基于随机森林可解释人工智能揭示储能与可再生能源在脱碳进程中的协同作用

Abstract The coordinated development of renewable energy (RE) and energy storage systems (ESS) is crucial for low-carbon transitions. Beyond optimal planning solutions, understanding the underlying reasons behind planning outcomes is essential to enhance decision-making transparency and reliability. This study investigates the evolving synergy between RE and MTES across decarbonization stages, proposing an explainable framework to attribute and analyze the factors influencing planning outcomes. By leveraging Random Forest (RF), the framework identifies key drivers behind RE-MTES synergies under diverse boundary conditions, such as carbon emission limits, resource endowments, and economic constraints. This approach provides a detailed understanding of how temporal and spatial factors shape planning decisions. A case study on representative Chinese provinces illustrates the dynamic evolution of RE-MTES collaboration: long-duration energy storage (LDES) supports seasonal balancing in RE-rich regions, while short-term energy storage (STES) mitigates intraday fluctuations in thermal-dominated areas. The RF-based analysis reveals that, at various decarbonization stages, LDES storage time, typically exceeding 100 h, significantly impacts system economics and efficiency. With a 20 % reduction in carbon emissions, the power generation structure plays a key role. However, beyond a 40 % reduction, carbon costs become the dominant factor in determining the economic viability of RE-MTES planning decisions. By offering actionable insights into the drivers of planning outcomes, this study advances the explainability of collaborative RE-MTES strategies, sup-porting more transparent and region-specific decision-making for low-carbon transitions.