Bidding on Tomorrow: Dispatch Algorithms for Energy Storage Resources

Abstract

Rapidly increasing electricity demand and growing penetration of variable renewable energy have significantly increased volatility in power markets. These conditions make the control of energy storage resources (ESRs) increasingly important: when dispatched well, ESRs can absorb low-cost renewable energy during periods of excess supply and discharge during high-price, carbon-intensive periods, improving asset revenues and grid flexibility while reducing emissions. This project develops improved dispatch algorithms for ESR participation in the Pennsylvania-New Jersey-Maryland (PJM) Interconnection. It first evaluates existing dispatch strategies and formulates linear programming methods for co-optimizing participation across energy, regulation, and capacity markets. It then collects and processes market and weather datasets to train electricity price forecasting models for real-time control, using a unified testing pipeline that evaluates more than 230 models across 20 statistical, machine learning, and deep learning architectures. The project also introduces multi-model ensemble forecasting paired with stochastic optimization to better represent price uncertainty and improve dispatch performance. In sum, this work produces a high-performance dispatch algorithm that is estimated to outperform leading industry software by over 20%, and contributes novel methods for market co-optimization, forecast evaluation, and ensemble-based control of storage assets.