并行混沌神经网络建模方法应用研究

摘要

针对开关磁阻电动机的非线性特点及其现有建模方法存在初始网络权值参数随机给定和易于陷入局部最小点的缺点,提出了一种采用并行优化混沌BP神经网络的建模方法.该方法首先利用混沌系统对神经网络权值向量、阈值向量进行初始优化,然后利用BP神经网络的Levenberg-Marquardt算法进行收敛训练,如果陷入局部最小点则再次使用并行混沌搜索进一步优化模型,使模型具有精度高、速度快的特点.模型训练和开关磁阻电动机调速系统动态仿真结果表明,采用该方法建立的模型运行平稳,系统动态性能好,响应速度快.%In view of nonlinear characteristic of switched reluctance motor and existing modeling method has shortcomings of random initial weights of network parameters and is easy to fall into local minimum point,the paper put forward a modeling method using parallel optimization chaotic and BP neural network.Firstly,the method uses chaotic system to optimize neural network weight vector and initial threshold vector,and then uses Levenberg-Marquardt algorithm of BP neural network to train convergence.If it drops into the local minimum point,then it needs to use parallel chaotic search to optimize model again,so as to make the model have characteristics of high precision and fast speed.The dynamic simulation results of training model and speed-regulation system of switched reluctance motor show that the model established by the method has stable operation,good dynamic performance,and fast response speed.