A short-term load forecasting method based on similar day algorithm with dynamic weight value and GM- GRNN model is proposed. Traditional ways to select similar days can not effectively identify the dominant factors. Aiming at this problem, the dynamic weight similarity day selection algorithm based on modified fruit fly optimization algorithm (MFOA) is proposed, which is based on the optimization algorithm of different regions and seasons. The method improves adaptability and effectiveness of the algorithm. Moreover, the accuracy of load forecasting is enhanced. After the selection of similar days, the short-term load forecasting method of grey model and generalized regression neural network (GRNN) is proposed, which improves the stability of the single model forecasting accuracy. In order to further optimize the prediction model, the GRNN smoothing factor is optimized by the modified cuckoo search (MCS) algorithm. Simulation results verify the validity of the proposed method. This work is supported by National Natural Science Foundation of China (No. 51267001), Guangxi Scientific Research and Technological Development Program (No. 14122006-29), and Natural Science Foundation of Guangxi Province (No. 2014GXNSFAA118338).