Since the traditional PID control algorithm has many problems in parameter selection, it cannot meet specific requirements in engineering practice. However, the PID control algorithm after BP neural network tuning can realize adaptive learning and further improve the control ability, but due to the randomness of its initial weight selection, It is easy to lead to inconsistent training results and affect system stability. To solve these problems, this paper proposes to use the improved mayfly algorithm (IMA) to optimize BP neural network. By taking advantage of the powerful advantages of the improved mayfly algorithm in global search, the optimal position is found as the initial weight of BP neural network. Compared with the traditional method, the overshoot of IMA-BP-PID is only 0.28% in the second order control system, and the overshoot is greatly reduced without steady-state error, which can be better applied to the actual control system.

No abstract