LMBP神经网络PID控制器在暖通空调系统中应用研究.docVIP

LMBP神经网络PID控制器在暖通空调系统中应用研究 　　摘要:针对BP神经网络学习过程收敛速度慢及易陷入局部极小值的缺陷,研究了阻尼最小二乘算法(即LM算法),为解决LM算法中学习速率的选择和逆矩阵的求解这两个严重影响训练时间和收敛精度的问题,采用LU分解法对LM算法进行改进和优化,并通过MATLAB语言编程实现,将得到的LMBP神经网络PID控制器应用于暖通空调冷冻水循环的控制回路中,将其控制效果与PID控制算法、BP神经网络PID控制算法进行仿真对比研究。研究结果表明,采用LMBP神经网络PID控制器在减少超调量、加快收敛速度、减少稳态误差等方面的性能都得到了明显地改善和提高。 　　关键词:最小二乘算法;反向传播;神经网络;比例积分微分控制;暖通空调系统 　　中图分类号:TP273.2 文献标识码:A 　　Study on Application of LMBP Neural Network PID Controller in HVAC System 　　WANG Yan1,2,LIU Hong-li1#8224;,YANG Ke2 　　(1College of Electrical and Information Engineering, Hunan Univ, Changsha Hunan 410082, China ; 　　2 College of Electrical Engineering, Nanhua Univ, Hengyang, Hunan 421001, China) 　　Abstract: In allusion to the defects of the slowly converging and easily immerging in partial minimum in the learning process of Back Propagation (BP) Neural Network (NN), a Levenberg-Marquardt (LM) algorithm is presented. In order to solve the two problems which seriously affect on both training time and converging accuracy in terms of the choice of learning rate and the inverse matrix solving in LM algorithm, the paper proposed LU decomposition to improve the LM algorithm simulated by MATLAB. The LMBP Neural Network PID controller is applied in cooling water cycle of heating ventilating and air conditioning (HVAC) system. The results of LMBP neural network PID controller, PID controller, and BP neural network PID controller are simulated and compared on the paper.The performance indicates obviously that the BP Neural Network PID controller optimized by improved LM algorithm is better than other algorithms in terms of reducing the overshoot, speeding up the convergence rate, and reducing the steady-state error. 　　Key words: Levenberg-Marquardt algorithm; back propagation; neural network; PID control; HVAC 　　 　　误差反向传播神经网络(简称BP网络),所具有的任意非线性表达能力,可以通过对系统性能的学习来实现具有最佳参数组合的PID控制[1]。 　　虽然BP神经网络的理论依据坚实,推导过程严谨,通用性强,但是由于其算法是基于最陡梯度下降算法、以误差平方和为目标函数的,所以其不可避免地存在一些问题,极大地影响了BP网络的进一步发展和应用[2]。 　　为此,本文通过改进的阻尼最小二乘算法(即LM算法),对