Therefore, the purpose of BP algorithm improvement is to speed up training and avoid falling into local minimum. There are many improved methods, such as momentum-driven factor algorithm, adaptive learning rate, variable learning rate and action function retraction method.