Residential Demand Response: Dynamic Energy ...

IEEE TRANSACTION ON POWER SYSTEMS

1

Residential Demand Response: Dynamic Energy Management and Time-Varying Electricity Pricing Matteo Muratori, and Giorgio Rizzoni, Fellow, IEEE,

Abstract—Demand response programs are currently being proposed as a solution to deal with issues related to peak demand, and to improve the operation of the electric power system. In the demand response paradigm electric utilities provide incentives and benefits to private consumers as a compensation for their flexibility in the timing of their electricity consumption. In this paper a dynamic energy management framework, based on highly-resolved energy consumption models, is used to simulate automated residential demand response. The models estimate the residential demand using a novel bottom-up approach that quantifies consumer energy use behavior, thus providing an accurate estimation of the actual amount of controllable resources. The optimal schedule of all the controllable appliances, including plug-in electric vehicles, is found by minimizing consumer electricity-related expenditures. Recently, time-varying electricity rate plans have been proposed by electric utilities as an incentive to their customers with the objective of re-shaping the aggregate demand. Large-scale simulations are performed to analyze and quantitatively assess the impact of demand response programs using different electricity price structures. Results show that simple time-varying electricity price structures, coupled with large-scale adoption of automated energy management systems, might create pronounced rebound peaks in the aggregate residential demand. To cope with the rebound peaks created by the synchronization of the individual residential demands, innovative electricity price structures – called Multi-TOU and Multi-CPP – are proposed. Index Terms—Demand Response; Residential Energy Management; Electricity Pricing; TOU and CPP; Rebound Peaks.

I. I NTRODUCTION Nowadays, to match peak demand, follow seasonal and daily fluctuations, and ensure reliable operation of the electric power system, utilities are forced to maintain a substantial amount of underutilized power capacity. This capacity, often outdated and environmentally harmful, drives the cost of electricity as indicated by the exponential increase in wholesale electricity price during peak operation [1]. Instead of adapting electricity generation to match changes in demand, the demand itself could be made more flexible to reduce requirements on the electric power generation infrastructure and allow for an easier integration of non-dispatchable resources. Demand response is a promising techno-economical solution to make electricity demand more flexible, allowing private customers to modify their demand profiles to fit the needs of energy supply. In the demand response paradigm, This work has been performed at The Ohio State University – Center for Automotive Research supported by the National Science Foundation under Grant No. 1029337. The authors would like to thanks Prof. W. Zhang, Dr. E. Serra, and C.Y. Chang for their contribution to the early development of this study, and Prof. R. Sioshansi for his thorough review. Contact information: [email protected], [email protected]. Manuscript received July 18, 2014; revised November 04, 2014.

electric utilities provide some sort of incentive to their residential customers as a compensation for their flexibility in the timing of their energy consumption. Utilities also provide a signal to their customers (typically electricity price) that is intended to guide the power consumption so as to obtain an aggregate demand that better matches the needs of the power generation. Demand response has proved effective at shifting consumption away from peak hours, thus increasing system efficiency and stability, reducing the need for investment in peaking generation, and bringing several environmental and financial benefits [2]. A power system equipped with demand response capabilities can lead to a reduction in systems costs, CO2 emissions, and price volatility by shifting power consumption to periods characterized by low prices and high renewable power production [3]. The goal of demand response programs is to influence consumers to change their demand, in response to the needs of the supplier [4]. To achieve this objective the proper signal must be sent to the final customers. In order to develop proper electricity price structures advanced modeling, simulation, and optimization tools are needed to properly analyze a complex system that includes interactions between humans, energy infrastructures, and local conditions. In this paper, a state-of-the-art dynamic energy management framework is used to evaluate the potential of residential demand response. The dynamic energy management framework is based on highly-resolved personal energy consumption models developed using a novel bottom-up approach that quantifies consumer energy use behavior in the United States [5]. The highly-resolved personal energy consumption models capture the entire energy footprint of American households, including energy consumption for personal mobility. The dynamic energy management framework simultaneously optimizes the scheduling of controllable appliances and in-home charging of Plug-in Electric Vehicles (PEVs). The automated dynamic energy management framework introduced in this paper is decentralized, in the sense that each single household receives a signal from the electric utility, and independently optimizes its own demand. Even though this leads to a local optimum, the signal sent from the electric utility can be developed in such a way as to achieve one or more system-level objectives, such as reduce electricity generation and grid operation costs for electric utilities; manage demand peaks (better interaction between demand and generation); reduce overall pollutant and carbon dioxide emissions from electricity generation; improve overall grid efficiency and minimize primary energy consumption; or others. The study proposed in this paper improves existing resi-


dential energy management systems in five important ways. First, the optimization framework is based on highly-resolved models that capture consumer behavior. The incorporation of stochastic consumer behaviors provides more accurate estimation of the actual amount of available controllable resources, allowing for a better understanding of the potential of residential demand response programs. Second, the model captures the entire residential electricity consumption and the recharging of plug-in electric vehicles, allowing for estimating the impact of the gradual electrification of the fleet of passenger vehicles on demand response. Third, a realistic scenario of residential demand response deployment is simulated, where no coordination among consumers or direct control from the electric utilities are assumed, in line with the distributed nature of smart grids. Also, simple, transparent, and deterministic (i.e. known a priori and predictable by the customers) time-varying pricing schemes coupled to automated management systems are considered in this study, a required condition for consumer acceptance [6]. Forth, the model is intended to capture systemlevel impact of demand response, rather than local effects on a single consumer. Large-scale simulations are performed to explore and evaluate the impact of different electricity price structures on the aggregate residential electricity demand. Fifth, simulation results confirms prior evidence that demand synchronization among consumers might be created by the introduction of automated energy management systems when time-varying electricity pricing is used (phenomenon known as “rebound peaks”). To cope with these issues, innovative electricity price structures based on group pricing are proposed in this paper, called multi-time-of-use (Multi-TOU) and multicritical peak pricing (Multi-CPP). II. D EMAND R ESPONSE AND T IME -VARYING E LECTRICITY P RICING Demand response models are based on the assumption that demand is elastic, and that consumers respond to higher electricity price by changing their demand (in particular, the timing of their electricity consumption) to reduce electricity-related expenditures. Studies have shown that under certain conditions and in some markets this is the case. Espey and Espey [7] summarize several studies of residential electricity demand elasticities, confirming the elasticity of residential demand for electricity in the United States. Caves and Christensen [8] use data from five experimental implementations of residential demand response in the United States, and concluded that customers responded to higher prices during the peak period by reducing peak period usage and/or shifting it to less expensive off-peak periods. Torriti shows that Italian customers, when proposed with time-varying electricity price, also respond with a significant load shift. In this example demand response leads to higher average electricity consumption and lower payments by consumers [9]. In 2011 the demand response market in the United States generated approximately $6 billion in direct revenues for local businesses, industry, and households as well as enabling avoided investment costs. In the PJM interconnection, demand response allowed for cutting 7% of their seasonal peak [2].

2

Currently, demand response programs focus mainly on industrial and large commercial consumers, with fixed compensations attributed to small numbers of large end-users, using direct load control and interruptible loads. Walawalkar et al. [10] predict that in order for demand response to scale up, and enable greater level of renewable integration, technologies that are suitable for participating every day to demand response are needed. The vast potential of building energy management strategies is still untapped, not to mention future possibilities as home automation and smart appliances become standardized along with deployment of smart grid infrastructure [10]. Thus, in the future, large numbers of end-users, including small commercial customers and residential households could be involved in demand response programs, involving deliberate shifts in electricity demand in correspondence with peak loads (and thus high electricity price) [11]. Controllable appliances and intelligent demand-side energy management platforms make residential demand response an attractive option for residential customers, since these technologies allow for an automatic response, without requiring direct monitoring of the electricity price by a user and limiting the inconvenience intrinsic in changing the timing of the electricity consumption. Other than overall reduction of electricityrelated expenditures, residential customer would also benefit from a series of advantages deriving from a better operation of the system, including reduced power sags and interruptions, better service continuity and reliability, and improved power quality (reduced voltage and frequency variations and transients phenomena). Customer acceptance is a conditio sine qua non for the effective deployment of demand response programs. Empirical results show that consumers are open to dynamic pricing, but prefer simple programs to complex and highly dynamic ones [6]. Also, smart home technologies allowing for response automation (of the kind assumed in this study) are seen as a prerequisite for an active participation of residential customers in demand response programs [6]. Studies performed at EPRI1 identified automated costumer-side equipment able to intelligently monitor and manage residential energy use as a critical enabling technology needed to move toward realizing a smart grid [12]. Time-varying electricity price is the signal adopted by electric utilities to influence and guide residential energy consumption. The most common pricing structures include Time of Use pricing (TOU), Critical Peak Pricing (CPP), Peak Time Rebates (PTR), or Real Time Pricing (RTP) [13], [14], [15], [16]. TOU rates are defined as different electricity prices for different periods of the day or of the year. TOU programs can be characterized by two or more price tiers. CPP is essentially a TOU program, with a significantly higher price tier during peak periods. The objective of CPP is to enhance a TOU price structure with the ability of promptly responding to emergency events. Two variants of this type of price structure exist: one where the time and duration of the peak periods 1 The Electric Power Research Institute (EPRI) is an independent, nonprofit company performing research, development and demonstration in the electricity sector.


are predetermined and another where the time and duration of the peak periods may vary based on the needs of the electric power infrastructure. In this case customers are notified a few hours prior to a critical peak event, that typically lasts for no more than a few hours. Outside these peak periods TOU prices are typically in effect. In the PTR paradigm customers receive electricity bill rebates for reducing their electricity consumption during peak periods (established a priori by the electric utility) relative to a previously established baseline, which is determined for each individual customer. The baseline is usually identified using historical household electricity consumption. Wolak notes that PTR provides an incentive for house-holders to elevate their electricity use during the period in which baselines are established, and found evidence for such behavior [17]. A study conducted at MIT finds that demand response programs that pay customers for reducing consumption from a baseline generally provide excessive compensation and give customers incentives for strategic behavior [18]. RTP is a highly dynamic electricity price structure that adjusts electricity prices on an ongoing basis throughout the day, following the wholesale electricity generation cost, on an hourly or sub-hourly basis (generally 10-15 minutes time intervals). RTP is intended to convey actual generation cost to the final consumer allowing for optimal use of generation resources. Navigant Research estimates less than 3% of U.S. residential customers have access to time-varying pricing today and well below 1% have actually adopted it. In a best-case scenario, as predicted by Navigant, time-varying pricing will be available to 60% of residential customers in the U.S. by 2020, with about 20% participating to demand response [19]. Muratori et al. [1] review in dept the role of residential demand response in modern electricity markets and show that simple technical solutions (like time-varying pricing associated with decentralized automated energy management) may lead to undesired demand dynamics. LeMay et al. [20] suggest that there is a threat of rebound peaks in which consumers delay their demands to avoid a peak, but cause a new peak when trying to satisfy delayed demand. Similar trends are observed in an experiment performed by Pacific Gas and Electric to monitor the substation-level load impacts of end-use load control [21]. Lenhoff et al. [22] reports that if many consumers react to time-varying electricity pricing in an un-coordinated manner, the coincidence factor of load increases significantly and the electric system may face strongly increased load fluctuations. Mishra et al. [23] suggest that current pricing plans incentivize all consumers to shift their energy consumption during low-price periods. Thus, at large scales, simultaneous energy request during off-peak periods will trigger rebound peaks if prices do not change to reflect the resulting increases in off-peak demand. Load “pickup” effects and formation of rebound peaks are observed in several studies and pilot projects. For example, results from the EV Project2 show how the introduction of 2 The EV Project is the largest deployment of electric vehicle charge infrastructure in history. [online] available: http://www.theevproject.com/

3

a TOU rate plan does not necessarily prevent the charging demand from peaking [24]. If anything, the TOU rate plan increases and shifts the peak demand, as shown in Figure 1, where weekdays-charging demand profiles normalized per Electric Vehicle Supply Equipment (EVSE) for two locations are reported. Figure 1a shows how in Nashville, where the project participants do not have access to any TOU rate, the charging demand starts to increase gradually after 4 p.m. and peaks around 8 p.m. when residents are mostly at home. On the other hand, Figure 1b shows charging demand profiles in San Francisco where a special TOU rate plan is available. In this case, the introduction of the TOU rate plan has the effect of synchronizing the demand exactly at the moment when electricity price changes (at midnight). Therefore, the demand increase is steeper and presents a higher peak value (note the different scale on the y-axis). This example shows that TOU electricity plan can be successful at shifting the electricity demand, but could also incurs in large demand rebound peaks when the electricity price changes.

(a) Average charging demand with flat electricity price.

(b) Average charging demand with TOU electricity price.

Fig. 1: Weekday average time-of-day charging demand of Electric Vehicle Supply Equipment (EVSE). From The EV Project reports. The tools proposed in this paper allow studying and comparing the impact of different electricity price structures on automated distributed residential demand response. III. DYNAMIC E NERGY M ANAGEMENT F RAMEWORK In this paper, a state-of-the-art Dynamic Energy Management (DEM) tool is used to simulate automated decentralized residential demand response [25]. The proposed energy management framework relies on highly-resolved models of all the components, providing a more accurate estimation of the actual amount of available controllable resources. A detailed Residential Energy Eco-System (REES) model, developed at The Ohio State University [5], is used to simulate


4

residential electricity demand. The REES model captures the entire energy footprint of American households, to include all appliances, lighting, HVAC system, and in-home charging of plug-in electric vehicles, viewing residential and transportation energy needs as an integrated continuum. The model is based on a novel bottom-up approach that simulates and quantifies consumer energy use behaviors. The REES model is based on the integration of highly-resolved (10-minute resolution) bottom-up models for residential [26] and personal transportation [27] energy consumption in the United States. Residential power profiles are proven to have the same statistical features of metered data and show typical diurnal and seasonal patterns (for more details on the residential power demand model refer to Muratori et al. [26]). As an example, Figure 2 reports the per-household average weekly electric power profile of 100 residential households. This large-scale simulation is aimed at representing the total residential electric load of a heterogeneous group of households and related PEVs (the 100 households differ in terms of size, insulation, and number and demographic characteristics of household members). Two scenarios are depicted in the figure: a case where no PEVs are deployed (reference scenario representing the current status quo), and a second case assuming a 10% PEV market share, equally subdivided between PHEVs and EVs.3

Average Electric Power Demand [W]

2,500

REES model with 10% market penetration of PEVs REES model with conventional vehicles

2,000

1,500

1,000

500

0:00

12:00

Monday

24:00

12:00

24:00

Tuesday

12:00

24:00

Wednesday

12:00

24:00

Thursday

12:00

Friday

24:00

12:00

24:00

Saturday

12:00

24:00

Sunday

Fig. 2: Sample output of the REES model for a winter week in the Midwest region off the United States. Figure 2 clearly shows the typical trait of aggregate residential electricity demand, characterized by significant daily fluctuations. Significant seasonal variations are also present, which are not depicted in the figure. Starting from the REES model, an optimal control problem for the scheduling of all the controllable appliances (laundry machine, dryer, and dishwasher) and in-home charging of plug-in electric vehicles is formulated. The management problem is numerically solved using Dynamic Programming (DP), and considers the behavior of household members and their energy consumption, as predicted by the REES model. Residential energy management has been studied extensively in the literature. Kirschen [29] discusses demand-side 3 A PEV is defined by the U.S. Department of Energy as a vehicle that draws electricity from a battery with a capacity of at least 4 kWh and is capable of being charged from an external source [28]. The definition of PEV includes Plug-in Hybrid Electric Vehicles (PHEVs) and Electric Vehicles (EVs). In this paper a battery capacity of 16 kWh and 24 kWh is assumed for PHEVs and EVs, respectively.

aspects of electricity markets and concludes that enhancing the ability of the demand for electricity to respond to price signals could benefit not only the consumers who choose to participate actively in electricity markets, but would also help these markets operate more efficiently. The expected time-varying retail price motivates the studies of various algorithms to schedule smart appliances to minimize cost for the users. Mohesenian-Rad and Leon-Garcia [30] propose a residential energy consumption scheduling framework which attempts to achieve a desired trade-off between minimizing the electricity payment and minimizing the waiting time for the operation of each appliance in household in presence of a real-time pricing tariff by doing price prediction based on prior knowledge. Sou et al. [31] use mixed integer programming to minimize the electricity cost scheduling smart appliances. Kim and Poor [32] use stochastic dynamic programming to solve an appliance scheduling problem assuming statistical knowledge of future electricity price. The capability of scheduling thermostatically-controlled loads is at the core of several appliance commitment algorithms prosed in the literature. Du and Lu [33] propose an algorithm to control a water heater based on price and consumption forecasts considering user comfort settings to meet an optimization objective such as minimum payment or maximum comfort. Avci et al. [34] propose a two-stage cost and energy efficient HVAC load control strategy under a dynamic pricing. Livengood and Larson [35] describe a prototype model to illustrate the use of approximate stochastic dynamic programming to optimally control a few residential devices accounting for uncertain weather forecast. Nevertheless, these approaches do not focus specifically on solving a decentralized (not coordinated) residential energy management problem, and do not capture properly stochastic user behavior. Moreover, the study proposed in this paper is intended to capture system-level impact of demand response, rather than local effects on a single consumer. In this paper a dynamic energy management framework is proposed to simultaneously find the optimal schedule for all the controllable appliances and in-home charging of PEVs. The framework proposed is flexible enough that different cost functions (e.g. minimization of carbon footprint) and different price structures (e.g. TOU, CPP, or others) can be easily simulated to reproduce different policy decisions and evaluate their impact on the electricity demand. In this paper the optimization framework is aimed at minimizing cost for households owners. For each controllable appliance the enabling time E i (time at which a user enables the ith controllable appliance), the completion time C i (time required to complete the run-cycle of the ith controllable appliance), the deadline Di (time at which the user requires the ith controllable appliance run-cycle to be completed), and the maximum waiting time W i (time that can be waited before the run-cycle of the ith controllable appliance must start to match the desired deadline) are predicted by the REES model, and could differ for different appliances executions or charging events. Deadlines are set by the user and may be the result of a compromise between cost and


convenience. In this study it has been assumed that each PEV must finish charging before the next driving event, when possible, while dish-washing and laundry appliances are given an 8-hour window to complete their execution.4 The state vector of the optimal control problem, namely xt , is defined to capture the timing dynamics of the controllable appliances that are used to solve the control problem. xa is a vector including information on which activity is currently performed by each of the N household’s members, as predicted by the REES model. The controllable appliances timing dynamics, xt , are different for non-interruptible and interruptible appliances. In the former case, the timing dynamics are represented by a counter, as described by Equation 1.  −W i          1 i xit (k) + 1 xt (k+1) =       xi (k) + 1    t −∞

if (xa (k) = i) ∧ ui (k) = 0 ∧(xit (k) =−∞) if xit (k) < 0 ∧ ui (k) =1 if −∞ < xit (k)≤ 0 ∧ ui (k) = 0 if 0 < xit (k) 0), this value represents the portion of the appliance execution that has been already completed, terminating with C i , the completion time required to complete the run-cycle of the ith non-interruptible appliance. When xit = C i the job is completed and the state is re-initialized: xit = −∞. For all the interruptible appliances, the timing dynamics are represented by two counters, an up counter and a down counter, as reported by Equation 2.  i  xt1 (k) xi (k) + 1 xit1 (k + 1) =  t1 0

xit2 (k + 1) =

 −W i          −(W i + 1)   xit2 (k)     xi (k) + 1    t2 −∞

if ui (k) = 0 if ui (k) = 1 if xit1 (k) = C i if (xa (k) = i) ∧ ui (k) = 0 ∧(xit2 (k) = −∞) if (xa (k) = i) ∧ ui (k) = 1 ∧(xit2 (k) = −∞) if ui (k) = 1 if ui (k) = 0 if xit1 (k) = C i (2)

4 Laundry is modeled as two separate parts: washing and drying [26]. The latter has been constrained in this framework not to start until the former has been completed.

5

The first part of Equation 2, xit1 , is an up counter that increases whenever the appliance is on (e.g. battery of a plug-in electric vehicle is being charged), until reaching the completion time C i required to complete the run-cycle (fully charging PEV battery) of the ith interruptible controllable appliance. When xit1 = C i the job is completed and the states are re-initialized: xit1 = 0; xit2 = −∞. The second part of Equation 2, xit1 , is a down counter that, starting from the maximum waiting time −W i , increases until reaching the value of 0, threshold at which the appliance (e.g. PEV charging station) must start running in order to complete its run-cycle before the deadline. The state vector is initialized so that xit (0) = −∞ for all the non-interruptible controllable appliances and xit1 (0) = 0, xit2 (0) = −∞ for all the interruptible appliances. An exogenous signal, sent by the electric utility, is needed as input to the DEM tool. In this case, the price of electricity p(k) is used to define the cost function in the following form:

M

g(k) =

1 X i u (k) · p(k) · Υi 6 i=1

(3)

where M is the number of controllable appliances, ui is the control variable of the ith appliance (u = 1 when the appliance is running), and Υi is the wattage of the ith appliance, expressed in [W]. Since the problem has been implemented on a 10-minute time basis, the term 1 /6 is used to convert the electric energy consumption to [kWh]. The cost function described by Equation 3 represents a scenario in which electric utilities send a time-varying electricity price signal to residential customers as a way to steer the demand. Each consumer receives a price signal and tries to optimally manage his/her electricity consumption to minimize cost. The optimal control policy µ∗ (k, xt (k)) is defined as the optimal control u∗ (k) with the initial state being xt (k), and it is found by solving the cost-to-go function, Vk (xt (k)), as shown in Equation 4, where gk is the cost function reported in Equation 3 evaluated at time k. DP method solves µ∗ (k, xt (k)) by using backwards induction. The cost-to-go function at time K (VK (xt (K))) is exactly gK (xt (K)). The algorithm proceeds from k = K −1 to k = 0 considering every possible state xt (k) per each time-step k, finding the optimal control policy µ∗ (k, xt (k)) for every statetime pair. The proposed automated optimization framework finds the optimal schedule for all the controllable appliances so as to minimize the cost function—based on the dynamic programming algorithm. When finding the optimal schedule for the controllable appliances, more than one solution may result in the same minimum cost. If this is the case, the dynamic energy management framework implemented in this study schedules the controllable appliances to run as early as possible during lowest price periods, so as to minimize customer inconvenience.


6

( min

E

u(i)∈U (xt (i)),k≤i≤K−1

IV. I MPACT OF D IFFERENT E LECTRICITY P RICE S TRUCTURES ON AUTOMATED R ESIDENTIAL D EMAND R ESPONSE Recently, time-of-use rate plans have been proposed to residential costumers with the objective of re-shaping the aggregate demand. Nevertheless, several studies and smart grid demonstration projects show that, when each household independently optimizes its demand leveraging off-peak electricity prices in order to reduce its own cost, the resulting aggregate demand may be affected by an even higher rebound peak shifted toward the off-peak period (see Figure 1). The modeling proposed in this study allows for exploring the implications of automated residential demand response, considering the impact of time-varying electricity pricing on residential demand response. The goal of the simulations presented in this section is to assess the impact of widespread adoption of decentralized automated residential demand response programs, assuming that all residential customers adopt the DEM framework proposed in Section III (each single customer optimizes his own energy consumption so as to minimize cost and customer inconvenience). The impact of electricity price structures on the aggregate residential demand is analyzed and quantitatively assessed via large-scale simulations. First, a two-tier TOU electricity pricing system is considered, where price of electricity is high between 7 a.m. and 10 p.m. (High-Tier Price), and low between 10 p.m. and 7 a.m. (Low-Tier Price). Figure 3 shows the same simulation depicted in Figure 2, after each household optimizes its own demand to minimize the electricity-related expenditure assuming TOU electricity price.


2,500

REES model with 10% market penetration of PEVs REES model with conventional vehicles Electricity Price

2,000

1,500

1,000

500 High−Tier Price Low−Tier Price 0:00

12:00

Monday

24:00

12:00

24:00

Tuesday

12:00

24:00

Wednesday

12:00

24:00

Thursday

12:00

Friday

24:00

12:00

24:00

Saturday

(4)

E {gk (xt (k), µ(k, xt (k))) + Vk+1 (fk (xt (k), µ(k, xt (k))))}

12:00

24:00

Sunday

Fig. 3: Aggregate residential electric power demand for a winter week in the Midwest region of the United States assuming TOU electricity price and automated demand response. Even though the demand has been deferred towards low cost periods, thus filling the load valleys (this phenomenon would be accentuated if more loads where deferrable or distributed energy storage were present), Figure 3 also shows

that whenever the electricity price drops a peak appears in the aggregate demand. Such peaks, called rebound peaks, are higher and steeper than the peaks originally present in the residential demand that the TOU electricity pricing structure was intended to eliminate. This occurs because all the deferrable activities wait for the price to drop before starting, leading to a contemporaneous request of power when the electricity price changes. Figure 4 shows the impact of the considered TOU electricity price structure on the statistical distribution of the electricity demand when an automated energy management system is introduced. REES Model with Conventional Vehicles 20 15 10 5

500

Optimized REES Model with Conventional Vehicles Percentage of time

u(k)∈U (xt (k))

gk (xt (i), µ(i, xt (i)))

1000

1500

2000

2500

20 15 10 5

500

REES Model with 10% PEVs 20 15 10 5

500

1000

1500

2000


1000

1500

2000

2500

Optimized REES Model with 10% PEVs Percentage of time

min

)

i=k

Percentage of time

=

gK (xt (K)) +

Percentage of time

Vk (xt (k)) =

K−1 X

2500

20 15 10 5

500

1000

1500

2000

2500


Fig. 4: Distribution of the average aggregate electric power demand assuming TOU electricity price.

Whether or not plug-in electric vehicles are present, dynamic energy management changes the statistical distribution of the residential electricity demand, with the effect of shrinking the demand towards lower power regions, while introducing peaks of higher demand that were not present before. Overall, the introduction of TOU electricity pricing eliminates the smoothing effect due to the natural stochastic features of residential demand, forcing demand synchronization among all the REESs. Even though the demand is being effectively deferred toward night periods, results show that pronounced rebound peaks are created in the aggregate demand. The introduction of CPP shows effects similar to those obtained using TOU electricity rates. Still, adopting CPP electricity pricing, electric utilities have the opportunity of calling a critical peak event, with the objective of drastically reducing the demand and relieving the electric power infrastructure in case of emergency. This is shown in Figure 5, where the same simulation depicted in Figure 2 is reported, but a CPP electricity price structure is assumed and each household optimizes its own demand to minimize the electricity-related expenditure. The CPP structure is identical to the TOU pricing previously


7

considered, but a critical event (Critical Peak Price) is called during the highest demand peak, namely between 9:00 p.m. and 11:30 p.m. of Saturday night for the representative week shown in Figure 5.


2,500

REES model with 10% market penetration of PEVs REES model with conventional vehicles Electricity Price

2,000

1,500

1,000

Critical Peak Price

500

High−Tier Price Low−Tier Price 0:00

12:00

Monday

24:00

12:00

24:00

Tuesday

12:00

24:00

Wednesday

12:00

24:00

Thursday

12:00

Friday

24:00

12:00

24:00

Saturday

12:00

24:00

Sunday

Fig. 5: Aggregate residential electric power demand for a winter week in the Midwest region of the United States assuming CPP electricity price and automated demand response. The introduction of a critical event is proven to be effective at removing one of the peaks in the demand, which is essential for guaranteeing the reliable operation of the electric grid in case of emergency. Overall, CPP can be considered as a case of TOU price structure, leading to the creation of similar rebound peaks, but with the ability of effectively dealing with a few emergency events during the year. Table I summarizes the statistical features of the perhousehold aggregate residential electricity demand (averaged over the total number of households) in the six scenarios considered so far: residential eco-systems with only conventional vehicles (REES CV) versus residential eco-systems with 10% market penetration of plug-in electric vehicles (REES 10% PEV) assuming flat electricity price, two-tier TOU price, and critical peak price (CPP). In the table, α(x) is a severity term, introduced in this study, defined as the percentage of time during which the demand is higher that x times the average demand. For example, α(1.5) represents the percentage of time during which the aggregate residential demand was higher than 1.5 times its average. This severity term can be used to rapidly evaluate the “peakiness” of a power profile. The results summarized in Table I suggest that when an automated decentralized (un-coordinated) dynamic energy management system is largely adopted by residential customers, the introduction of TOU electricity price decreases the overall demand variability and pushes the demand toward the average load. α(1.25) is significantly reduced compared to the reference case of flat electricity price. This partially achieves the provider’s goal of having a residential demand as flat as possible, filling demand valleys and reducing peaks. Nevertheless, higher demand peaks are introduced, and α(1.75) increases compared to the reference case. The adoption of a CPP electricity price produces results similar to those achieved with time-of-use rates, but CPP can be used as a means for effectively dealing with a few emergency events. These effects are accentuated by the presence of plugin electric vehicles, that represent a substantial portion of deferrable load.

The main goal in the deployment of demand response programs is to smooth consumption and reduce overall “peakiness” of the aggregate demand. However, as the results show, introduction of either TOU or CPP rates in a system where each individual consumer automatically optimizes demand responding to the same signal may actually exacerbate the very problem that the demand response program was designed to address. Still, CPP rates are proven to be extremely effective in dealing with a limited number of emergencies. Different techno-economic solutions can be used to cope with the rebound peaks created by simple time-varying electricity price structures that lead to the synchronization of the individual residential demands. For example, electricity price structures based on group pricing can be used. In such a structure the customers are divided in different groups, and each group is proposed with a different pricing scheme (e.g. the same time-varying pricing staggered in time) so as to avoid the synchronization of the aggregate demand. Alternatively, the starting time of the controllable appliances could be randomized within low-price periods, or some sort of timebased or location-based randomization can be added to the price signal. The development of effective demand response programs results from a compromise between costumers’ acceptance (i.e. keeping the electricity price simple and predictable) and achieving the system-level objectives of the electric utilities. In this paper the use of deterministic electricity price structures based on group pricing, called multi-time-of-use (Multi-TOU) and multi-critical peak pricing (Multi-CPP), is explored. Due to their intrinsically simple structure and predictability these pricing schemes are more likely to be adopted by American residential customers during the early roll-out of demand response programs, while making demand response appropriate for increasingly complex supply–demand balancing. A MultiTOU structure is essentially a TOU structure in which the residential customers are divided into groups. Each group sees different TOU electricity prices, leading to a multi time-of-use price structure [5]. In particular, a two-tier TOU electricity pricing system is assumed, where the price of electricity is high during day hours (High-Tier Price), and low during night hours (Low-Tier Price). The moment at which price changes between low and high tier is deferred by one hour going from one group of customers to the following. For example, for the first group of customers the daily (Low-Tier) price goes from 6 a.m. to 9 p.m., for the second group it goes from 7 a.m. to 10 p.m. and so forth. In this way each customer sees a TOU rate, but not all the customers are synchronized. To avoid disparities between customers, each can rotate among the four groups during the year, so that each residential customer is faced with the same overall electricity price during a calendar year. Results of the introduction of this Multi-TOU rates are shown in Figure 6, in which the simulations described in Figure 2 are repeated, after each household optimizes its own demand in order to minimize the electricity-related expenditure assuming Multi-TOU electricity price. The adoption of Multi-TOU pricing does not lead to the creation of any rebound peak when automated energy manage-


8

TABLE I: Statistical features of the per-household aggregated residential electricity demand in the six scenarios considered. Scenario REES REES REES REES REES REES

St. Dev. [W]

Min. [kW]

Max. [kW]

α(1.25) [-]

α(1.75) [-]

1.02 1.02 1.02 1.08 1.08 1.08

270 257 287 305 296 326

0.53 0.52 0.52 0.54 0.54 0.56

1.70 2.02 2.18 1.88 2.28 2.36

20.1 9.9 15.2 22.8 9.2 14.8

0 0.9 1.5 0 3.0 3.3

CV CV TOU Price CV CPP Price 10% PEV 10% PEV TOU Price 10% PEV CPP Price

REES model with 10% market penetration of PEVs REES model with conventional vehicles Electricity Price − Group 1 Electricity Price − Group 2 Electricity Price − Group 3 Electricity Price − Group 4

2,500


Mean [kW]

2,000

1,500

1,000

500 High−Tier Price Low−Tier Price 0:00

12:00

24:00

Monday

12:00

24:00

Tuesday

12:00

24:00

Wednesday

12:00

24:00

12:00

24:00

Friday

Thursday

12:00

24:00

Saturday

12:00

24:00

Sunday

Fig. 6: Aggregate residential electric power demand for a winter week in the Midwest region of the United States assuming Multi-TOU electricity price and automated demand response.

ment is introduced. Moreover, the natural “peakiness” of the demand shown in Figure 2 appears to be significantly reduced, leading to a smoother aggregate demand. This achieves the objective for which residential demand response programs are designed, which is to alleviate the requirements on the electric system to follow the fluctuations in the demand. Figure 7 shows the impact of Multi-TOU electricity price on the statistical distribution of the electricity demand when an automated energy management system is introduced.

15 10 5

500

Optimized REES Model with Conventional Vehicles Percentage of time

Percentage of time

REES Model with Conventional Vehicles 20

1000

1500

2000

2500

20 15 10 5

500

20 15 10 5

500

1000

1500

2000


1000

1500

2000

2500

Optimized REES Model with 10% PEVs Percentage of time

Percentage of time

REES Model with 10% PEVs

2500

20 15 10 5

500

1000

1500

2000

2500


Fig. 7: Distribution of the average aggregate electric power demand assuming Multi-TOU electricity price. Figure 7 confirms that the introduction of a Multi-TOU electricity price is effective in reducing the variability in the demand, with the effect of shrinking the demand towards lower power regions. In this case no rebound peak is introduced. Table II summarizes the statistical features of the perhousehold aggregate residential electricity demand (averaged

over the total number of households) when Multi-TOU electricity rates are introduced. From the table it appears that when an automated decentralized (un-coordinated) dynamic energy management system is introduced, the adoption of multi-time-of-use electricity pricing (Multi-TOU) decreases the overall demand variability and pushes the demand toward the average load. α(1.25) is significantly reduced compared to the reference case of flat electricity price. Also, peaks in the demand are significantly reduced, and α(1.75) is maintained at 0. The maximum demand during the simulated week is reduced by 12% when no PEV are considered, and by 17% when a PEV market share of 10% is considered. The simulated results suggest that Multi-TOU electricity pricing could be an effective policy to achieve the objective of electric utilities to have residential demand as flat as possible, filling demand valleys and reducing peaks. A multi-CPP price structure has the same effects, also providing the ability of dealing with emergency events, as shown previously. V. C ONCLUSIONS In this paper an automated dynamic energy management framework is proposed to find the optimal schedule of residential controllable appliances, including in-home charging of plug-in electric vehicles. The management framework is decentralized, in that each individual household independently optimizes its own demand, without any coordination among different consumers. The optimal control problem is solved using dynamic programming, finding the global solution that minimizes a cost function. The algorithm is general and different cost function could be selected to achieve different objectives. For the purpose of simulating demand response programs the cost function is chosen to minimize consumer electricity-related expenditure, thus providing some sort of compensation to the residential costumers for their flexibility in timing the energy consumption. This work is among the first that systematically considers stochastic correlations among different end-use activities in the design of the energy management framework [5]. The incorporation of stochastic consumer behaviors provides more accurate estimation of the actual amount of available controllable resources, and hence enables better estimation of the impact of automated demand response. Moreover, the proposed study of integrated load modeling for both PEVs and other smart appliances represent a significant advance over most existing works in aggregate load modeling that have mainly focused on first-order thermostatically controlled loads. Large-scale simulations are performed to explore and evaluate the impact of different electricity price structures on the


9

TABLE II: Statistical features of the per-household aggregated residential electricity demand assuming Multi-TOU electricity price. Scenario REES REES REES REES

CV CV Multi-TOU Price 10% PEV 10% PEV Multi-TOU Price

Mean [kW]

St. Dev. [W]

Min. [kW]

Max. [kW]

α(1.25) [-]

α(1.75) [-]

1.02 1.02 1.08 1.08

270 214 305 211

0.53 0.52 0.54 0.58

1.70 1.50 1.88 1.56

20.1 12.0 22.8 9.0

0 0 0 0

aggregate residential electricity demand. In particular, results show that when an automated decentralized (un-coordinated) dynamic energy management system is largely adopted by residential customers, the smoothing effect due to the natural stochastic features of residential demand is eliminated, forcing demand synchronization among all the consumers. Even though the demand is being effectively deferred toward night periods, results show that pronounced rebound peaks are created in the aggregate demand, that are higher and steeper than the original demand peaks that the time-varying electricity pricing structures were intended to eliminate. To cope with the rebound peaks created by simple timevarying electricity price structures, that lead to the synchronization of the individual residential demands, the use of Multi-TOU and Multi-CPP electricity pricing is proposed. In the Multi-TOU paradigm residential customers are divided into groups, and each group sees different TOU electricity prices. In this way each customer sees a time-varying electricity price, but not all the customers are synchronized. The adoption of Multi-time-varying electricity pricing does not lead to the creation of any rebound peak when automated energy management is introduced. Moreover, the natural “peakiness” of the residential demand appears to be significantly reduced. This achieves the objective for which residential demand response programs are designed, which is to alleviate the requirements on electric power generation infrastructure to follow the fluctuations in the demand. That is, instead of adapting electricity generation to match changes in the demand, the demand itself is made more flexible to reduce requirements on electric power generation. This also allows for an easier integration of non-dispatchable renewable resources. Simulation results show that the modeling proposed in this paper can serve as a tool to study energy policy solutions, including evaluating and comparing the effects of different electricity price structures, and developing effective residential demand response programs. R EFERENCES [1] M. Muratori, B.-A. Schuelke-Leech, and G. Rizzoni, “Role of Residential Demand Response in Modern Electricity Markets,” Renewable and Sustainable Energy Reviews, vol. 33, no. 0, pp. 546 – 553, 2014. [2] Smart Energy Demand Coalition, The Demand Response Snapshot - The Reality for Demand Response Providers Working in Europe today, 2011. [3] J. Morales, A. Conejo, H. Madsen, P. Pinson, and M. Zugno, Integrating Renewables in Electricity Markets: Operational Problems. Springer. International Series in Operations Research and Management Science, Vol. 205, 2014. [4] U.S. Department of Energy, Benefits of Demand Response in Electricity Markets and Recommendations for Achieving Them. A report to the United States Congress Pursuant to Section 1252 of the Energy Policy Act of 2005, February 2006.

[5] M. Muratori, “Dynamic management of integrated residential energy systems,” Ph.D. dissertation, The Ohio State University, 2014. [6] E. D¨utschke and A.-G. Paetz, “Dynamic electricity pricingwhich programs do consumers prefer?” Energy Policy, vol. 59, pp. 226–234, 2013. [7] J. A. Espey and M. Espey, “Turning on the lights: a meta-analysis of residential electricity demand elasticities,” Journal of Agricultural and Applied Economics, vol. 36, no. 1, pp. 65–82, 2004. [8] D. W. Caves, L. R. Christensen, and J. A. Herriges, “Consistency of residential customer response in time-of-use electricity pricing experiments,” Journal of Econometrics, vol. 26, no. 1-2, pp. 179 – 203, 1984. [9] J. Torriti, “Price-based demand side management: Assessing the impacts of time-of-use tariffs on residential electricity demand and peak shifting in northern italy,” Energy, vol. 44, no. 1, pp. 576–583, 2012. [10] R. Walawalkar, S. Fernands, N. Thakur, and K. R. Chevva, “Evolution and current status of demand response (DR) in electricity markets: insights from PJM and NYISO,” Energy, vol. 35, no. 4, pp. 1553–1560, 2010. [11] J. Torriti, M. G. Hassan, and M. Leach, “Demand response experience in europe: Policies, programmes and implementation,” Energy, vol. 35, no. 4, pp. 1575–1583, 2010. [12] C. W. Gellings and R. J. Lordan, “The power delivery system of the future,” The Electricity Journal, vol. 17, no. 1, pp. 70 – 80, 2004. [13] J. R. Roncero, “Integration is key to smart grid management,” in SmartGrids for Distribution, 2008. IET-CIRED. CIRED Seminar. IET, 2008, pp. 1–4. [14] G. R. Newsham and B. G. Bowker, “The effect of utility time-varying pricing and load control strategies on residential summer peak electricity use: a review,” Energy Policy, vol. 38, no. 7, pp. 3289–3296, 2010. [15] S. Borenstein, M. Jaske, and A. Ros, “Dynamic pricing, advanced metering, and demand response in electricity markets,” Journal of the American Chemical Society, vol. 128, no. 12, pp. 4136–4145, Mar. 2002. [16] H. Chao, “Price-responsive demand management for a smart grid world,” The Electricity Journal, vol. 23, no. 1, pp. 7–20, 2010. [17] F. A. Wolak, Residential customer response to real-time pricing: The anaheim critical peak pricing experiment, 2007, UC Berkeley: Center for the Study of Energy Markets. [Online], Available. [Online]. Available: http://escholarship.org/uc/item/3td3n1x1 [18] J. G. Kassakian and R. Schmalensee, The future of the electric grid: An interdisciplinary MIT study, 2011. [19] Dynamic pricing adoption will be very low without action from utilities. Navigant - Electric Light and Power, October 9, 2013, [Online], Available. [Online]. Available: http://www.elp.com/articles/2013/10/ dynamic-pricing-adoption-will-be-very-low-without-action-from-utilities. html [20] M. LeMay, R. Nelli, G. Gross, and C. A. Gunter, “An integrated architecture for demand response communications and control,” in Hawaii International Conference on System Sciences, Proceedings of the 41st Annual. IEEE, 2008, pp. 174–174. [21] G. Heffner and D. Kaufman, “Distribution substation load impacts of residential air conditioner load control,” Power Apparatus and Systems, IEEE Transactions on, no. 7, pp. 1602–1608, 1985. [22] S. Lehnhoff, O. Krause, and C. Rehtanz, “Dezentrales autonomes energiemanagement,” at-Automatisierungstechnik Methoden und Anwendungen der Steuerungs-, Regelungs-und Informationstechnik, vol. 59, no. 3, pp. 167–179, 2011. [23] A. Mishra, D. Irwin, P. Shenoy, and T. Zhu, “Scaling distributed energy storage for grid peak reduction,” in Proceedings of the 4th international conference on Future energy systems. ACM, 2013, pp. 3–14. [24] S. Schey, D. Scoffield, and J. Smart, “A first look at the impact of electric vehicle charging on the electric grid in the ev project,” in 26th Electric Vehicle Symposium (EVS-26), Los Angeles, 2012. [25] M. Muratori, C.-Y. Chang, G. Rizzoni, and W. Zhang, “Dynamic Energy Management of A Residential Energy Eco-System,” in Proceedings


[26] [27] [28] [29] [30]

[31] [32] [33] [34]

[35]

of the ASME 2013 Dynamic Systems & Control Conference (DSCC), October 21–23, 2013. M. Muratori, M. C. Roberts, R. Sioshansi, V. Marano, and G. Rizzoni, “A highly resolved modeling technique to simulate residential power demand,” Applied Energy, vol. 107, no. 0, pp. 465 – 473, 2013. M. Muratori, M. J. Moran, E. Serra, and G. Rizzoni, “Highly-Resolved Modeling of Personal Transportation Energy Consumption in the United States,” Energy, vol. 58, no. 0, pp. 168 – 177, 2013. U.S. Department Of Energy, Alternative Fuels Data Center, 2013, [Online], Available. [Online]. Available: http://www.afdc.energy.gov/ laws/law/NC/9355 D. Kirschen, “Demand-side view of electricity markets,” IEEE Transactions on Power Systems, vol. 18, no. 2, pp. 520–527, 2003. A.-H. Mohsenian-Rad and A. Leon-Garcia, “Optimal residential load control with price prediction in real-time electricity pricing environments,” Smart Grid, IEEE Transactions on, vol. 1, no. 2, pp. 120 –133, sept. 2010. K. C. Sou, J. Weimer, H. Sandberg, and K. H. Johansson, “Scheduling smart home appliances using mixed integer linear programming,” in IEEE Conference on Decision and Control, Orlando, FL, Dec. 2011. T. T. Kim and H. V. Poor, “Scheduling power consumption with price uncertainty,” IEEE Transactions on Smart Grid, vol. 2, no. 3, pp. 519– 527, Sept. 2011. P. Du and N. Lu, “Appliance commitment for household load scheduling,” Smart Grid, IEEE Transactions on, vol. 2, no. 2, pp. 411–419, June 2011. M. Avci, M. Erkoc, and S. S. Asfour, “Residential HVAC Load Control Strategy in Real-Time Electricity Pricing Environment,” in 2012 IEEE EnergyTech. Cleveland, OH, USA: Institute of Electrical and Electronics Engineers, 29-31 May 2012. D. Livengood and R. Larson, “The energy box: Locally automated optimal control of residential electricity usage,” Service Science, vol. 1, no. 1, pp. 1–16, 2009.

Matteo Muratori is a researcher at the Pacific Northwest National Laboratory - Joint Global Change Research Institute, where he works on evaluating and assessing the role new technologies in responding to global energy issues, including climate change and energy security. Prior to joining PNNL, Dr. Muratori worked for The Ohio State University – Center for Automotive Research where his research focused on on sustainable mobility, residential energy systems modeling, energy management techniques, smart grids and demand response programs, and energy policy. Dr. Muratori holds B.S. and M.S. summa cum laude degrees in Energy Engineering from Politecnico di Milano (Italy) and M.S. and Ph.D. degrees in Mechanical Engineering, along with a minor degree in statistics, from The Ohio State University.

Giorgio Rizzoni , the Ford Motor Company Chair in ElectroMechanical Systems, is a Professor of Mechanical and Electrical Engineering, and an Adjunct Professor of Industrial Design at The Ohio State University. He received his B.S. (ECE) in 1980, his M.S. (ECE) in 1982, his Ph.D. (ECE) in 1986, all from the University of Michigan (UM). Since 1999 he has been the director of The Ohio State University Center for Automotive Research (CAR), an interdisciplinary university research center. CAR conducts research on advanced automotive and transportation technologies and systems engineering, focusing on sustainable mobility, advanced propulsion systems, human safety and the environment.

10