equation (2), provides insights on the recharging infrastructure

The given Equation (2), insights on recharging infrastructure utilization. The ratio, use time ratioby is the connected time, which represent the sum of idle The use use time time ratio, given bycalculated Equationusing (2), provides provides insights on the the recharging infrastructure time and charging time, divided by the total number of connectors multiplied by the corresponding utilization. The use ratio is using the time, which the utilization. The use time time ratio is calculated calculated using the connected connected time, which represent represent the sum sum of of idle idle hours. The ratios shown in Table 2 have been calculated by road segment. The results show that the time and charging time, divided by the total number of connectors multiplied by the corresponding time and charging time, divided by the total number of connectors multiplied by the corresponding installed infrastructures primary and been secondary roads segments aresegment. less used. On the other hand, hours. The ratios ratios showninininTable Table have beencalculated calculated road results show hours. The shown 22have byby road segment. TheThe results show thatthat the the residential use time ratio shows that 12.14% of the time, the connectors are plugged to EVs. Such the installed infrastructures in primary secondary segments areused. less used. the hand, other installed infrastructures in primary and and secondary roadsroads segments are less On theOn other a high value should not be completely attributed to a proportional higher energy take from the hand, the residential useratio time shows ratio shows that 12.14% the time, the connectors are plugged to Such EVs. the residential use time that 12.14% of theof time, the connectors are plugged to EVs. infrastructure, as can be seen in Table 3. The idle time in the residential road segment is higher Such a high value should not be completely attributed to a proportional higher energy take a high value should not be completely attributed to a proportional higher energy take from the because of the fact that cars stay connected overnight, making use of this time for both charging and the infrastructure, as can be seen in Table 3. The idle time in the residential road segment is higher because infrastructure, as can be seen in Table 3. The idle time in the residential road segment is higher parking. of the factofthat stay cars connected overnight, making use of this time for both and parking. because the cars fact that stay connected overnight, making use of this timecharging for both charging and parking.

Energies 2018, 11, x Energies 2018, 11, 1869

12 of 19 12 of 18

Table 2. Use time ratio by road segment. Road Segment Time (%)segment. Connectors Table 2. UseUse time ratioRatio by road Residential 12.14 1974 Road Segment Use Time Ratio (%) Connectors Primary 5.72 67 Secondary 8.7 279 Residential 12.14 1974 Primary 5.72 67 Tertiary 9.5 451 Secondary 8.7 279 Total 11.21 2771 Tertiary 9.5 451 Total

11.21

2771

As previously mentioned, each StartCard is assumed to be assigned to a specific EV. The total number of StartCards for the year 2014 and 2015 42,400. Among these users, the average number As previously mentioned, each StartCard is is assumed to be assigned to a specific EV. The total of recharges in two years is 17.31, with a lower median value of four charging events (max of 1105). number of StartCards for the year 2014 and 2015 is 42,400. Among these users, the average number The low mean by the fact that only 47.4% ofevents EVs (20,069) of recharges inand twomedian years isvalues 17.31, can withbeaexplained lower median value of four charging (max ofcharge 1105). more than 24 kWh in two years. If only these EVs are considered, the average number of charges per The low mean and median values can be explained by the fact that only 47.4% of EVs (20,069) charge each StartCard is 34.09 instead of 17.31. Figure 10 only considers such EVs that have been identified more than 24 kWh in two years. If only these EVs are considered, the average number of charges per as frequent users (>24 kWh) of of the17.31. ElaadNL infrastructure. It shows average number recharges each StartCard is 34.09 instead Figure 10 only considers suchthe EVs that have been of identified as in the whole infrastructure per week day. There is a marked difference between the number frequent users (>24 kWh) of the ElaadNL infrastructure. It shows the average number of recharges of in recharges during week days (median of 45), and the weekend days where it drops to 36 on Saturday the whole infrastructure per week day. There is a marked difference between the number of recharges and 29 on Sunday. The use time ratio hence well distributed theto all36weekdays, with a lower during week days (median of 45), andisthe weekend days whereacross it drops on Saturday and 29 on expected use on weekends. However, the energy use ratio can further help verifying this Sunday. The use time ratio is hence well distributed across the all weekdays, with a lower expected observation. use on weekends. However, the energy use ratio can further help verifying this observation.

Figure Figure 10. 10. Boxplot Boxplotof ofthe theaverage average recharging recharging events events on on the the network network per per week week day. day.

3.6. 3.6. Energy Energy Use Use Ratio Ratio The The energy energy use use ratio ratio is is given given by by Equation Equation (1). (1). Observing Observing Table Table 3, 3, even even if if the the average average maximum maximum power is considered to estimate the total infrastructure capacity, only 3.56% of the power is considered to estimate the total infrastructure capacity, only 3.56% of the energy energy is is actually actually being drawn. The residential segment appears again with a higher share, as it was expected from being drawn. The residential segment appears again with a higher share, as it was expected from the the use use time time ratio ratio observed observed in in Table Table 2. 2. Even Even though though the the residential residential segment segment chargers chargers are are those those with with the the higher for the the higher higher amount amount of of energy energy provided. provided. higher idle idle time, time, they they are are also also responsible responsible for Table Table 3. 3. Energy Energy use use ratio ratio by by road road segment. segment. Road Segment Road Segment Residential connectors: Residential connectors: Primary connectors: Primary connectors: Secondary connectors: Secondary connectors: Tertiary connectors: Tertiary connectors: Total Total

Energy Use (%) Connectors Energy Use (%) Connectors 3.75 1974 3.75 1974 2.36 67 2.36 67 3.24 279 3.24 279 3.11 451 3.11 451 3.56 2771 3.56 2771

Figure provided byby thethe network for for eacheach day day of the The Figure 11 11illustrates illustratesthe theaverage averageenergy energy provided network ofweek. the week. mean dailydaily energy demand is 400iskWh working week days. the amount of energy The mean energy demand 400 during kWh during working weekLikewise, days. Likewise, the amount of required during the weekends is lower when compared with working days, by 22% (31.2 kWh) on energy required during the weekends is lower when compared with working days, by 22% (31.2 kWh) Saturday andand 35%35% (260(260 kWh) on Sunday, on average. The fact is more energy drawndrawn from on Saturday kWh) on Sunday, on average. Thethat fact there that there is more energy

Energies 2018, 11, 1869 Energies 2018, 11, x

13 of 18 13 of 19

fromnetwork the network during weekdays is substantiated higher numberofofcharging chargingevents events when the during weekdays is substantiated by by thethe higher number compared with weekend days.

Figure 11. Average energy demand to the network per week day. Figure 11. Average energy demand to the network per week day.

3.7. Total Service Ratio 3.7. Total Service Ratio The classic service ratio refers to the relation between the amount of cars per chargers or The classic service ratio refers to the relation between the amount of cars per chargers or chargers chargers per car. However, the functional unit of analysis should be energy related and if a direct per car. However, the functional unit of analysis should be energy related and if a direct comparison comparison of vehicles to the number of points is to be made, it should contemplate the number of of vehicles to the number of points is to be made, it should contemplate the number of connectors, connectors, not columns. The number of vehicles per number of positions (34,380 positions of not columns. The number of vehicles per number of positions (34,380 positions of normal power in normal power in the Netherlands) in European Alternative Fuel Observatory (EAFO) [3] is 3.59 the Netherlands) in European Alternative Fuel Observatory (EAFO) [3] is 3.59 vehicles per charger. vehicles per charger. From the dataset under analysis, there are 42,400 vehicles using the network, From the dataset under analysis, there are 42,400 vehicles using the network, which correspond to which correspond to 25.19 vehicles per column. However, if the connectors are considered (2785), 25.19 vehicles per column. However, if the connectors are considered (2785), this implies 15.22 vehicles this implies 15.22 vehicles per connector. It should be highlighted that the vehicles do not use the per connector. It should be highlighted that the vehicles do not use the charging network exclusively, charging network exclusively, meaning that they can be assigned to other providers’ connectors as meaning that they can be assigned to other providers’ connectors as well. As stated before, more than well. As stated before, more than half of the EVs (22,331) in the dataset used less than an average full half of the EVs (22,331) in the dataset used less than an average full charge (24 kWh) in two years. charge (24 kWh) in two years. This shows one more reason why the vehicles alone should not be This shows one more reason why the vehicles alone should not be used as an indicator. Considering used as an indicator. Considering only the EVs charging more than 24 kWh will yield a ratio of 7.2 only the EVs charging more than 24 kWh will yield a ratio of 7.2 EV/connector (20,069/2785). There are EV/connector (20,069/2785). There are just over 119,000 [1] EVs in the Netherlands, and as we know just over 119,000 [1] EVs in the Netherlands, and as we know that the dataset under analysis (2014 that the dataset under analysis (2014 and 2015) corresponds to 8.1% (2785 connectors/34,380) of the and 2015) corresponds to 8.1% (2785 connectors/34,380) of the total network in 2018, if, in exclusivity, total network in 2018, if, in exclusivity, 8.1% of the vehicles would be allocated to the network, then 8.1% of the vehicles would be allocated to the network, then it would correspond to 3.46 (119,000 × it would correspond to 3.46 (119,000 × 0.081/2785) EVs per connector. This value, given the 0.081/2785) EVs per connector. This value, given the proportion, is in line with the EAFO data (3.59). proportion, is in line with the EAFO data (3.59).

3.8. Carbon Intensity of the Infrastructure 3.8. Carbon Intensity of the Infrastructure Equation (4) calculates the global warming potential (GWP) from the life cycle analysis (LCA) of Equation calculates warming potential (GWP) from the life cycle analysis (LCA) materials used (4) in the chargersthe perglobal total energy flowing in CO 2eq /MJ (MJ in final energy). This is given of used in theofchargers per charger total energy flowing COinventory, 2eq/MJ (MJ in final energy). This is by materials the carbon intensity the normal (187.04 gCO2 in /MJ) based on Lucas et al. [7]. given by the carbon intensity of the normal charger (187.04 gCO 2/MJ) inventory, based on Lucas et The inventory value is based on a column with two connectors. Evidently, the energy flowing through al. The inventory value based on a will column with two connectors. Evidently, energy flowing the[7]. chargers that have twoisconnectors be higher, hence contributing to itsthe overall reduction. through the is chargers that have connectors ofwill be higher, hence contributing to its overall Such a value then multiplied bytwo the percentage normal chargers (%Nch SR ) of the total number of reduction. Such a value is then multiplied by the percentage of normal chargers (%Nch ) of the total chargers-per-vehicle and by the chargers-per-vehicle ratio (SRTotal ). The numerator isSRthen divided number of chargers-per-vehicle and by the the chargers, chargers-per-vehicle ratio (SRby Total). The numerator is then by the total energy that flows through which is calculated multiplying the average divided by the total energy that flows through the chargers, which is calculated multiplying energy consumed by a car per year (AvrcarMJ /year = TTWEnergy X km/year) bybythe percentagethe of average energy consumed by a car per year (Avrcar MJ/year = TTWEnergy X km/year) by the percentage charge actually being performed in this type of charger (%Nch ) during its lifetime period (LTimeNch ). of charge actually performed in this istype of charger (%Nch) during period In the present study, being in which an infrastructure assessed, no assumptions have toits be lifetime made regarding (LTime Nch). In the present study, in which an infrastructure is assessed, no assumptions have to be the proportion of chargers, because it is given by the dataset. The number of chargers corresponding made regarding the proportion of chargers, because it is given by the dataset. The number of to the %Nch SR × SRTotal part in the numerator is 1683. The energy in the denominator is the sum of chargers corresponding %NchSR ×inSR partbasis. in the numerator is 1683. energy in the the two years, which hasto to the be estimated anTotal annual The lifetime used wasThe 10 years. denominator is the sum of the two years, which has to be estimated in an annual basis. The lifetime used was 10 years. 2 × 187.04 × 103 × %NchSR × SRTotal Nc = (4) 3 ×%N Avrcar 2×187.04 × 10× %Nch SRTotal Nch chSR×× LTime MJ/year NC = (4) AvrcarMJ/year × %Nch × LTimeNch

Energies 2018, 11, 1869

14 of 18

Energies 2018, 11, x

14 of 19

The uncertainty associated with the parameters used is based on the literature [7]. There is no data on the expected lifetime of the the parameters product, however, we on have it and considered Theexact uncertainty associated with used is based theincreased literature [7]. There is no datato onaccount the exactfor expected lifetime of the product, however, we have increased it and 10 to 10 years technology evolution and material degradation. Crystal Ballconsidered [27] was used years the to account for technology evolution and material Crystal Ball the [27] forecast was usedistogiven introduce uncertainty statistical distributions. After degradation. running 100,000 trials, introduce the mean uncertainty distributions. After running is 100,000 trials, the forecast is given in Figure 12. The valuestatistical for the carbon intensity indicator 4.44 gCO 2eq /MJ and the standard in Figure 12. The mean value for the carbon intensity indicator is 4.44 gCO2eq/MJ and the standard deviation is 0.87, with minimum and maximum being 2.09 and 10.65, respectively. The value found deviation is 0.87, with minimum and maximum being 2.09 and 10.65, respectively. The value found is 30% higher, but in line to the ones reported in the literature [26]. For the Portuguese case study, is 30% higher, but in line to the ones reported in the literature [26]. For the Portuguese case study, the the authors found a mean of 3.4 gCO2eq/MJ. /MJ. Evidently, this occurs because there is on average less authors found a mean of 3.4 gCO2eq Evidently, this occurs because there is on average less energy flowing through the charging network the Portuguese Portuguesestudy. study. energy flowing through the charging networkthan than in in the

Figure 12. Carbon intensity forecast and sensitivity analysis of the infrastructure. GWP—global

Figure 12. Carbon intensity forecast and sensitivity analysis of the infrastructure. GWP—global warming potential. warming potential. Table 4 shows a summary of the indicators discussed and analysed in this section. In total, eight indicators were developed. A fair comparison can discussed only be done in relative terms to other datasets. Table 4 shows a summary of the indicators and analysed in this section. In total, However, in absolute terms, one can observe that there is a discrepancy between the use time eight indicators were developed. A fair comparison can only be done in relative terms to otherratio datasets. and the useterms, ratio. This dueobserve to low charging time comparedbetween with idle the time. The fact ratio However, inenergy absolute oneiscan that there is when a discrepancy use time that there is high availability means that overall, there is low energy demand when compared with and the energy use ratio. This is due to low charging time when compared with idle time. The fact what the network could provide. This causes a low allocation of the carbon intensity, which seems to that there is high availability means that overall, there is low energy demand when compared with be higher regarding values in the literature. The values reveal that the network could be over whatdimensioned, the networkwhich couldisprovide. This a low allocation of the carbon intensity, which seems normal for the causes early adoption stage. Nevertheless, because of the energy use to beand higher regarding values in the literature. The values reveal that the network could be service ratio assessed, one may observe that 52.6% of the vehicles charge less than 24 kWh in over dimensioned, forofthethe early adoption stage. Nevertheless, of thethat energy two years,which whichisisnormal a misuse network. In addition to this, it maybecause be observed the use correlations to strong, which of seems to indicate that less the network is well and service ratio assessed assessed,are onemoderate may observe that 52.6% the vehicles charge than 24 kWh in two is of also moderate In charger and to geographic concentration in that termsthe of correlations energy years,distributed. which is aThere misuse thea network. addition this, it may be observed demand concentration. Still, there is high enough charging andisdispersion to promote assessed are moderate to strong, which seems to indicate thatavailability the network well distributed. There is EV adoption. also a moderate charger and geographic concentration in terms of energy demand concentration. Still,

there is high enough charging availability and dispersion to promote EV adoption. Table 4. Summary of all charging infrastructure indicators analysis. EVs—electric vehicles Table 4.Indicator Summary 1—Energy Demand

Performance of all charging infrastructure indicators analysis. EVs—electric vehicles

6685.72 MJyear/charger and 4040.23 MJyear/connector Higher intensity in six urban areas. 50% of energy supplied observed in 19.6% of 2—Energy Use Intensity Indicator Performance the chargers. 6685.72 MJyear /charger and lots 4040.23 MJyear /connector 1—Energy Demandintensity distribution 3—Chargers Fuel Stations Corr—75.7%; Parking Corr—73.19%; Population density Higher intensity in six urban areas. 50% of energy supplied observed in 19.6% of (correlations) Corr—62.51% 2—Energy Use Intensity the chargers. 4—Nearest Neighbour distance and Primary—9462 (94.27%); Residential—1101 (87.86%); Secondary—3985 (91.28%); Fuel Stations Corr—75.7%; Parking lots Corr—73.19%; Population density 3—Chargers intensity distribution (correlations) Availability Tertiary–2850 (90.49%); Total—968 (88.78%) Corr—62.51% 5—Use time ratio (Time related) 11.21% Primary—9462 (94.27%); Residential—1101 (87.86%); Secondary—3985 (91.28%); 4—Nearest Neighbour distance and Availability 6—Energy use ratio (Total) 3.56% Tertiary–2850 (90.49%); Total—968 (88.78%) 5—Use time ratio (Time related) 11.21% consuming more than 24 kWh in two years (15.22 if 7—Total Service ratio 7.2 EVs/connector 6—Energy use ratio (Total) 3.56% vehicles/charging positions considering all EVs ever connected) EVs/connector consuming more than 24 kWh in two years (15.22 if 8—Carbon intensity 4.44 gCO7.2 2eq/MJ 7—Total Service ratio vehicles/charging positions considering all EVs ever connected) 4.44 gCO2eq /MJ 8—Carbon intensity

4. Impact of Public Policies

Energies 2018, 11, 1869

15 of 18

4. Impact of Public Policies At this stage of EV market penetration, policy support is still indispensable for lowering the barriers for adoption. A supportive policy environment enables market growth by making EVs appealing to consumers, reducing risks for investors, and encouraging manufacturers willing to develop EV business to start implementing them. Policy support mechanisms can be grouped into four major categories: supporting research and development of innovative technologies; targets, mandates, and regulations; financial incentives; and other instruments (primarily enforced by the cities). Naturally, cities and municipalities benefit from, and are constrained by, policy at the regional and national level. Such targeted policies are best developed and adapted to the unique, local mobility conditions of each urban area. Popular examples are the following: exemptions from access restrictions to urban areas, free public charging, EV-friendly building and parking codes, special road or lane access, car registration benefits, parking benefits, and local purchase incentives [2,33,34]. Such policies have a direct effect on the infrastructure use, and hence on the assessed indicators discussed here. As an example, EV support actions promoting parking benefits are commonly found in major cities such as Amsterdam, Shanghai, Utrecht, Oslo, San Jose, Shenzhen, or Taiyuan [2]. They are, however, only effective in a promotion stage, because they could allow a potential misuse of the network by increasing the idle time, thus decreasing the availability to other users. By assessing the indicators presented in this article, infrastructures both already in place and in the design stage may benefit from information towards optimization. Variables can be changed and the effect they may have on the indicators can be studied. Effect on Indicators by Changing Parameters We provide two examples to show how some indicators may be influenced by changing target variables of public policies. The first case considers a public policy aiming for a 50% reduction of idle time. One may approach this exercise by fixing different parameters. In this case, we assume that the use time ratio, and hence availability of the system, is maintained. We will recalculate indicators referred as 1, 6, and 8 in Table 4, which are the ones directly impacted. Regarding the first, the energy demand indicator, if actions are taken to reduce the idle time by 50%, maintaining the use time ratio, this means that the connected time is also maintained. Because the connected time is the sum of the idle time plus the charging time, the corresponding decrease in the first will lead to the same increase in the second. The connected time (in minutes) is 5.27 × 106 = 3.38 × 106 (total sum of idle) + 1.89 × 106 (total sum of charging times) being 50% of 3.38 × 106 = 1.62 × 106 min or 27.0 × 103 h. This extra time, if used as charging time, at an average power of 3.87 kWh (from the dataset), means an extra flow of energy of 103.8 × 103 kWh, which means 11.45 × 106 MJ/year. Dividing this by the number of charging columns or connectors provides the first indicator output of 6802.08 MJyear /column and 4110.56 MJyear /connector. Following the same principle for the other indicators, considering the new amount of energy demand, indicator 6 becomes 4.68% and indicator 8 becomes 4.36 gCO2eq /MJ. Figure 13a shows the variation on the assessed indicators with policy aiming for a reduction of 50% of the idle time in place. For the second example, increasing the NN distance would correspond to eliminating certain columns. Decreasing the number of chargers means that the ratio of cars/charger will be higher. This also means that the energy use ratio will increase and the availability will decrease, therefore, an acceptable compromise must be found. We consider the residential road segment, which has lower average distance between connection points (1101 m) and columns (1838 m). Considering, for instance, that an increase of 20% of the distance between columns would increase it to 2205 m. Firstly the distances of the existing residential chargers dataset are organized in ascending order. Secondly, the ones that are lower are removed and the new average computed in an iterative process until the 2205 m distance is reached. Once this is done, the number of chargers removed is obtained, which is 103, corresponding to 6.12% of all the chargers. The total energy demand is assumed to remain constant. The energy demand indicator becomes 7121.56 MJyear /chargers in this case, the use time

Energies 2018, 11, 1869

16 of 18

Energies 2018, 11, xis now 11.45% of total and 12.57% for residential. In terms of the energy use 16 of 19 ratio indicator ratio, increasing the NN distance results in ratio of 3.85% and the infrastructure carbon intensity indicator reducestoto 4.10 2eq/MJ. Figure 13b presents the corresponding variation of the assessed reduces 4.10 gCOgCO 2eq /MJ. Figure 13b presents the corresponding variation of the assessed indicators, indicators, when the residential distance is by 20%with compared when the residential NN distanceNN is increased byincreased 20% compared the basewith case.the base case.

Figure13. 13.Indicator Indicator variation variation with with 50% 50% reduction reduction of of idle idle time time (a) (a)and andindicator indicatorvariation variation with with 20% 20% Figure increase of residential NN distance compared with the base case (b). increase of residential NN distance compared with the base case (b).

These examples show how policy implementation may affect the indicators, and hence be used These examples show how policy implementation may affect the indicators, and hence be used as as a straightforward tool to manage the operation of charging networks. a straightforward tool to manage the operation of charging networks. Conclusions 5.5.Conclusions The article article presents presents aa methodology methodology based based on on eight eight indicators indicators to to analyse analyse the the EV EV charging charging The infrastructure features and performances. The methodology is developed around the principle that infrastructure features and performances. The methodology is developed around the principle that energydemand demandfrom fromEVs, EVs,rather ratherthan thanthe themere merenumber numberof ofchargers, chargers,should shouldbe beused usedas asthe thefunctional functional energy unit to build the indicators characterising the charging infrastructure. The analysed dataset revealsaa unit to build the indicators characterising the charging infrastructure. The analysed dataset reveals high availability and a low energy use ratio in the infrastructure under study. This means that low low high availability and a low energy use ratio in the infrastructure under study. This means that energyflows flowsthrough throughthe thenetwork. network.This Thiscauses causesaalow lowallocation allocationof of the the carbon carbon intensity, intensity,which whichseems seems energy high regarding values in the literature. Moreover, the fuel station, parking lots, and population high regarding values in the literature. Moreover, the fuel station, parking lots, and population density density correlations assessed are moderate to strong, which seems that to indicate that distribution the network correlations assessed are moderate to strong, which seems to indicate the network distribution is appropriate. The values reveal that the network could be over dimensioned, which is appropriate. The values reveal that the network could be over dimensioned, which may sound may sound reasonable at the early stage of EV adoption. There is nevertheless moderate charger and reasonable at the early stage of EV adoption. There is nevertheless moderate charger and geographic geographic concentration in terms of energy Still,enough there is high enough charging concentration in terms of energy demand. Still, demand. there is high charging availability and availability and dispersion to promote EV deployment. In terms of public policies, idle time was dispersion to promote EV deployment. In terms of public policies, idle time was observed to be one observed to be one parameter to negatively impact the infrastructure size and use. As EV parameter to negatively impact the infrastructure size and use. As EV penetration starts to increase, penetration increase,should policies as free should besized phased An inadequately policies suchstarts as freetoparking be such phased out. parking An inadequately gridout. negatively impacts sized grid negatively impacts all indicators. If undersized, the EV adoption may be inferior, as the all indicators. If undersized, the EV adoption may be inferior, as the user perception of an adequate user perception adequate infrastructure may an be over undermined. On network the otherwill hand, an over infrastructure mayofbean undermined. On the other hand, dimensioned increase the dimensioned network will increase the cost, social impact, and environmental burden. cost, social impact, and environmental burden. AuthorContributions: Contributions:Resources Resourcesand andData DataCuration, Curation,Writing—Original Writing—OriginalDraft DraftPreparation, Preparation,Conceptualization, Conceptualization, Author Investigation,and andMethodology, Methodology, A.L.; Software, G.P.,M.G.F.; and M.G.F.; Validation, E.K.; Formal Investigation, A.L.; Software, A.L.,A.L., G.P., and Validation, A.L. andA.L. E.K.;and Formal Analysis, A.L., G.P., and M.G.F.; & Editing, A.L.,&G.P., M.G.F., E.K.,G.P., G.F.,M.G.F., and M.M.; Visualization, A.L., Analysis, A.L., G.P., Writing—Review and M.G.F.; Writing—Review Editing, A.L., E.K., G.F., and M.M.; G.P., and M.G.F.; Supervision, G.F., and M.M.; Project Administration, M.M.Administration, M.M. Visualization, A.L., G.P., and E.K., M.G.F.; Supervision, E.K., G.F., and M.M.; Project Acknowledgments: We would like to thank ElaadNL for the data availability and cooperation. Acknowledgments: We would like to thank ElaadNL for the data availability and cooperation. Conflicts of Interest: The authors declare no conflict of interest Conflicts of Interest: The authors declare no conflict of interest

Abbreviations Abbreviations AFI AFI Avrcar MJ/year AvrcarMJ/year DC DC DOE DOE EAFO EC EU

Alternative Fuel Infrastructure Alternative Fuel Infrastructure Annual Average of energy use per car Annual Average Direct Current of energy use per car Direct Currentof Energy Department Department of Energy European Alternative Fuel Observatory European Commission European Union

Energies 2018, 11, 1869

EAFO EC EU EV GWP LCA Ltime MS NCS NN POI SD SEDAC SOC SRTotal TTW T0 WGS %NchSR %Nch

17 of 18

European Alternative Fuel Observatory European Commission European Union Electric Vehicle Global Warming Potential Life Cycle Analysis Life Time Member State Nippon Charge Service Nearest Neighbour Point of interest Standards Deviation Socioeconomic Data and Applications Centre State of Charge Chargers-per-vehicle ratio Tank to Wheel Time period of observations in hours World Geodetic System Percentage of normal chargers Percentage of charge in normal charger

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15.

16.

International Energy Agengy. Global EV Outlook 2018; IEA Publications: Paris, France, 2018; p. 144. Hall, D.; Cui, H.; Lutsey, N. Electric Vehicle Capitals of the World: What Markets are Leading the Transition to Electric? ICCT Briefing: Washington, DC, USA, 2017; p. 9. EAFO. Available online: http://www.eafo.eu/electric-vehicle-charging-infrastructure (accessed on 15 June 2018). Hall, D.; Lutsey, N. Emerging Best Practices for Electric Vehicle Charging Infrastructure; ICCT WP: Washington, DC, USA, 2017; p. 54. Sierzchula, W.; Bakker, S.; Maat, K.; van Wee, B. The influence of financial incentives and other socio-economic factors on electric vehicle adoption. Energy Policy 2014, 68, 183–194. [CrossRef] Harrison, G.; Thiel, C. An exploratory policy analysis of electric vehicle sales competition and sensitivity to infrastructure in Europe. Technol. Forecast. Soc. Chang. 2017, 114, 165–178. [CrossRef] Lucas, A.; Neto, R.C.; Silva, C.A. Energy supply infrastructure LCA model for electric and hydrogen transportation systems. Energy 2013, 56, 70–80. [CrossRef] Gkatzoflias, D.; Drossinos, Y.; Zubaryeva, A.; Zambelli, P.; Dilara, P.; Thiel, C. Optimal Allocation of Electric Vehicle Charging Infrastructure in Cities and Regions; JRC Report; EU Joint Research Centre: Luxembourg, 2016. Gagarin, A.; Corcoran, P. Multiple domination models for placement of electric vehicle charging stations in road networks. Comput. Oper. Res. 2017, 96, 69–79. [CrossRef] Liu, J. Electric vehicle charging infrastructure assignment and power grid impacts assessment in Beijing. Energy Policy 2012, 51, 544–557. [CrossRef] Huang, K.; Kanaroglou, P.; Zhang, X. The design of electric vehicle charging network. Transp. Res. Part D Transp. Environ. 2016, 49, 1–17. [CrossRef] Zhu, Z.H.; Gao, Z.Y.; Zheng, J.F.; Du, H.M. Charging station location problem of plug-in electric vehicles. J. Transp. Geogr. 2016, 52, 11–22. [CrossRef] Hafez, O.; Bhattacharya, K. Optimal design of electric vehicle charging stations considering various energy resources. Renew. Energy 2017, 107, 576–589. [CrossRef] Guo, S.; Zhao, H. Optimal site selection of electric vehicle charging station by using fuzzy TOPSIS based on sustainability perspective. Appl. Energy 2015, 158, 390–402. [CrossRef] Cai, H.; Jia, X.; Chiu, A.S.F.; Hu, X.; Xu, M. Siting public electric vehicle charging stations in Beijing using big-data informed travel patterns of the taxi fleet. Transp. Res. Part D Transp. Environ. 2014, 33, 39–46. [CrossRef] Wirges, J. Planning the Charging Infrastructure for Electric Vehicles in Cities and Regions; KIT Scientific Publishing: Karlsruhe, Germany, 2016.

Energies 2018, 11, 1869

17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28.

29. 30. 31. 32. 33. 34.

18 of 18

Vazifeh, M.M.; Zhang, H.; Santi, P.; Ratti, C. Optimizing the Deployment of Electric Vehicle Charging Stations Using Pervasive Mobility Data; MIT: Boston, MA, USA, 2015. Yi, Z.; Bauer, P.H. Optimization models for placement of an energy-aware electric vehicle charging infrastructure. Transp. Res. Part E Logist. Transp. Rev. 2016, 91, 227–244. [CrossRef] Xiong, Y.; Gan, J.; An, B.; Miao, C.; Bazzan, A.L.C. Optimal Electric Vehicle Charging Station Placement. In Proceedings of the 24th IJCAI, Buenos Aires, Argentina, 25–31 July 2015. Helmus, J.R.; Spoelstra, J.C.; Refa, N.; Lees, M. Assessment of public charging infrastructure push and pull rollout strategies: The case of the Netherlands. Energy Policy 2018, 121, 35–47. [CrossRef] Elaad. Available online: https://www.elaad.nl/ (accessed on 15 June 2018). European Commission. Directive of the european parliament and of the council on the deployment of alternative fuels infrastructure. Off. J. Eur. Union 2014, 12, 1–38. Chademo. Available online: https://www.chademo.com (accessed on 15 June 2018). TEN-T Projects. Available online: https://ec.europa.eu/inea/en/ten-t/ten-t-projects (accessed on 15 June 2018). Green Emotion. Available online: http://www.greenemotion-project.eu/ (accessed on 15 June 2018). ENOA. Available online: https://www.enova.no/ (accessed on 15 June 2018). Crystal Ball. Available online: https://www.oracle.com/applications/crystalball/index.html (accessed on 15 June 2018). European Commission. Towards the Broadest Use of Alternative Fuels-an Action Plan on Alternative Fuels Infrastructure under Article 10(6) of Directive 2014/94/EU. Available online: http://ec.europa.eu/ transparency/regdoc/?fuseaction=list&coteId=1&year=2017&number=652&version=F&language=en (accessed on 15 June 2018). GeoFabrik. Available online: http://download.geofabrik.de/europe.html (accessed on 15 June 2018). Eurostat. Available online: http://ec.europa.eu/eurostat/statistics-explained/index.php/Population_grids (accessed on 15 June 2018). SEDAC. Available online: http://sedac.ciesin.columbia.edu/data/set/gpw-v4-population-density-rev10/ data-download (accessed on 15 June 2018). Grid Zones of the World. Available online: http://www.dmap.co.uk/utmworld.htm (accessed on 15 June 2018). International Energy Agengy. Global EV Outlook 2017: Two Million and Counting; IEA Publications: Paris, France, 2017. Heidrich, O.; Hill, G.; Neaimeh, M.; Huebner, Y.; Blythe, P.; Dawson, R. How do cities support electric vehicles and what difference does it make? Technol. Forecast. Soc. Chang. 2017, 123, 17–23. [CrossRef] © 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).