Data-driven real-time home energy management system based on adaptive dynamic programming

Abstract

Real-time optimal control is crucial for the efficacy of Home Energy Management Systems (HEMS) in residential settings during actual operation. The time-varying and nonlinear nature of smart households — characterized by fluctuations in renewable energy generation, real-time electricity pricing, and load consumption — presents substantial challenges for both prediction and real-time control within HEMS. To tackle these issues, this paper introduces a real-time optimal control algorithm, augmented by predictive scheduling for HEMS. More specifically, the proposed real-time HEMS framework integrates an adaptive dynamic programming (ADP) algorithm, which is complemented by predictions of renewable energy generation and load consumption. Initially, data-driven methodologies generate accurate forecasts using available data collected and processed in real time. Gated Recurrent Unit (GRU) neural networks utilizing a range of data inputs such as electricity prices, battery charge/discharge rates, load consumption, and renewable energy generation, the system computes the optimal performance index function. Following this, we employ the ADP algorithm to reduce total electricity costs. This paper confirms the convergence properties of the value iteration ADP algorithm., demonstrating a monotonic approach of the iterative performance index function towards the optimal solution. The efficacy of the proposed algorithm is supported by numerical experiments, which verify its that solar energy efficiency has increased to 98% and electricity costs have been reduced by 64%.