Wind turbine condition monitoring and wind power prediction both rely heavily on power curves. Firstly，to increase the modeling accuracy of wind turbine power curves，the random forest technique was used to screen the important variables that influence wind energy capture ability. Then，the screened variables were fed into the improved Gaussian process（GP） model，which improved computational efficiency. Finally，four separate metrics were used to evaluate the model's correctness，and the entropy weight approach was used to resolve any potential conflicts between the metrics，resulting in a comprehensive assessment metric that measured the quality of the power curve model. The suggested approach's effectiveness was validated using supervisory control and data acquisition （SCADA） data from a wind farm in the United Kingdom，and the findings reveal that the proposed method improves model accuracy when compared to the current six types of conventional methods.