Evaluation of Historical Electric Vehicle (EV) Driving Data to Suggest Improvements in Driving Efficiency Benjamin Pichler Institute for Pervasive Computing Johannes Kepler University Linz 4040 Linz, Austria
[email protected] Andreas Riener Institute for Pervasive Computing Johannes Kepler University Linz 4040 Linz, Austria
[email protected]
Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the Owner/Author. Copyright is held by the owner/author(s). AutomotiveUI’15 Adjunct, September 01–03, 2015, Nottingham, United Kingdom. ACM 978-1-4503-3858-5/15/09
http://dx.doi.org/10.1145/2809730.2809756
Abstract Increasing the operating distance (range) and, in particular, the accuracy of charge state/remaining range displays is very important for battery electrical vehicles (BEVs) to gain market penetration and satisfy customers. Battery drain of BEVs is dictated to a great extent by external conditions (elevation profile, outdoor temperature, precipitation), internal factors (consumption of appliances like A/C, heating) and the individual style of driving. The high variation caused by these factors is, however, not considered in most range displays of current BEVs. This was the main motivation for us to examine data from a large database of real-world BEV trips and to take into account a lot more influencing parameters than any other similar project before. Our analysis revealed, for instance, that the average battery drain nearly doubles between trips driven at moderately warm versus cold outdoor temperatures. Preliminary results allow 1) to optimize accuracy/precision of remaining range displays and 2) to use the information with a gamification approach to motivate drivers to drive more economically.
Author Keywords Battery electric vehicles (BEVs); Behavior change; Charge state; Gamification; Optimizing operating distance (range).
ACM Classification Keywords Human-centered computing: [Human computer interaction (HCI)]: Interactive systems and tools, User interface management systems.
Introduction
1) External/Environmental Outdoor temperature Insulation Wind Atmospheric conditions (e.g., precipitation) Road type and condition Topology (elevation profile) Traffic condition (rush hour) 2) Vehicle dependent Inside temperature EV body (air resistance) Empty weight Tire tread patterns (rolling friction) Total battery capacity, age and actual charge state 3) Personal Occupants (payload) Individual driving style Distance to front vehicles (influence of slipstream) Plugged electric devices State of electric consumers such as heating, A/C, lights, entertainment systems, etc.
Table 1: Factors limiting battery electric vehicles (BEV) range.
Although battery electric vehicles (BEVs) become more and more popular, there are still a number of drawbacks and problems compared to traditional combustion engine vehicles, such as long charging periods, limited range or “range anxiety” [4]. According to previous work and our own preliminary analysis [9], battery capacity is exposed to tremendous variation caused by changing environmental conditions [10], individual driving style [1], and other factors as outlined in Table 1. In range displays of actual BEVs, most of these factors are not taken into account. The remaining range is calculated based on current battery charge state and the average energy consumption over the last few trips, which is known to be not very accurate and precise. Research approach The aim of this research project is twofold. First, to provide – based on extensive data from real BEVs – detailed insights on the impact of the most influential factors on battery drain or operating distance (range). Second, to use this information to implement (and evaluate) a more precise range display based, e. g., on historic trips with similar characteristics. In addition, gamification concepts are integrated to further enhance range based on recommendations for more economic driving. Regarding 1), to allow for a first, easy-to-use preselection and analysis of BEV driving data according to different characteristics (outdoor temperature, precipitation, spatial/temporal route length, etc.), we developed an interactive exploration tool. Thanks to our project partner [3], the tools’ power results from thousands of electric vehicle trip data recorded with special on-board units (OBUs), which were further enriched with climate data, trip elevation profile, and tire conditions (Figure 1). Details about the implementation and the processing toolchain are explained at length in [9], the user interface and preliminary evaluation results are shown below (Figure 3, “Preliminary
Results” section). According to our findings, there is a clear potential to improve operating distance estimation based on the incorporation of various range-limiting factors. In our ongoing research 2) we develop and evaluate now an enhanced range display (add-on for the dashboard) aiming to help drivers to better understand the impact of a) different external factors as well as b) the personal driving style on battery drain or energy consumption. Using a recommendation system with gamification concepts, it should be finally possible to demonstrate an increased operating distance based on a change in the individual driving behavior. Related work Various attempts have been made to measure range-limiting factors of BEVs under both real and laboratory conditions. Single components like batteries were examined in [10] and the performance of BEVs as a whole was researched under varying conditions by [7]. Studies have also shown wide differences between operating distances advertised by car manufacturers (based on standard driving cycles like the New European Driving Cycle, NEDC) and real empirical measurements [8]. Providing more accurate range information is thus essential to reduce range anxiety and customer satisfaction with BEVs. Attempts to compute consumption and remaining range on the basis of additional parameters (like driving speed and heating settings) have been shown already, with variation computed based on theoretical models [5]. In contrast, our approach is to examine a large database of real-world historical driving data and to take for analysis a lot more parameters into account than any other similar project before. The result of knowledge acquisition is an exploration tool [9] that is used to deeply investigate BEV driving characteristics and to identify coherence in thous- ands of BEV trips. This enables us now to implement more fine-grained applications such as improved range displays.
Figure 1: Processing steps executed on each BEV trip to make it usable for later interactive exploration.
Processing and Analysis Toolchain
Step 1. Recording raw trip data: Special on-board units (OBUs) connected to the Internet are used to collect live data (charge state, GPS position, driver ID (via NFC keycards)) from BEVs and store it in our DB.
“Google Maps Directions API” to iteratively eliminate GPS inaccuracies. Once the corrected route is matching the real driving path (Figure 2), all distance-dependent measures (charge state, driven distance, etc.) are mapped back to the right position in the route (method of least-distance or linear mapping). In the last step, each track point is enriched with additional information not available during live recording in the past. Different services are used to gather (historic) climate data from a weather database, elevation/altitude information related to a track and tire conditions (friction, tread patterns) for a specific BEV, and the information is finally merged into the track DB.
Step 2. Track processing: Processing of raw track data is divided into three subtasks, 1) preprocessing, 2) route correction, and 3) merging. In the first step, recorded data for complete tracks are analyzed, outlier or missing values are corrected and corrupt data sets filtered out. To handle inaccurate or missing GPS data of waypoints (e. g., due to signal reflections, GPS outage in tunnels, or too low sampling rate), a route correction algorithm needs to be applied to finally get correct, exact route information. A divide and conquer algorithm is used together with queries of the
Step 3. Interactive exploration tool: Through its interactive multi-view concept, the user interface allows track analysis via three interconnected forms of visualization (Figure 3): The list view (A) displays tracks in a tabular listing with date/time, duration, distance, battery drain (driving efficiency in %/km), weather conditions (temperature, precipitation) and elevation data. A map view (B) based on Google Maps [2] can be used to easily determine spatial properties of tracks and to select track subsets by defining radial filter areas within the map. The visual analysis
With an intuitive interface for exploration and comparison of trips with diverse characteristics, a user can understand and reason about complex data and connections between data sets more easily. To provide a rough understanding about the power and flexibility of our exploration tool (Figure 1), we will briefly explain here the main properties of the system, for details we refer to [9].
Figure 2: A path formed by GPS coordinates originally recorded in the car (top) and the corrected route using Google Maps API (bottom).
view (C), based on the “parallel coordinate visualization” concept (using “Data-Driven Documents” [6]), was implemented to explore (and filter) high-dimensional track data in a two-dimensional graphical representation.
Preliminary Results
battery drain (%/km)
As a foundation and motivation for further research, this section presents some of our findings on BEV range- influencing factors. Using our tool to explore real-world data provided by our partner, an Austrian car-sharing provider [3], allows us to highlight some of the current problems and limitations of BEVs, but also to indicate potentials to increase energy efficiency and extend operating range. The following results are based on recorded trip data from four different vehicles that are running at three different areas in Austria. In total, more than 50 registered people have been driving these cars on a regularly basis from mid 2013. After preprocessing (and outlier removal), more than 4,000 individual tracks remain for further evaluation. For unbiased comparisons, only (about 1,200) tracks of one particular vehicle are used below. all routes
one route
min
0.23
0.64
max
1.92
1.17
mean
0.83
0.89
var
0.05
0.01
stdev
0.23
0.12
avdev
0.17
0.10
Table 2: Basic statistics of average energy consumption/track over all/arbitrary or one specific route.
Simply quantitative statistics (Table 2) indicate tremendous variation in battery drain (expressed in % per km) over all tracks with values ranging from 0.23 to 1.92 %/km (average deviation 0.17 %/km). When comparing the mean energy consumption rates over all routes (all tracks) to those of just one specific route (selected via location-based map filtering), it can be clearly seen that BEV efficiency is highly depending on the driven route. This is not surprising as each route consists of different range-influencing factors: elevation profile (impacts battery drain and recuperation), road condition and surface (friction resistance), speed limits, and general driving dynamics (different on winding roads or roads with frequent traffic jams). In this example, one (hand-picked) route accounts for an average increase in energy consumption of ca. 7 % as compared to the median of
Figure 3: Interface of the web-based interactive exploration tool. Track data is visualized based on three interconnected visualization concepts: (A) list, (B) map, and (C) visual analysis views.
all other routes. The most influential factor is, however presumably outdoor temperature (in combination with tire tread patterns). Figure 4 clearly shows the coherence within a temperature range from −5 to 35 ◦C. Average temperatures and the corresponding battery drain of each track is depicted as a color-coded point to differentiate between summer and winter tires. The average battery drain/km within
battery drain (energy consumption, %/km)
2 1.8
summer tires
1.6
winter tires
coherence of battery drain and driving style (anticipatory driving, utilization of recuperation, avoidance of stop&go).
1.4
Towards Energy-Efficient Driving
1.2 1 0.8 0.6 0.4 0.2
0
-5
0
5
10
15
20
25
30
35
outdoor temperature (°C)
Figure 4: Coherence of energy consumption and outdoor temperatures of real-world BEV tracks (with summer/winter tires).
driver
temperature (°C)
average battery drain (%/km) -5 to 0
1.10
0 to 5
0.98
5 to 10
0.86
10 to 15
0.73
15 to 20
0.69
20 to 25
0.66
25 to 30
0.67
5
0.89
7
0.73
8
1.10
11
1.01
21
0.78
27
0.97
Table 3: Average battery drain based on different temperature intervals or specific drivers. (Battery drain of 1.10 %/km means that 1.10% of the battery power was used for 1 km on average.)
an interval from −5 to 0 ◦C was found to be nearly twice as high than that within 20 to 25 ◦C (Table 3). This coherence might be caused by several reasons: Lower temperatures can have a negative effect on the battery capacity of BEVs [10], reducing operation range and increases the average consumption. Temperature also affects the efficiency of power delivery drop along the wires and we can further assume that heating systems consume much more energy with cold outdoor temperatures [7] as compared to the transitional season. However, in summer time, the A/C system might again consume a lot of energy, reversing the effect for very warm or hot temperatures. Also friction resistance of tires was found to have a huge impact on BEV battery drain and range. Temperature range 9 to 11 ◦C (Figure 4) shows that winter tires account for an additional 9 % battery drain compared to summer tires. Despite the obvious fact that a driver cannot directly influence these factors, there is still a substantial difference in the style of different drivers. Table 3 shows that an economical driver might easily save 30 % energy compared to an inefficient driver. It is our goal to strengthen drivers’ awareness on range-limiting factors and to help understanding
Driven by the initial results, the second aim of this project is to bring the acquired knowledge into the car, providing a more accurate range display and – by using gamification concepts – helping the driver to drive more economically. Our database can be used to derive models and develop systems that improves range estimation (i. e., more accurate predictions based on trips with similar characteristics) and optimizes range itself by encouraging drivers to change their driving styles. In addition, personal accounts and hints to support economic driving shared in the “Hypermiling blog” (http://ecomodder.com/blog) might be a good starting point for concept development. Our next step is to develop a novel dashboard application that analyzes the actual journey and provides a ranking list based on the current energy efficiency compared to other trips (or drivers) with similar characteristics. For meaningful results, comparison must incorporate only historical trips within a narrow range of matching conditions including a similar elevation profile, same weather conditions, similar tire conditions (friction resistance, tread pattern), comparable trip length, etc. This way, drivers should be encouraged to drive more economically by adapting their driving style and/or switching off non-essential electrical appliances. It is not only that we want pique drivers to adapt their driving styles, we basically want to increase the awareness of drivers on the tremendous impact that different internal and external factors have on battery drain and operating range.
Conclusion and Further Work In order to gain knowledge and increase the awareness on the influence of different factors in and around battery electric vehicles (BEVs), an explorative tool to analyze, vi-
sualize, and compare real-world trip data from a car-sharing network was presented. Evaluation results suggest strong coherence of internal and external factors on battery drain (or energy consumption) of BEVs. It was demonstrated, for instance, that average battery drain under moderately cold (−5 ◦C) vs. warm (20 ◦C) outdoor conditions are set apart by almost a factor of two. In addition, there seems to be a huge gap on the average driving economy of drivers based on their individual styles of driving. Comparing the best driver to the worst (under similar conditions) results in a difference of about 50% regarding operating distance (range) – a sound foundation and opportunity for further R&D. Our immediate perspective is to develop (and evaluate) a novel in-car dashboard (Tablet app) to provide driving efficiency rankings based on influencing factors such as weather condition, elevation profile, trip duration, tire conditions, and other vehicle-specific characteristics. In the planned field study we hope – based on the fact that humans naturally tend to compete to each other – to find evidence that accurate range displays and gamification concepts (incentives, ranking lists) can be employed to increase satisfaction as well as the operating distance of electric vehicles.
Acknowledgments We would like to thank IBIOLA Mobility Solutions GmbH [3] for access to both the car-sharing network database and real-time data as well as for their continuous support.
References [1] R. Alvarez, A. Lopez, and N. De la Torre. 2014. Evaluating the effect of a driver’s behaviour on the range of a battery electric vehicle. Proceedings of the Institution of Mechanical Engineers (October 2014), 13. [2] Google Inc. 2015. Google Maps API. [online]. (27 July, 2015). https://developers.google.com/maps/.
[3] IBIOLA Mobility Solutions GmbH. 2015. [online]. (27 July, 2015). http://www.ibiola-mobility.com/. [4] M. Jung, D. Sirkin, T. Gür, and M. Steinert. 2015. Displayed Uncertainty Improves Driving Experience and Behavior: The Case of Range Anxiety in an Electric Car. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (CHI ’15). ACM, New York, NY, USA, 2201–2210. [5] A. Lundström. 2014. Differentiated Driving Range: Exploring a Solution to the Problems with the GuessO- Meter in Electric Cars. In Adjunct Proceedings of AutomotiveUI 2014. ACM, Seattle, US, pp. 8. [6] M. Bostock. 2015. Data-Driven Documents. [online]. (27 July, 2015). http://d3js.org/. [7] N. Meyer, I. Whittal, M. Christenson, and A. LoiselleLapointe. 2012. The impact of the driving cycle and climate on electrical consumption and range of fully electric passengers vehicles. In Proceedings of EVS, Vol. 26. Los Angeles, CA, US, pp. 11. [8] P. Mock, J. German, and A. Bandivadekar et al. 2013. From Laboratory to Road: A Comparison of Official and ‘Real-world’ Fuel Consumption and CO2 Values for Cars in Europe and the United States. White paper. Int. Council on Clean Transportation (ICCT). pp. 88. [9] B. Pichler and A. Riener. 2015. An Interactive Exploration Tool for Detailed E-Vehicle Range Analysis. In Adjunct proceedings of Mensch & Computer 2015, Workshop Automotive HMI. Walter de Gruyter Verlag GmbH, Stuttgart, Germany, pp. 8. [10] Y. Zhang and C. Wang. 2009. Cycle-life Characterization of Automotive Lithium-ion Batteries with LiNiO2 Cathode. Journal of the Electrochemical Society 156, 7 (2009), A527–A535.