Abstract: Aiming at the problem of oscillation caused by excessive selection of learning rate in the learning process of traditional BP neural network, we proposed a new adaptive tuning algorithm for PID (proportional-integral-differential) parameters of BP neural network. BP neural network was used to adjust and optimize PID parameters adaptively, and momentum factor was used to optimize learning rate and increase momentum term to restrain oscillation phenomenon in BP neural network training, so as to accelerate convergence speed. The experimental results show that the proposed algorithm can effectively alleviate the oscillation phenomenon and accelerate the convergence speed of the algorithm.

Key words: PID parameter self-tuning, neural network, learning rate, momentum factor