摘要
针对递归模糊神经网络(Recurrent fuzzy neural network,RFNN)的递归量难以自适应的问题,提出一种基于小波变换–模糊马尔科夫链(Wavelet transform fuzzy Markov chain,WTFMC)算法的RFNN模型.首先,在时间维度上记录隐含层神经元的模糊隶属度,并采用小波变换将该时间序列进行分解,通过模糊马尔科夫链对子序列的未来时段进行预测,之后将各预测量合并后代入递归函数中得到具有自适应性的递归量.其次,利用梯度下降算法更新RFNN的参数来保证神经网络的精度.最后,通过非线性系统建模中几个基准问题和实际污水处理中关键水质参数的预测实验,证明了该神经网络模型的可行性和有效性.
Aiming to solve the problem that the recursive variable in the recurrent fuzzy neural network(RFNN)is difficult to be self-adaptive,this paper proposes an RFNN structure model based on wavelet transform fuzzy Markov chain(WTFMC).Firstly,it records the fuzzy membership degree of hidden layer neurons in time dimension,and decomposes the time series by wavelet transform.The future period of the subsequence is predicted by fuzzy Markov chain,and the adaptive recursive variables are obtained by combining the predictors into the recursive function.Secondly,the gradient descent algorithm is utilized to update the parameters of RFNN in order to ensure the accuracy of neural network.Finally,the feasibility and validity of the neural network model are demonstrated by several benchmark problems in nonlinear system modeling and the prediction of key water quality parameters in the practical wastewater treatment.
作者
乔俊飞
丁海旭
李文静
QIAO Jun-Fei;DING Hai-Xu;LI Wen-Jing(College of Electronic Information and Control Engineering,Beijing University of Technology,Beijing 100124;Beijing Key Laboratory of Computational Intelligence and Intelligent System,Beijing 100124)
出处
《自动化学报》
EI
CSCD
北大核心
2020年第11期2367-2378,共12页
Acta Automatica Sinica
基金
国家自然科学基金(61533002,61603009)
北京市自然科学基金(4182007)
北京市教委科技一般项目(KM201910005023)
北京工业大学日新人计划(2017-RX(1)-04)资助。
关键词
递归模糊神经网络
小波变换
模糊马尔科夫链
动态建模
污水处理
Recurrent fuzzy neural network
wavelet transform
fuzzy Markov chain
dynamic modeling
wastewater treatment
