AI-based carbon peak prediction and energy transition optimization for thermal power industry in energy-intensive regions of China

• Developed an Artificial Intelligence-based framework for energy transition research. • Established explicable carbon emission indicators for the thermal power industry. • Predicted multi-scenario carbon emission peaks of thermal power by machine learning. • Analyzed spatial patterns of carbon peaks and proposed energy transition strategies. As the largest carbon emitter, China faces an increasingly critical trade-off between the economy and the environment. Despite its recent increasing adoption of renewable energy, China continues to generate excessive emissions, particularly from its dominant thermal power sector. Against this background, this study selected the East China Region, where energy consumption is permanently highest, to implement an AI-based three-step “Indicator Screening − Scenario Prediction − Policy Optimization” framework. Firstly, a highly explanatory system of carbon emission impact indicators in the thermal power industry was established utilizing an Optimal Parameters-based Geographical Detector. Secondly, multi-scenario predictions of carbon emissions from the thermal power industry were conducted based on robust Random Forest models. Lastly, the tailored energy transition strategies were suggested according to the spatial distributions of carbon peak time nodes under each scenario. The results showed that, compared to the baseline, the carbon peak under the Economic Development Scenario will be delayed by three years, with an additional 92.74 Mt CO 2 ; while under the Environmental Protection and Energy Transition Scenarios, the peak will be advanced by five and three years, with 106.48 and 73.86 Mt CO 2 reductions, respectively. Leveraging multi-source data-driven AI models, this study efficiently provided reliable quantitative support for measuring policies with various priorities, emphasizing the necessity of implementing balanced energy transition strategies. Furthermore, through intelligent scenario simulation and optimal decision-making, the proposed replicable and scalable methodological framework facilitates achieving relevant Sustainable Development Goals (e.g., SDG 7, 12, and 13) across different industries and regions.