基于改进PSO⁃BP神经网络的PID参数优化方法

PID parameter optimization method based on improved PSO⁃BP neural network

针对传统PID控制器在面对实际对象时在线整定困难的问题,提出采用BP神经网络与PID控制器相结合,并采用粒子群算法对其网络权值矩阵进行优化,但在用粒子群算法优化BP神经网络PID控制器的参数时存在收敛速度不够快,易陷入局部最优解等问题。提出通过改进粒子群算法中惯性权重由常用的线性递减改为随机权重,然后将其最优粒子用于优化BP神经网络PID控制器的网络初始权值矩阵以得到更优的参数。最后通过仿真实验得到其相较于标准粒子群算法有更好的适应度函数曲线,并且其超调量为10.4%,调节时间为0.31 s,均小于同一传递函数下的BP神经网络PID和用标准粒子群算法优化的BP神经网络PID。结果表明该方法相较于BP神经网络PID和用标准粒子群算法优化的BP神经网络PID更具有优越性。

It is difficult for the traditional PID controller to implement online tuning when it is faced with actual objects,so the researchers have proposed a way to combine the BP neural network with the PID controller,and the particle swarm optimization(PSO)algorithm is adopted to optimize the network weight matrix of the PID controller.However,when optimizing the parameters of the BP neural network PID controller by the PSO algorithm,the PSO algorithm′s convergence speed is not fast enough,and the PSO algorithm is susceptible to falling into the local optimal solution.In this paper,the inertia weight in the improvement particle swarm optimization(IPSO)algorithm is changed from the commonly⁃used linear decreasing to random weight,and then the optimal particles obtained by IPSO are used to optimize the network initial weight matrix of the BP neural network PID controller to obtain better parameters.It is obtained in the simulation experiment that the proposed method has the better fitness function curve than the standard PSO algorithm,and its overshoot(10.4%)is smaller and its tuning duration(0.31 s)is shorter than those of the BP neural network PID under the same transfer function and the BP neural network PID optimized by the standard PSO algorithm.The results show that the method is superior to BP neural network PID and BP neural network PID optimized by standard PSO algorithm.