It is important for the energy conservation and emissions reduction to accurately predicate the power of Photovoltaic micro-source in a certain period of time in the future. In this paper, by comparing the power and meteorological history data, analyzes the factors such as weather, solar radiation and temperature in the photovoltaic power generation power prediction ,based on the global optimization searching performance of the genetic algorithm and the time-frequency localization of the wavelet neural networks, microgrids photovoltaic power generation forecasting model has been established. Through case analysis, the results show that wavelet neural network based on genetic algorithm has better learning ability and generalization ability. And in the aspect of microgrids photovoltaic power, the forecasting model is more valuable in practical application.