Abstract: An optimal control strategy for thermal power-new energy coupled power systems considering rolling update of new energy load shedding ratios was proposed. First, the overall control architecture for the thermal power and new energy coupled system was designed, and evaluation metrics for both operational economy and regulation performance of thermal power plants were established. For the thermal power-new energy coupled system model, combined with the ultra short term prediction results of photovoltaic/wind power, the optimal load shedding ratio of new energy was obtained through exhaustive optimization, and then a training dataset was constructed. After that, a single-step prediction method for load shedding ratios based on BP neural networks and autoencoders was proposed, with a test dataset constructed using actual unit data. Results show that the proposed algorithm enables autonomous switching and adjustment of operating modes, which can balance the operational economy and regulation effectiveness of the thermal power-new energy coupled system. Finally, comprehensive performance of the system has been improved.