Current Control of EAST Fast Control Power Supply Based on BP Neuron Network Predictive PI

Abstract

Abstract The dynamic response speed of output current is one of the most important performance indexes of EAST fast control power supply. To further improve the response speed of output current, a predictive PI control algorithm of BP neural network is proposed to overcome the shortcomings of large computation and slow convergence in traditional BP neural network. A branch current prediction model of EAST fast control power supply is established. Based on linear PI control, the BP neural network structure is combined, and the current prediction model is used as the adaptive objective function. To simplify the calculation process of the processor and speed up the convergence rate, a piecewise linear activation function is adopted. The learning rate of BP neural network structure and the activation function factor of the output layer are adjusted in real time according to the error between the predicted current and the reference signal. The online predictive adaptive setting of PI control parameters is realized by the steepest gradient descent method. Simulation and experiments results show that the proposed control algorithm has faster dynamic response speed than the traditional PI control.