Real-time optimal control of integrated power systems via deep neuro-dynamic programming
Bei Ye, Xin Zhang, Yan Gao, Hossein Salehfar, Nan Wu, Yong Hou · Energy Sources Part B Economics Planning and Policy · 2025
This paper introduces a real-time optimal control strategy for an integrated power system focused on demand-side management and involving two distinct energy sources. The proposed approach features a dual-component model: one on the user side, focused on welfare maximization and operating independently of time, and another on the provider side, which dynamically adapts based on real-time user data. The model’s complexity, particularly due to its interactive nature between user demands and energy supply, poses a significant challenge in terms of conventional solution methods. In response, we introduce a bilateral interactive algorithm that incorporates deep learning techniques, significantly enhancing the model’s capability to manage user welfare and provider responses dynamically. This approach includes a deep learning-enhanced neuro-dynamic programming architecture, featuring a critic network and multiple action networks, which together address the complex interactions between user demands and energy supply. The simulation results demonstrate the approach’s rapid convergence and its feasibility in optimizing control strategies, significantly enhancing the decision-making efficiency in power systems.