Abstract: The current experimental simulations usually use environmental temperature or fixed value instead of photovoltaic module temperature which result in a deviation of modeling results. To solve this problem,the paper is based on numerical weather prediction（NWP）,takes use of adaptive partial least squares regression method（APLSR）,BP neural network method and ensemble prediction method to establish a prediction model of PV module temperature,to analysis and predict temperature of 17 PV modules with different angles. The result show that,The ensemble prediction method has the best prediction effect. In autumn and winter,the prediction error of 0°-25° inclination angle and the west wall 90° are lower than the others,with RMSE no more than 3 ℃. The variation of inclination error in spring and summer is relatively stable,as for ensemble prediction method,the RMSE is between 2.1 ℃（0°）and 3.0 ℃（45°）in spring;as for summer,from 0° to 90°,the RMSE decreases,and RMSE of 90° is the lowest（1.9 ℃）;for different weather conditions,the prediction error is relatively minimal in sunny day,and the prediction error fluctuates in raining and cloudy days.Finally,the paper also verified the optimal title angle of high and low temperature days in Wuhan. The RMSE in high temperature day is1.2 ℃ and in low temperature day is 1.6 ℃.