节点文献

神经网络在非线性预测控制中的应用研究

Research on Application of Neural Network in Nonlinear Predictive Control

【作者】 李睿

【导师】 黄西平; 刘军;

【作者基本信息】 西安理工大学 , 模式识别与智能系统, 2005, 硕士

【摘要】 预测控制自上世纪70年代产生以来,因其对模型要求低、具有较强鲁棒性等特点,在生产过程中获得了良好的应用。目前,对线性系统预测控制算法的研究已经比较深入,相对理论也比较成熟。而对于非线性系统,因其自身的复杂性,使得寻找一种统一的非线性预测控制方法很困难。本文围绕着预测模型、反馈校正、滚动优化三项预测控制的基本原理对神经网络技术在非线性预测控制中的应用进行了深入的研究和探讨,并最终给出了相应的研究成果。主要研究工作概况如下:1. 对预测控制算法中典型的动态矩阵控制算法进行了研究,并通过实例仿真分析了在预测控制中预测时域P、控制时域M以及其他参数对控制性能的影响。2. 针对在非线性预测控制中广泛应用的BP神经网络存在收敛速度慢,容易陷入局部最小的缺点,本文提出了一种基于并行拟牛顿优化算法的并行拟牛顿神经网络。同BP神经网络和BFGS拟牛顿神经网络相比,该神经网络具有收敛速度快,模型精度高的特点,更适合于实时非线性控制。此外,通过对局部动态递归网络中的Elman网络中上下文单元层各神经元增加自反馈功能的方法,提高西安理工大学硕士学位论文 了Elman网络对动态非线性系统的泛化能力。3.针对神经网络递推多步预测会随着预测时域的增人而产生累积模型误差的问 题,本文提出了一种模糊白补偿式在线反馈校正方法。该方法利用专家经验建 立模型误差随时间变化的规则表,并通过杳表对补偿模型误差进行修正,提高 了对神经网络预测模刑输出的校正质量。4.在研究并分析当前有关基于神经网络预测模烈的滚动优化方法的基础上,提出 了一种神经网络白适应Pl预测控制策略。该方法对系统内在参数变化和模型的 不确定性问题通过在线的预测白适应调节增加到PI控制器中,使系统响应速 度变快,并且能够在较短的时间达到系统稳态值。关键词:神经网络:非线性系统;预测控制;井行拟牛顿算法;神经网络自适应 P工预测控制

【Abstract】 Based on the study status of model predictive control and the problem on application neural network in nonlinear predictive control, some problems are studied and discussed in the article, also the corresponding results are given. The main contents are as follows:1. Based on the basic structure and theory of dynamic matrix predictive control, the predictive model, the method of revising feedback, receding horizon optimization and the influence on performance of controlling of the predictive horizon P, control horizon M and others parameters is analyzed in detail.2. To improve the rate of convergence of the BP neural networks for nonlinear system identification in nonlinear predictive control, a novel parallel quasi-Newton optimization Technique is proposed and as multi-step predictive model of nonlinear industrialized process. The simulation results confirm the algorithm is able to increase the precision of nonlinear predictive model greatly and improve the rate of rate of convergence of neural network. Moreover, a novel recurrent neural network - Elman be given for dynamic nonlinear system predictive model.3. To improve the revising feedback performances, a fuzzy self-compensating correction method for nonlinear multi-step prediction is presented. The simulation results showing the effectiveness fend robustness of the algorithm in nonlinear predictive control and dynamic of the controlled system.4. Based on the analysis present receding horizon optimization methods for neural network predictive model, a neural network adaptive PI controller is proposed. The parameters of the PI controller are tuned by the error of predictive output and the real system output. The simulation results prove the method is effective in adaptation and tracking of the different process operating.

  • 【分类号】TP273.2
  • 【被引频次】14
  • 【下载频次】953
节点文献中: