一种结合PSO及改进BP网络的辨识方法

当辨识神经网络的类型和结构确定后，初始权值等辨识参数直接影响到辨识效果，而依靠先验知识试凑而得的参数值往往难以达到最佳效果。针对这一问题，提出了一种结合粒子群（PSO）算法及引入动量项的改进BP网络的辨识方法，利用PSO对改进BP网络辨识的初始权值/偏置、学习率、动量系数进行寻优，并将优化后的神经网络模型用在控制系统中进行修正，进一步完善辨识模型。应用在热工系统中，仿真结果表明了该辨识方法的有效性。

Identification method combining PSO and improved BP neural network

After type and structure of identification neural network have been determined,identification effect can be directly influenced by identification parameters like neural network weighting initial value.However,it＇s hard to acquire satisfactory identification performance with parameters determined by method of trial and error according to experience.Aiming at this problem,an identification method combining Particle Swarm Optimization（PSO） algorithm and improved BP Neural Network（NN） introducing momentum coefficient in learning algorithm is put forward.This method uses particle swarm optimization algorithm to optimize the parameters of NN weighting initial value,NN offset initial value,speed of learning and momentum coefficient in using improved BP neural network to identify.Subsequently,the optimized neural network model is modified in control system,which perfects the identification model.The application in thermal process shows the identification method is effective.