Characterizing the Response of Commercial and Industrial Facilities to Dynamic Pricing Signals from the Utility

We describe a method to generate statistical models of electricity demand from Commercial and Industrial (C&I) facilities including their response to dynamic pricing signals. Models are built with historical electricity demand data. A facility model is the sum of a baseline demand model and a residual demand model; the latter quantifies deviations from the baseline model due to dynamic pricing signals from the utility. Three regressionbased baseline computation methods were developed and analyzed. All methods performed similarly. To understand the diversity of facility responses to dynamic pricing signals, we have characterized the response of 44 C&I facilities participating in a Demand Response (DR) program using dynamic pricing in California (Pacific Gas & Electric’s Critical Peak Pricing Program). In most cases, facilities shed load during DR events but there is significant heterogeneity in facility responses. Modeling facility response to dynamic price signals is beneficial to the Independent System Operator for scheduling supply to meet demand, to the utility for improving dynamic pricing programs, and to the customer for minimizing energy costs. INTRODUCTION Electricity markets are unusual in that electricity demand is relatively inelastic, in part because most electricity consumers do not monitor electricity prices. In addition, when generation nears its capacity supply also becomes increasingly inelastic. Traditionally, Independent Systems Operators (ISOs) balance electricity supply and demand by calling on the supply-side to produce electricity to meet demand. The result is extreme electricity price volatility on days when the load is high [1]. Several studies have attempted to understand and quantify the benefits of a more responsive demand-side [2-4]. One way of achieving Demand Response (DR) is through dynamic electricity pricing, defined here as any electricity pricing scheme in which prices are known to the customer no more than a day in advance. Per this definition, examples of dynamic pricing would include Critical Peak Pricing (CPP) and Real Time Pricing (RTP). Time of Use (TOU) pricing is not dynamic pricing under our definition. Commercial and Industrial (C&I) facilities using Energy Management Control Systems (EMCS) are likely candidates for dynamic pricing tariffs because responses to varying electricity prices can be automated. Responding to dynamic pricing includes both shedding and shifting electricity demand. Examples of load shedding strategies include changing Heating, Ventilating, and Air Conditioning (HVAC) set points and dimming/switching-off noncritical lighting. Examples of load shifting strategies include pre-cooling and water heating during times of the day when energy costs are low. From the perspective of the ISO and the utility there is a strong need to accurately predict characteristics of C&I facilities’ demand sheds and shifts (including their variability) resulting from dynamic pricing signals. Better shed/shift forecasts are important to the ISO for scheduling electricity supply to meet demand. Predictions also allow the utility to understand how characteristics of dynamic pricing signal responses vary across facilities and which facilities are best suited to different DR programs. In addition, understanding facilities’ responses to dynamic prices allows the utility to model, analyze, and compare new dynamic pricing programs and their ability to achieve the utility’s goals (e.g., peak demand reductions, highly-predictable demand reductions, energy savings, etc.). It is also advantageous for C&I facilities to be able to predict characteristics of their own demand sheds/shifts. Facility managers would like to minimize energy costs subject to available resources, constraints (e.g., occupant comfort), and dynamic pricing information. Understanding how a facility responds to pre-programmed dynamic pricing response strategies is essential for minimizing energy costs. For instance, data presented in this paper show that the time interval between receipt of a demand response signal and complete execution of a demand reduction (referred to as ‘ramp time’ in this paper), using an automated demand response system, varies from less than three minutes to more than two hours. Facility managers should consider the ramp time of their facility when devising optimal price response strategies. Another important reason to understand facility sheds/shifts resulting from dynamic pricing signals is to determine the optimal lead-time (defined here as the time between when prices are published and when they take effect) at which dynamic prices should be announced to the customer. It is advantageous to the facility to receive pricing information with large lead-times so that the facility can plan its response, leveraging all of its DR resources, to minimize total energy costs. However, to participate in real-time electricity markets, facilities would receive pricing information at short lead times (e.g., one hour), which would not allow them to leverage all of their DR resources (e.g., pre-cooling) resulting in smaller demand sheds. There are clearly tradeoffs in facility response to dynamic pricing as pricing information lead-time changes. Though this is not the subject of this paper, we are currently researching these tradeoffs. One way to predict demand sheds and shifts, and their variability, is by modeling C&I facilities. Facility models can be built with physical equations or historical electricity demand data. Existing physical equation-based facility models can accurately predict certain facility loads (e.g., HVAC loads, some lighting loads) if the characteristics of the facility are known and the model is calibrated with actual demand data. However, physical equation-based models seldom accurately predict human-controlled loads (e.g., some lighting loads, humancontrolled process loads, plug loads, etc.) and their variability. Alternatively, statistical models (which we employ in this paper) built with historical electricity demand data capture all facility loads and do not require calibration or knowledge of facility characteristics. However, historical electricity demand data are not generally sub-metered so different types of loads cannot be disaggregated. Also, statistical models capture only historical facility behavior and cannot be used to predict future facility behavior (though statistical models can be updated as more data become available). In this paper, we describe a method to generate statistical models of C&I facilities including their response to dynamic pricing signals. Facility models comprise a baseline demand model and a residual demand model that predicts deviations from the baseline model due to dynamic pricing signals. To understand the diversity of facility responses to dynamic pricing signals, we have characterized the response of 44 C&I facilities participating in Pacific Gas and Electric’s (PG&E’s) CPP Program. We begin with a brief review of relevant research on DR. We then detail our approach and describe the data used in our analysis. We present our preliminary results and conclude. DEMAND RESPONSE DR is defined by DOE as “a tariff or program established to motivate changes in electric use by end-use customers in response to changes in the price of electricity over time, or to give incentive payments designed to induce lower electricity use at times of high market prices or when grid reliability is jeopardized” [3]. DR has many benefits. It can be used to reduce price volatility during peak periods [4]. It also contributes to system reliability [5]. In addition, DR could be used to provide ancillary services (A/S) such as spinning reserve [6] and regulation/load following [7].