摘要
针对控制系统参数在线整定(自适应自优化)方法在工程应用中存在的稳定性和性能指标欠佳与进一步智能化问题,首先提出结合小波分析、BP、RBF神经网络优点和自适应机制构造了自适应小波神经网络方法 (AWNN),进而提出将AWNN结合经典控制方法构成的AWNN-PID方法,最终提出了采用自回归移动平均向量时间序列算法预测输出替代控制系统实时输出,从而构成基于向量时间序列预测自适应小波神经网络的控制参数在线整定方法 (VARMA-WNN-PID);进而选择工程应用中最为常见的多阶延迟被控对象,对BPNN-PID、RBF-PID与该研究提出的AWNN-PID、VARMA-WNN-PID等4种方法进行计算机仿真对比实验(结合针对神经网络学习效率、惯性系数和预测算法阶数、步长的对比试验),验证了新方法具有可行性、工程应用可靠性、更好的快速性、更低的静差和更灵活的控制参数调整能力。
In order to solve instability, poor control performance and advanced intelligence problems, Adapted wavelet neural networks method, （AWNN, which combines the advantages of BPNN, RBF and wavelet neural network methods） is proposed firstly; Secondly, AWNN based online tuning PID control method （AWNN--PID, which combines the advantages of AWNN method and PID control method） is proposed; 3） Finally, Vector time series forecasting adaptive wavelet neural network PID control method （VARMA--AWNN--PID, which adjusts the control parameters according to the predicted system output） is proposed. Based on the comparative simulation experi- ments, with different parameters such as neural network learning efficiency, inertia coefficient, order of prediction algorithms and predict steps, between the BPNN--PID, RBF--PID, AWNN--PID and VARMA--WNN--PID control system, better feasibility, reliability, speed, lower static error, more flexible parameter adjustment ability are verified.
出处
《计算机测量与控制》
北大核心
2014年第2期427-430,共4页
Computer Measurement &Control
基金
国家自然科学基金资助项目(61174109)
国家发明专利(201110023946.6)
实用新型专利(201120020012.2)
关键词
在线整定
小波神经网络
向量时间序列
指标控制
预测控制
online--tuning
wavelet neural network
vector time series
index control
forecasting control
