An inference-based tabular approximate dynamic programming for online energy management of home energy management system with energy storage

With the development of smart home technologies, home energy management systems (HEMS) with energy storage are expected to participate in online energy management (OEM) to reduce costs under solar and load forecast uncertainties. This paper proposes an inference-based tabular approximate dynamic programming (ITADP) algorithm for OEM of HEMS. The conventional tabular approximate dynamic programming (TADP) algorithm directly utilizes learned value estimates for decision-making without considering their estimation accuracy, which may lead to degraded decision quality under time-varying stochastic conditions. To address this issue, the proposed ITADP algorithm employs a Bayesian inference mechanism to adaptively refine inaccurate value estimates, thereby improving decision quality. In addition, different from the traditional TADP algorithm that needs to repeatedly call the solver to compute the near-optimal decision, the proposed algorithm can directly obtain it through the proposed analytical expression method, thereby achieving higher computational efficiency. Numerical simulations on the OEM in HEMS demonstrate that the proposed algorithm reduces cumulative cost by 16.4%, 11.2%, 6.0%, and 5.3% compared to Myopia, MPC, TADP, and DDQN algorithms, respectively.