The Economic Value of Distributed Storage at Different Locations on an Electric Grid.

• Author: Jeon, Wooyoung

1. INTRODUCTION

   Although increasing the dependence on Renewable Energy Sources (RES) for generating electricity is expected to lower operating costs, the expected savings may not be fully realized due to the operational challenges caused by the inherent variability of RES. One effective way to mitigate this uncertainty is to use storage capacity as a buffer. For example, one could collocate batteries at wind farms to provide a source of generation that is smoother and more predictable. However, having batteries dedicated solely to this purpose is relatively expensive. In contrast, distributed storage located at load centers can mitigate the variability of RES and also modify load profiles, and in particular, reduce the peak system load and congestion on the grid. Consequently, distributed storage can lower the amount of Conventional Generating Capacity (CGC) committed for reserves and ramping and also the amount of installed CGC needed to ensure generation adequacy for the grid. The basic economic questions addressed in this paper are (1) how is the capacity of distributed storage allocated between providing ramping services and shifting load from peak to off-peak periods, and (2) what are the implications of this allocation on total system costs, including the capital cost of installed CGC.

   Ramping services are needed because of the natural variability of RES and the difficulty of forecasting generation from RES accurately. This uncertainty increases operating costs and steps have been taken in some regions to internalize the associated ramping costs (e.g., Wang and Hobbs, 2016), but even if this is done, congestion on the grid can still prevent CGC and RES at remote locations from participating effectively in the market. When suitable incentive mechanisms are provided for installing storage capacity, the storage can improve the utilization of all generation assets on the grid. The economic and environmental value of using storage in the wholesale market is clearly recognized in the literature (e.g., Sioshansi and Denholm, 2010; Denholm et al., 2013; RTO Insider, 2016), and recently, attention has also been directed to using storage for regulation due to its rapid response capabilities (Parkinson, 2017; Washington Post, 2017). In general, the actual way storage capacity is used is affected by local conditions on the grid, and this determines the system benefits derived from the storage at a specific location. The overall implication is that the benefits derived from storage at different locations vary because the storage is used differently at different locations. Understanding the causes of these differences is a central theme for this paper.

   We have argued in other papers (Jeon et al., 2015) that deferrable demand' is an attractive form of distributed storage because it is typically less expensive than batteries. Examples of deferrable demand include (1) Thermal Storage (TS) for space heating/cooling, (2) electric water heaters, and (3) smart-charging of the batteries (2) in Electric Vehicles (EVs). EVs can improve the management of the grid by charging during low demand periods with excess generation from RES, reducing the spillage of this resource, and, if vehicle-to-grid capabilities are supported, discharging when generation from RES is less than expected. TS works in a similar way to EVs but its potential effect is much larger (e.g., Xu et al., 2013). The stored TS energy can be used to reduce the system peak by, for example, displacing the compressors used for air-conditioning when the load on the grid is high. Furthermore, the potential for TS is large because the amount of energy used for heating and cooling accounts for roughly 40 percent of total energy consumption in the residential and commercial sectors (39 quadrillion British Thermal Units in 2016 for the U.S., (EIA, 2015, 2017). There are, in fact, numerous recent contributions to the literature that recognize this potential and focus on the control of thermostatic load and the possibility of coupling it with RES (e.g., Callaway and Hiskens, 2011; Pourmousavi et al., 2014).

   Past research has shed light on the efficiency issues raised by, for example, the widespread use of flat-rate structures for retail rates and the perverse incentives that they provide for investing in deferrable demand (Simshauser and Downer, 2016). In fact, experience with regulations that increase the responsiveness of demand by using real-time prices shows that this is the most promising way to develop two-sided markets (e.g., Bushnell et al., 2009; Chen and Kleit, 2016), and there is a large literature looking at the welfare gains from real-time prices (e.g., Borenstein, 2005; Faruqui et al, 2014) and also from block pricing (e.g., You and Lim, 2017). Overall, tariffs that are reflective of the true system costs lead to gains in welfare, and in this article, our focus is on how to reduce the total system costs. Note that information available in the prices can help improve the forecasting of other variables of interest, notably the load (Forbes and Zampelli, 2014). The effects on lowering retail rate structures will be addressed in future work. One reason for this is that we conclude our article by questioning whether minimizing the expected operating costs during peak periods, and ignoring capital costs, leads to long-run efficiency in the use of storage capacity.

   The specific objective of this work and our main contribution is to determine the value of deferrable demand at different locations. The analysis uses a stochastic form of multi-period Security Constrained Optimal Power Flow (SCOPF) that maximizes the expected social surplus for operations over a 24-hour period. This model assumes that a System Operator (SO) determines how all generating capacity is committed and the charging/discharging profiles of deferrable demand (EVs and TS). We simulate the daily operations of the grid using a reduced network representing New York State and New England with different assumptions about the amount of deferrable demand that is available. This type of application leads naturally to differences in the shadow prices at different locations. Since each shadow price represents the expected marginal cost of delivering energy at a specific location, it determines, to a large extent, the optimum use of any deferrable demand at that location. Our main empirical results show that there are major differences in the use of deferrable demand at two large load centers, representing Boston and New York City. We focus on wind energy and the coupling effects it has with deferrable demand, due to the diurnal pattern of this resource and the characteristics of the location for our empirical application, the Northeastern U.S. Our analytical approach and conclusions can inform plans for other regions with similar prospective levels of wind generation to our empirical area of study. These results have economic policy implications for the design of electricity tariffs and for improving the net social benefits of the grid (e.g., Woo and Seeto, 1988; Hobbs, 1991; Borenstein, 2007). However, the main practical conclusion from the analysis is to demonstrate that distributed storage (i.e., deferrable demand at load centers) can be an effective way to provide ramping services to the grid to mitigate the inherent uncertainty of the generation from wind farms on the grid. It is not necessary to colocate storage at the wind farms.

   The remainder of the article is organized as follows. Section 2 provides a description of the analytical framework and explains the spatial characteristics of inputs for the model. Section 3 presents the information exchanges that define the market and operations on the grid, and the physical specifications of deferrable demand for TS and EVs. Section 4 presents the empirical results for five different cases for an application in which a SO optimizes operations on the grid and the management of deferrable demand. Section 5 concludes and provides suggestions for further research.

2. FORMULATION AND DATA

   This work belongs to the literature on integrating RES into grid operations, focusing on the problem from the viewpoint of the SO (e.g., Hirth, 2015). Most of this recent literature uses either stochastic programming (e.g., Arroyo and Galiana, 2005) or robust optimization (e.g., Bertsimas et al., 2013) for the simulation analysis. The high computational requirements for these methods have prompted a discussion of novel solutions approaches (e.g., Siano et al., 2012) or approximations to the distribution of the uncertain parameters (e.g., Warrington et al., 2012). This article uses a unique, probabilistic, multi-period Security Constrained Optimal Power Flow (pS-SC-OPF). The SO's objective is to maximize the total expected welfare of the participants in the system associated to the provision of energy subject to a set of economic criteria focused on reliability, and the technical constraints of the system. A particularly relevant contribution of this model is the determination of shadow prices of electricity that internalize the variability and uncertainty from RES (Lamadrid et al., 2015). The main differences between our model and other stochastic formulations used to study similar problems can be summarized in four areas:

1. The distinction between different classes of uncertainty in the system. The first class is related to events with a low probability of occurrence (e.g., an outage due to a "tree strike" on a transmission line, or contingencies). The second class is related to events that have high probability of occurrence (e.g., generation from RES drastically changing due to clouds passing in the case of solar energy, or states). The uncertainty from events in both classes affects system operations and the overall expected cost. In this case, an information layer (e.g., a Supervisory Control and Data Acquisition System, SCADA) can provide the SO with the...