改进的小脑模型神经网络及其在时间序列预测中的应用  

Improved Cerebellar Model Neural Network and Its Application in Time Series Prediction

作　　者：乔俊飞 董敬娇 李文静 QIAO Junfei;DONG Jingjiao;LI Wenjing(Beijing Key Laboratory of Computational Intelligence and Intelligent System,Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学信息学部计算智能与智能系统北京市重点实验室,北京100124

出　　处：《北京工业大学学报》2021年第6期598-606,共9页Journal of Beijing University of Technology

基　　金：国家自然科学基金资助项目(61890930-5,61533002,61603009);国家重点研发计划资助项目(2018YFC1900800-5);北京市自然科学基金资助项目(4182007);北京市教育委员会科技计划资助项目(KM201910005023)。

摘　　要：针对小脑模型神经网络(cerebellar model neural network,CMNN)中泛化能力与存储空间容量之间的冲突这一关键问题,提出了一种改进的小脑模型神经网络——模糊隶属度小脑模型神经网络(fuzzy membership cerebellar model neural network,FM-CMNN),用于解决非线性动态系统的时间序列预测问题.首先,FM-CMNN在保留原始CMNN输入变量的地址映射方式的情况下,在CMNN存储空间中引入铃型模糊隶属度函数,从而保证在不需增加量化级数的情况下提高网络的泛化能力.然后,使用梯度下降算法对网络权值进行更新,提高网络的逼近强度.最后,通过非线性时间序列预测基准实验和污水处理中水质参数预测实验,验证了FM-CMNN性能的可靠性.To tackle the key problem of the conflict between generalization ability and storage space capacity in the cerebellar model neural network(CMNN),an improved cerebellar model neural network-fuzzy membership cerebellar model neural network(FM-CMNN)was proposed to solve the time series prediction problem of nonlinear dynamic systems.First,FM-CMNN,while retaining the address mapping mode of the original CMNN input variables,the bell fuzzy membership function was introduced into the CMNN storage space,so as to improve the generalization ability of the network without increasing the quantization series.Then,the gradient descent algorithm was used to update the weights of the network to improve the approximation strength of the network.Finally,the reliability of the network was verified by the nonlinear time series prediction benchmark experiment and the water quality parameter prediction experiment in sewage treatment.

关 键 词：小脑模型神经网络 地址映射 模糊隶属度函数 泛化能力 非线性时间序列 预测精度 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]