Elsevier

Applied Energy

Volume 210, 15 January 2018, Pages 987-1001
Applied Energy

Day ahead optimization of an electric vehicle fleet providing ancillary services in the Los Angeles Air Force Base vehicle-to-grid demonstration

https://doi.org/10.1016/j.apenergy.2017.07.069Get rights and content

Highlights

  • We present an overview of an ongoing EV vehicle-to-grid demonstration project.

  • We discuss practical issues of bidding EV capacity in frequency regulation markets.

  • We formulate a MILP optimization model to plan EV charging and day-ahead bidding.

  • We analyze the model sensitivity and bidding strategy to variation of key inputs.

  • Model behavior is highly sensitive to predicted resource utilization and prices.

Abstract

The Los Angeles Air Force Base Electric Vehicle Demonstration is a currently ongoing vehicle-to-grid demonstration project with the objective of minimizing the cost of operation of a fleet of approximately 30 electric vehicles (EVs) through participation in the California Independent System Operator (CAISO) frequency regulation market. To accomplish this, a hierarchical control system has been developed to optimize, plan, and control the charging, market bidding, and response to grid system operator control of the EVs. This paper presents an overview of the day-ahead optimization model component of the hierarchy. The model is a mixed integer linear program that optimizes daily EV charging and regulation capacity bids strategies in order to minimize operation costs and maximize ancillary service revenue. A deterministic approach is used due to several practical concerns of the demonstration project, including model complexity and the availability and uncertainty of input data in day-ahead decision making, and the limited size of the fleet. The model includes additional user-defined parameters to tune model behavior to better match real-world conditions and minimize the risks of uncertainty.

The paper conducts scenario analysis to explore the impact of these parameters on high level model behavior and resulting bid strategy. The parameters explored include hourly regulation prices, local load conditions leading to retail demand charges, forced symmetry constraints for regulation bids, SOC penalty values to reserve higher states-of-charge in vehicles, and expected regulation resource utilization while providing reserves. These analyses show significant sensitivity in the frequency regulation bidding strategy to the regulation utilization, as well as large differences in the regulation prices between regulation up (discharging capacity) and regulation down (charging capacity). Results also suggest enforcing symmetry in regulation appears to have significant impacts in regulation revenue when there is large relative disparities between prices in the up and down direction. Finally, imposing a small cost on low SOC values significantly impacts the fleet-wide average SOC, making the system more resilient to uncertainty in the mobility demands gathered at the time of making day ahead decisions.

Introduction

Electric vehicles (EVs) have been touted as a panacea for our carbon-hungry, energy importing transportation sector. Their ability to shift the energy production burden away from distributed, inefficient internal combustion engines to the electricity sector supports national priorities in energy security and public health, and opens up opportunities for decarbonization of personal mobility [1], [2]. Further, their commercial renaissance corresponds with significant introduction of intermittent, renewable energy resources into the electricity grid. As renewable generation becomes more prominent, some electricity system decision-makers are looking to increase storage capacity [3], and EVs appear to be a promising, low-cost energy storage resource for the grid. However, the interaction of individual EVs with electricity grid and market operators can be far too onerous from such a small resource to warrant electric vehicle participation directly. This creates a niche for an entity that aggregates a population of electric vehicles to present them to the market operator in a size that is useful for grid operations.

The EV aggregator will play a number of important roles in vehicle-to-grid (V2G) services offered into markets. They will need to understand the availability of the EVs that they represent, take positions and assume the financial risks associated with providing the services in a market, manage their resources in a way to meet any capacity and energy obligations made (e.g. ensure that there is adequate energy stored prior to a service provision period), and finally to determine which vehicles will provide the requested grid service in real-time. All of these must be accomplished while ensuring that the mobility needs of vehicle owners are met, and the cost of EV ownership is reduced. For consumers and fleet operators, the deployment of EVs also creates opportunities for operational cost reduction (e.g. from demand side management) and revenue from new and existing markets, by employing novel planning and control strategies to leverage idle EV capacity.

In the present paper, we focus on a real-world demonstration of one such aggregation at the Los Angeles Air Force Base Electric Vehicle Demonstration (LAAFB EVD). The LAAFB EVD integrates a mixed fleet of roughly 30 electric vehicles capable of bi-directional charging into the wholesale frequency regulation market run by the California Independent System Operator (CAISO) to minimize the net cost of operating the fleet [4]. The demonstration is the first of its kind to take an operational vehicle fleet, replace it with electric vehicles, and participate as a full market resource (subject to all rules and financial obligations) in a frequency regulation market in the US. The hierarchical control system that enables many EV aggregator functions in the LAAFB EVD project is composed of a fleet scheduling tool to gather input data, day-ahead and hour-ahead charging and market participation optimization models based on LBNL’s Distributed Energy Resources – Customer Adoption Model (DERCAM) [5], [6], and a real-time myopic optimal controller for charging instructions described in [7]. Of particular interest to this discussion is the formulation and design choices made in the development of the day-ahead market participation optimization model.

The costs of owning and operating an electric vehicle can be reduced through offering vehicle-to-grid services.1 Service opportunities in current markets are found either in the management of retail electricity purchases, or in wholesale electricity market participation. Retail bill management falls into two major categories: (1) taking advantage of time-varying electricity tariffs by charging/discharging to minimize retail electricity costs; and (2) managing peak electricity demand charges, set by the highest consuming 15-min interval in a month, which can account for nearly 50% of retail electricity bills for commercial account customers. While savings on retail electricity can be a significant opportunity for vehicle-to-grid capable EVs, wholesale market revenue opportunities are more varied.

Grid services in wholesale markets include offering planning capacity, energy, and operational reserves. Planning capacity is an offer to participate in the wholesale energy market during a period of performance in the planning horizon (months to years). In contrast, wholesale energy and operational reserves are offered day-ahead or hour-ahead. Energy offers are simply for buying or selling a quantity of electricity, and the market matches buyers and sellers to determine the price and quantity of electricity each is awarded. Energy markets offer the opportunity for EV owners to charge when the cost of electricity is lowest and even arbitrage energy purchases between high and low priced periods. Operational reserves are used to maintain balance between supply and demand in the event of unexpected changes in either. Operational reserve based ancillary services, such as synchronous reserve and frequency regulation reserve, are of particular interest to EVs with bi-directional capability, as they typically require low energy commitments with reasonable payment for being held in reserve [8].

In the LAAFB EVD, the EV fleet aggregation is allowed to participate in the wholesale market for ancillary services, but all energy consumed by the vehicles is settled at a Southern California Edison retail rate, per guidance by the California Public Utilities Commission. The LAAFB EVD targeted provision of the most valuable ancillary service in CAISO: frequency regulation reserve [8]. Frequency regulation is a centrally controlled service that continuously attempts to balance supply and demand between the five minute ISO economic dispatch to minimize both deviations in system frequency from nominal and unscheduled electricity interchanges with neighboring balancing authorities. The control system, known as Automatic Generation Control (AGC), sends real power setpoint commands from the system operator to resources providing the reserve every 2–10 s. The commands will be within the market-awarded reserve capacity range, in units of megawatts held in reserve for one hour or MW-h. In most markets in the US, reserve capacity offers are symmetrical, meaning that the resource that offers any amount of capacity is offering to either decrease or increase their output up to the offer amount around their energy market award. In CAISO, AGC offers are separate for up (generation) and down (load) capacity and setpoints are delivered on a 4 s interval.

To capture value in the electricity market, an EV aggregator must decide the size of their service offer within the constraints of market timelines. In many ancillary service markets in the United States, this offer must be bid at least one day before providing the service [8]. The challenge is determining an economically optimal schedule of charging and grid service offers in the face of significant day-ahead uncertainty. Uncertain quantities include the energy demands on the battery (both from providing grid services and for mobility), the time in which a vehicle is plugged into its EV supply equipment (EVSE), and the prices that will be received for providing services. To manage these uncertain decisions, the aggregator can leverage optimization techniques that either explicitly include the stochasticity [9], [10], [11], [12] of the resource or attempt to robustly manage the representations of the risk in a deterministic optimization framework [13], [14]. The latter was selected for the LAAFB EVD to limit complexity in an ambitious demonstration project.

The opportunity for electric vehicles to participate in providing grid services has been well examined from a technical potential, optimization and control perspective. Kempton and Tomić performed seminal work describing the opportunity for vehicle grid interactions [15] and much work, such as that performed by Galus et al., attempts to further refine the role that electric vehicles will have in grid operations [2].

Many attempts to develop optimal grid participation strategies for electric vehicles have been proposed in the literature. In some, unidirectional modulated charging grid interactions are considered in an optimal control framework [16], while others have used electric vehicles as a proxy for a more general distributed resource in a model predictive control type approach to providing grid services [17], and still others approach the problem with generalized bi-directional or unidirectional grid services that EVs can provide [18], [19], [20]. All of these employ some simplification to the modeling of either the electric vehicles (such as uni-directional power flow) or the gride services offered (generalizing grid services without a market context), however some do explicitly consider the service of interest in the present paper, frequency regulation [9], [11], [14], [21]. Some models explicitly consider the point of view of the charging services aggregator [11], [13], [18], [21] or market participation, constraints and bidding timelines [13], [22], [23], however these models focus on available price data, often from the PJM market, and are never interested in participation rules of any specific market context, which can vary considerably [8].

Most work in optimization of V2G services has been deterministic, relying on average input values from large aggregations of thousands of vehicles to provide services [21] or perfect forecasts [13] to determine optimal charging and reserve offers. More recent work has focused on methods to directly handle the uncertain input parameters for vehicle-to-grid optimization through the use of stochastic dynamic programming [9], fuzzy optimization [11], robust [14] and stochastic optimization [10], [12]. However, most of these approaches require significant historical data to create models of the uncertainty in their inputs. This requirement may be unreasonable for small aggregations with limited historical data.

The present paper describes the day ahead optimization model being applied to the LAAFB EVD and examines the optimization’s sensitivity to parameters designed to manage some of the uncertainty inherent in the input data. A deterministic approach to optimization of vehicle charging and frequency regulation provision is taken due to the lack of historical data for the frequency regulation signal and electric vehicle usage, as well as to reduce complexity and required run-time in a live demonstration. Parameters that penalize low battery states of charge (SOC) and estimate energy content of the frequency regulation signal to handle these optimization uncertainties. Scenario analysis for frequency regulation prices, non-EV site electricity load, and EV travel requirements are evaluated to see the sensitivity to inputs. Finally, the imposition of symmetric frequency reserve provision is examined for comparison to other market contexts. Results help evaluate the impact of these parameters and inputs on expected economic benefits from V2G activities in the context of the real-world demonstration in the LAAFB EVD project. The paper offers an approach to optimization of small aggregations of electric vehicles fleets, as will likely be seen in near to mid-term applications, and suggests parameters that can help account for uncertain inputs into an aggregator’s optimal scheduling algorithm for V2G offers.

The remaining sections of the paper are structured as follows: Section 2 describes in detail the optimization formulation developed for the LAAFB EVD. Section 3 introduces the design of scenarios that were analyzed. Section 4 presents results and discussion of the scenario results. Lastly, Section 5 concludes the paper, highlighting important findings and suggesting opportunities for future work.

Section snippets

Day ahead optimization formulation

The optimization formulation minimizes the cost of operation for an EV fleet, subject to constraints that account for the dynamics of energy storage in the vehicles, physical infrastructure constraints, and market participation constraints. The decision variables are the charge/discharge power of each vehicle for each interval in the optimization horizon, the regulation up and down capacity that the fleet may provide for each hour, as well as how the expected impact of regulation is distributed

Scenario design

Regulation bidding strategies and charging schedules generated by the model are dependent on the availability and energy requirements of the vehicles. They are also driven to a lesser extent by other inputs, such as the forecasted hourly regulation prices, AGC utilization, and building loads, as well as other, user defined model parameters such as the SOC penalty. The scenario analyses presented in this paper are used to evaluate the sensitivity of model outputs to several key input parameters.

Base scenario results

The base case scenario run set is composed of 150 optimization results corresponding to the 30 EV fleet schedules and five regulation price scenarios described above. Fig. 4 illustrates the impact of regulation price on hourly bidding behavior of the base case parameters by plotting the range of values in each hour for up regulation, down regulation, and energy bidding. In each of these plots, the solid line indicates the median hourly values for the runs within that subset. The colored shaded

Conclusion

This paper presents an overview and mathematical formulation of the day-ahead optimization model being deployed at the Los Angeles Air Force Base Electric Vehicle Demonstration. This optimization identifies economically optimal charging patterns and frequency regulation bidding strategies for the approximately 30 bi-directional charging capable EVs in the LAAFB EVD. It requires input data that include forecasts of CAISO market prices, building load, and intended EV fleet trip requirements

Acknowledgements

This work was funded jointly by the US Department of Defense’s Environmental Security Technology Certification Program (ESTCP) and California Energy Commission (CEC). Lawrence Berkeley National Laboratory is operated for U.S. Department of Energy under Contract Grant No. DE-AC02-05CH11231.

The authors would like to thank their project partners at Kisensum, especially Paul Lipkin and Bob Barcklay, as well as former LBNL staff, Chris Marnay and Sila Kiliccote, for their work during the early

References (24)

  • MacDonald J, Cappers P, Callaway D, Kiliccote S. Demand response providing ancillary services: a comparison of...
  • J. Donadee et al.

    Stochastic optimization of grid to vehicle frequency regulation capacity Bids

    IEEE Trans Smart Grid

    (2014)
  • Cited by (98)

    View all citing articles on Scopus
    View full text