Abstract: The output of wind and solar power stations is intermittent and fluctuating, but their complementarity can be fully realized with reasonable wind-solar capacity ratio. The inaccurate theoretical power calculation of power station affects the extraction of true features of wind and solar, and then leads to large capacity matching error. The model-generator method commonly used in theoretical power calculation of new energy power stations is improved in this study. The abnormal data are firstly identified and reconstructed, and then the abnormal model-generators are identified and updated. Furthermore, according to the actual operation of non-model-generators, the dynamic information window is selected. By using the measured power of model-generator and non-model-generator in the dynamic information window, the non-model-generators are grouped and the scale coefficient of each group of non-model-generator is dynamically identified, and thus the theoretical power of new energy power stations is calculated. Based on the calculation of historical theoretical power for many years, the characteristics of new energy power stations are analyzed. The random output scenario is simulated accordingly, and the optimization method of wind-solar capacity ratio based on the risk of source-load mismatching is established. The accuracy of the improved model-generator method is verified by examples. By using this method, the improved theoretical power data of wind farm and photovoltaic power stations in a certain area of northwest China are obtained, and the optimal ratio of wind-solar capacity in this area is optimized. The improved model machine method proposed in this paper can effectively improve the calculation accuracy of theoretical power of new energy power stations, and the optimal ratio of wind-solar capacity calculated based on this method can provide reference for wind-solar power generation capacity of regional power grid.