Supervised Optimization Framework for Charging and Discharging Controls of Battery Energy Storage

Although residential houses have widely adopted battery energy storage (BES) in conjunction with solar photovoltaic (PV) panels, it has been challenging to optimize BES controls owing to the uncertainty inherent in electricity prices, renewable energy resources, and electricity demand loads. The main objective of this study is to propose a supervised optimization (SO) framework designed to enable supervised learning to infer efficient BES operations in conjunction with mathematical optimization. Based on the proposed SO framework, a mathematical optimization model is formulated and solved to generate optimal charging and discharging controls given historical data in an offline optimization fashion. Then, a supervised learning model can be trained based on the optimal charging and discharging controls to infer efficient BES charging and discharging controls in an online optimization fashion. Specifically, an objective function and constraints driven by the mathematical optimization model are reflected in the course of training and inferring by supervised learning models so that the SO framework can infer feasible and efficient BES controls. Numerical experiments are conducted with data collected from a residential community in Texas, and the results reveal that the proposed SO framework results in better performance compared to the benchmark model, such as reinforcement learning and model predictive control.