Considering the duality of cyclical changes and environmentally sensitive in micro-grid power output, micro source control (MSC)is used into the learning algorithm of wavelet neural network model to predict distributed power grid power. Based on the flexibility of dealing with the local and cycle nature of the power, and combined with the impact of environmental factors on the power change, the correlation factor is introduced into the algorithm to optimize the weight and gets the final prediction results. Through the actual network system simulation test, and compared with BP neural network and GRNN model, the results show that: the relative errors of MSC-WNN model in three tests were within -1% to 1%, showing its high precision and good robust performance.