文化基因算法优化PID神经网络系统辨识

Memetic Algorithm Optimize PID Neural Network System Identification

摘要:

PID神经网络(PIDNN)是一种融合比例、微分、积分环节,结构简单固定,且具备动态网络特点的神经网络模型,适合于非线性系统辨识。但是网络对初始权值和样本质量敏感,参数难以选定,导致网络收敛速度慢,容易陷入局部极小。提出一种采用文化基因算法(Memetic Algorithm)优化网络权值的方法。在差分进化(DE)算法全局寻优结果基础上,通过混沌局部搜索算法,进一步优化网络权值；根据PIDNN特性,在优化过程中加入先验知识,采用L1正则项,对目标函数正则化,避免算法搜索到无潜力解,保证网络模型泛化能力。对一杂非线性系统进行辨识仿真,仿真结果表明优化后的神经网络辨识精度高,有良好的泛化能力。

Abstract:

PID neural network (PIDNN) is a neural network model that integrates proportional, differential and integral links, and has simple structure . It is suitable for nonlinear system identification. However, the BP algorithm used in the network is sensitive to the initial weight and sample quality, and it is difficult to select the parameters, which leads to the slow convergence of the network and easy to fall into the local minimum. A method of optimizing network weight by cultural genetic algorithm is proposed. Differential evolution algorithm in global optimization based on the results, the chaotic local search algorithm, and further optimize the network weights; according to the prior knowledge, using L1 regularization, the objective function of the regularization algorithm to search, to avoid potential solutions, ensure the generalization ability of network model. A hybrid nonlinear system is identified and simulated. The simulation results show that the optimized neural network has high recognition precision and good generalization ability.