Setpoint Tracking with Partially Observed Loads

Demand response (DR) is an important source of flexibility for the electric power grid [1], [2], [3]. In this work, we use online convex optimization (OCO) to design algorithms for tracking setpoints with uncertain flexible loads in DR programs. Uncertainty is a key challenge in DR [4]. This uncertainty arises from weather, human behavior and unknown load models. In addition, loads are time-varying, e.g., a heater may not have much flexibility during cold nights. The uncertain, time-varying nature of loads means that load aggregators must deploy loads for DR to observe their capabilities.