With the rapid development of economy, short-term power load forecasting has become an indispensable task for power system in China. In this paper, firstly, considering daily weather factor, a prediction model is established by BP neural network to realize the prediction of power load. Then a genetic algorithm (GA) optimized BP neural network (GA-BP) is developed to achieve the power load prediction. The better prediction results are achieved, compared with BP neural network. Due to the shortcoming of poor local search ability and easy precocity in GA, a improved genetic algorithm optimized BP neural network (IGA-BP) is proposed to optimize power load forecasting model. The example shows that the IGA-BP can effectively improve the accuracy of short-term power load forecasting.