Estimation of Regulation Reserve Requirements in California ISO: A Data-driven Method

This paper proposes a data-driven method to estimate real-time regulation reserve requirements in the California Independent System Operator (CAISO) balancing authority area (BAA). The approach is based on the statistical analysis of actual historical area control error (ACE) and regulation procurement in the CAISO system. The CAISO baseline method uses day-ahead (DA) information and does not statistically relate requirements in real time. In this work, we examine the benefit of setting requirements in real time, and the proposed method presents advantages rather than the present DA approach by using additional 1-hour-ahead (1HA) independent variables to estimate requirements. By adopting a time series forecasting technique, the approach estimates regulation reserve requirements on an hourly basis. More specifically, a neural network based upon a nonlinear autoregressive exogenous model (NARX), which accounts for temporal continuity and autocorrelation, is leveraged for regulation requirements estimation. The proposed approach incorporates multiple time-series variables that might affect the system balance (e.g., solar, wind, and load). Two metrics are adopted to evaluate method performance: 1) frequency of shortage, and 2) oversupply of the regulation reserve requirement. By dynamically tuning the parameters of the procurement strategies, system regulation reserve requirements were reduced without compromising system reliability, improving both system reliability and economics.