Article: Causal reinforcement learning-based operational performance evaluation of multi-source energy storage power stations Journal: International Journal of Energy Technology and Policy (IJETP) 2025 Vol.20 No.7 pp.109 - 126 Abstract: To address the challenges of multi-source heterogeneous data coupling, multi-scenario strategy dynamic adaptation, and small-sample modelling in the operational performance evaluation of multi-source energy storage power stations, this paper proposes a hierarchical physical-data fusion modelling and causal reinforcement learning collaborative evaluation method. Firstly, a mechanism-data hybrid model for underlying multi-type energy storage units is constructed, embedding physical constraints of energy storage units such as flywheel rotor dynamics equations and flow battery Nernst correction equations. Secondly, a scenario-driven multi-time-scale dynamic aggregation model is proposed, establishing an equivalent circuit network for multi-source energy storage power stations based on Kron reduction theory, and analysing the nonlinear coupling mechanism of power distribution in peak-shaving (minute-level) and frequency regulation (second-level) scenarios. Thirdly, a causal reinforcement learning framework is designed, integrating the structural causal model and deep deterministic policy gradient algorithm to achieve closed-loop optimisation of cross-scenario control strategies and power station operational performance indicators. The verification based on a provincial 'pumped storage-flywheel-lithium battery-flow battery' multi-source energy storage demonstration project shows that the method applies to operational performance evaluation in peak shaving, frequency regulation, and other scenarios. Inderscience Publishers - linking academia, business and industry through research

Title: Causal reinforcement learning-based operational performance evaluation of multi-source energy storage power stations

Authors: Junfang Sun; Weiqing Wang; Jin Jin; Xiaokan Gou; Junjun Ma

Addresses: State Grid Qinghai Information and Telecommunication Company, Xining 810008, Qinghai, China ' Qinghai Green Energy Data Co., Ltd., Xining 810001, Qinghai, China ' Qinghai Green Energy Data Co., Ltd., Xining 810001, Qinghai, China ' State Grid Qinghai Electric Power Company, Xining 810001, Qinghai, China ' Qinghai Green Energy Data Co., Ltd., Xining 810001, Qinghai, China

Abstract: To address the challenges of multi-source heterogeneous data coupling, multi-scenario strategy dynamic adaptation, and small-sample modelling in the operational performance evaluation of multi-source energy storage power stations, this paper proposes a hierarchical physical-data fusion modelling and causal reinforcement learning collaborative evaluation method. Firstly, a mechanism-data hybrid model for underlying multi-type energy storage units is constructed, embedding physical constraints of energy storage units such as flywheel rotor dynamics equations and flow battery Nernst correction equations. Secondly, a scenario-driven multi-time-scale dynamic aggregation model is proposed, establishing an equivalent circuit network for multi-source energy storage power stations based on Kron reduction theory, and analysing the nonlinear coupling mechanism of power distribution in peak-shaving (minute-level) and frequency regulation (second-level) scenarios. Thirdly, a causal reinforcement learning framework is designed, integrating the structural causal model and deep deterministic policy gradient algorithm to achieve closed-loop optimisation of cross-scenario control strategies and power station operational performance indicators. The verification based on a provincial 'pumped storage-flywheel-lithium battery-flow battery' multi-source energy storage demonstration project shows that the method applies to operational performance evaluation in peak shaving, frequency regulation, and other scenarios.

Keywords: multi-source energy storage power station; operational performance evaluation; cluster modelling; physics-data fusion; causal reinforcement learning.

DOI: 10.1504/IJETP.2025.151791

International Journal of Energy Technology and Policy, 2025 Vol.20 No.7, pp.109 - 126

Received: 28 Sep 2025
Accepted: 10 Dec 2025
Published online: 19 Feb 2026 *

Title: Causal reinforcement learning-based operational performance evaluation of multi-source energy storage power stations

Keep up-to-date