引用本文:韩改堂,乔俊飞,韩红桂.基于自适应递归模糊神经网络的污水处理控制[J].控制理论与应用,2016,33(9):1252~1258.[点击复制]
HAN Gai-tang,QIAO Jun-fei,HAN Hong-gui.Wastewater treatment control method based on adaptive recurrent fuzzy neural network[J].Control Theory and Technology,2016,33(9):1252~1258.[点击复制]
基于自适应递归模糊神经网络的污水处理控制
Wastewater treatment control method based on adaptive recurrent fuzzy neural network
摘要点击 3615  全文点击 2068  投稿时间：2015-12-07  修订日期：2016-09-30
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2016.50965
  2016,33(9):1252-1258
中文关键词  污水处理  递归模糊神经网络  自适应学习率  基准仿真模型(BSM1)
英文关键词  wastewater treatment  recurrent fuzzy neural network  adaptive learning rate  benchmark simulation model 1 (BSM1)
基金项目  国家自然科学基金项目(61622301, 61533002, 61225016), 北京市教育委员会科研计划项目(KZ201410005002, km201410005001), 教育部博士点 基金项目(20131103110016)资助.
作者单位E-mail
韩改堂 北京工业大学 hangaitang@emails.bjut.edu.cn 
乔俊飞* 北京工业大学 junfeiq@bjut.edu.cn 
韩红桂 北京工业大学  
中文摘要
      针对污水处理过程中具有的非线性、大时变等特征, 提出了一种基于自适应递归模糊神经网络(recurrent fuzzy neural network, RFNN)的污水处理控制方法. 该方法利用自适应RFNN识别器建立污水处理过程的非线性动 态模型, 建立的模型可以为RFNN控制器提供污水处理过程中的状态变量信息, 保证了控制器根据系统响应调整操 作变量的精确性; 并且RFNN辨识器及RFNN控制器基于自适应学习率进行学习, 确保了递归模糊神经网络的收敛 精度和速度, 并通过构造李雅普诺夫函数证明了此算法的收敛性; 最后, 基于基准仿真模型(benchmark simulation model 1, BSM1)平台进行仿真实验. 结果表明, 与PID、模型预测控制及前馈神经网络相比, 该方法对污水处理中溶 解氧浓度和硝态氮浓度的跟踪控制精度具有明显的提升.
英文摘要
      Due to the nonlinear and highly time-varying issues of wastewater treatment processes, a wastewater treatment control method based on adaptive recurrent fuzzy neural network (RFNN) is proposed. Firstly, the adaptive RFNN identifier is used to establish the nonlinear dynamic model of wastewater treatment process. The model can afford the state variable information of wastewater treatment process to RFNN controller, which can ensure the accuracy of manipulated variable is adjusted by controller. Secondly, RFNN identifier and RFNN controller are learning through gradient descent algorithm with an adaptive learning rate, which guarantee the convergence of learning process of RFNN, and a function is constructed by lyapunov theory to prove the convergence of this algorithm. Finally, the simulation experiment carried out based on BSM1 platform. Compared with PID, model predictive control and forward neural network control techniques, the simulation results show that the proposed method can improve obviously the control accuracy of wastewater treatment.