Relieving the Need for Bi-Level Decision-Making for Optimal Retail Pricing via Online Meta-Prediction of Data-Driven Demand Response of HVAC Systems

Price-based demand response (DR) of heating, ventilating, and air-conditioning (HVAC) systems is a challenging task, requiring comprehensive models to represent the building thermal dynamics and game theoretic interactions among participants. This paper proposes an online learning-based strategy for a distribution system operator (DSO) to determine optimal electricity prices, considering the optimal DR of HVAC systems in commercial buildings. An artificial neural network (ANN) is trained with building energy data and represented using an explicit set of linear and nonlinear equations, without physics-based model parameters. An optimization problem for price-based DR is then formulated using this equation set and repeatedly solved offline, producing data on optimal DR schedules for various conditions of electricity prices and building thermal environments. Another ANN is then trained online to directly predict DR schedules for day-ahead electricity prices, which is referred to as meta-prediction (MP). By replacing the DR optimization problem with the MP-enabled ANN, an optimal electricity pricing strategy can be implemented using a single-level decision-making structure, which is simpler and more practical than a bi-level one. In simulation case studies, the proposed single-level strategy is verified to successfully reflect the game theoretic relations between the DSO and commercial building operators, so that they effectively exploit the operational flexibility of the HVAC systems to make the DR application profitable, while ensuring the grid voltage stability and occupants thermal comfort.