In view of the fluctuation and the intermittence of the output power of the photovoltaic system, based on principal component analysis (PCA) and genetic algorithm (GA) optimization, a short term forecasting method of BP neural network power is proposed. Direct forecasting of output power is done by historical power data and real time meteorological factors. It uses principal component analysis to reduce the dimension of multiple original variables into a few independent variables, so that it can optimize the initial weights of back-propagation''s threshold and overcome the traditional BP neural network easy to fall into local minimum point, and the problems of slow convergence speed. The results of the comparison for different forecast models validate the effectiveness of the algorithm and proposed model. This work is supported by Natural Science Foundation of Liaoning Province (No. 2013020141).