基于LM-PSO算法和BP神经网络的非线性预测控制  

Nonlinear Predictive Control Based on LM-PSO Algorithm and BP Neural Network

作　　者：王炳萱 李国勇[1] 王艳晖[1] 

机构地区：[1]太原理工大学信息工程学院,太原030024

出　　处：《太原理工大学学报》2016年第2期207-211,共5页Journal of Taiyuan University of Technology

基　　金：国家自然科学基金资助项目:改善电液伺服系统动态特性的双自由度回路原理及控制方法(51075291)

摘　　要：针对非线性系统,提出了一种基于BP神经网络的预测控制方法。以BP神经网络建立多步预测模型并预测系统输出值,用LM(Levenberg-Marquardt)算法和PSO(Particle Swarm Optimization)算法组合的混合算法对目标性能指标函数进行滚动优化求解,得到非线性系统的最优控制量;利用误差修正参考输入法实现反馈矫正。通过将粒子群算法引入LM算法,克服了LM算法依赖初值和粒子群算法过早收敛于局部极值的问题,提高了求解的运行速度和精确度。通过对单变量非线性系统仿真实验,证明了该控制系统具有良好的稳定性、自适应性和鲁棒性。该方法可在数学模型不确定的情况下设计出有效的预测控制器。In this paper,a multistep predictive control method for nonlinear systems was proposed,which uses a Back Propagation（BP）neural network as a model.First,a multi-step predictive model based on BP neural network was constructed and applied to predict the output of the system.Then the optimal control values were obtained by the rolling LM-PSO optimization algorithm,which was combined with Levenberg-Marquardt（LM）algorithm and Particle Swarm Optimization（PSO）algorithm.Feedback correction was achieved by modifying reference input according to the error.The PSO algorithm was introduced into the LM algorithm to overcome the limitation of initial value,enhance the ability of escaping from local optima and improve the speed and precision of the solution.This method can be used to design effective predictive controllers for univariate nonlinear systems with uncertain mathematical models.The simulation results demonstrated the self-adaptive ability,robustness and efficiency of the proposed method.

关 键 词：非线性系统 预测控制 LM算法 粒子群算法 BP神经网络 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]