超临界机组燃水比GAP-DRFNN的优化控制  

Growing and pruning dynamic recurrent fuzzy neural network optimal control on firing rate to feed water ratio for supercritical unit

作　　者：周洪煜[1] 汪正海[1] 张振华 童明伟[1] 

机构地区：[1]重庆大学动力工程学院,重庆400044 [2]中国大唐集团科学技术研究院,北京100032

出　　处：《重庆大学学报（自然科学版）》2013年第6期84-90,共7页Journal of Chongqing University

基　　金：重庆市科委重大科技攻关项目(CSTC;2009AB1008)

摘　　要：分析某厂直流锅炉燃水比控制系统,设计出基于生长剪枝动态递归模糊神经网络(Growing and pruning dynamic recurrent fuzzy neural network,GAP-DRFNN)的燃水比优化控制系统,GAP-DRFNN以汽水分离器出口工质温度偏差最小为导出信号,综合学习燃水比控制主要状态参数,实时输出燃水比最佳控制量。随着当前主要相关状态参数的输入,GAP-DRFNN通过结构学习,自动增加和修剪神经元,而且根据梯度下降法,动态调整网络的隶属函数参数以及神经网络递归权值。实验结果表明:此方案中燃水比控制可兼顾快速性和准确性,在变工况时系统仍具有优异的动静态性能,控制效果优于传统PID控制。By analyzing a certain once through boiler＇s FR/FW control system,a new control scheme based on growing and pruning dynamic recurrent fuzzy neural network（GAP-DRFNN）is proposed.This GAP-DRFNN can synthetically study main relative state parameters about FR/FW control,so as to calculate the optimal FR/FW by using least temperature deviation value of outlet of moisture separator as its training signal.As the data of current main relative state parameters input,GAP-DRFNN through structure learning can automatically increase and pruning neurons,and adjust the parameters and the recurrent weight of neural network dynamically based on stochastic gradient descent algorithm.The experimental results show the good performance for the system in variable conditions and this scheme＇s celerity and precise on FR/FW control,it has better quality than traditional PID control method.

关 键 词：燃水比 动态递归模糊神经网络 生长剪枝 变结构控制 超临界机组 

分 类 号：TK223.7[动力工程及工程热物理—动力机械及工程]