Power Load Forecasting Based on BP Neural Network Optimized by New Improved Genetic Algorithm

The instability and low accuracy of short-term forecasting of electric power loads, this study deliberately proposes a new improved genetic algorithm through continuous rehearsal. In turn, the accuracy of forecasting model of BP neural network can be improved. It improves the BP neural network's weight initialization, smoothing factor and thresholds, increasing adaptive learning rate and additional momentum of BP neural network, improving the efficiency of searching. Due to the defects of genetic algorithm which is precocious and frequent traps in local optimal solutions, we improve the selection/crossover and mutation operators. Next, we verify the effectiveness of the (Im-GA) -BP model, the power load data of an area were used for training and forecasting, and the improved (Im-GA)-BP model, BP model and actual load were compared. The final data for the relative error of the improved model is reduced by 6.18% and the root mean square error is 5.11 % lower than before. Therefore, the method has a higher prediction accuracy compared to the general BP algorithm.