Optimization Research of Genetic Algorithm in Improving Back Propagation Neural Network

The results show that back propagation neural network (BP) has a strong non-linear mapping capability and flexible structural design, and it has wide application in many fields, such as function approximation, pattern recognition, classification, data compression and so on, but it still has some inherent defects. In order to overcome these limitations, in this paper, genetic algorithm (GA) is introduced to optimize BP initial weights and thresholds, and uses the global search capability, parallel computing advantage, adaptability and diversity of GA to accelerate the convergence rate of the network, enhance the generalization ability, increase the diversity of solutions, and avoid the local minimum trap. The results indicate that, compared with conventional BP, the BP optimized by GA shows faster convergence rate in the training process, and the model performance is significantly improved. Specifically, the R squared (${\mathrm {R}}^{2}$) is close to 1, and root mean square error (RMSE), mean absolute error (MAE), mean bias error (MBE) are all small, indicating low model prediction error, high fit degree, and small system deviation.