基于ODCGWO-Elman的生产线设备软测量模型  

Soft Measurement Modeling of Production Line Equipment Based on ODCGWO-Elman

作　　者：宋佩林 崔振丰 陆鹏 王清云 胡成威 李岩[1] SONG Peilin;CUI Zhenfeng;LU Peng;WANG Qingyun;HU Chengwei;LI Yan(College of Electrical and Electronic Engineering,Changchun University of Technology,Changchun Jilin 130012,China;Jilin Kaidi Technology Co.,Changchun Jilin 130000,China;Powertrain Plant,China FAW Group Co.,Ltd.,Changchun Jilin 130000,China)

机构地区：[1]长春工业大学电气与电子工程学院,吉林长春130012 [2]吉林省凯迪科技有限公司,吉林长春130000 [3]中国第一汽车股份有限公司动力总成工厂,吉林长春130000

出　　处：《机床与液压》2024年第21期155-161,共7页Machine Tool & Hydraulics

基　　金：吉林省科技发展计划项目(20210201073GX)。

摘　　要：由于设备本身的复杂多样性以及传感器应用范围和成本的限制,导致部分故障检测信息无法完整、准确获得。为解决上述问题,提出一种基于改进灰狼算法的Elman神经网络(ODCGWO-Elman)软测量模型,用于预测故障时间。该模型以Elman网络作为基础结构,将灰狼算法(GWO)与Elman网络结合,以克服网络因随机选取的权值和阈值不恰当对预测精度的影响,提高模型的学习能力和泛化性。同时,在灰狼算法中引入反向学习策略、基于余弦函数自适应的收敛因子和正则函数对其进行改进,提高灰狼算法的求解精度。实验结果表明:ODCGWO-Elman预测准确率为94.8%,相较于GWO-Elman和传统Elman网络,预测准确率分别提升了11.5%和22.3%,有效提升了设备故障预测的准确性。Due to the complexity and diversity of the equipment itself and the limitation of sensor application range and cost,some fault detection information cannot be obtained completely and accurately.To solve the above problems,a soft measurement model based on the improved grey wolf algorithm and Elman neural network(ODCGWO-Elman)was proposed to predict the fault time.In this model,Elman network was used as the infrastructure,and the grey wolf algorithm(GWO)was combined with Elman network to overcome the influence of the network on the prediction accuracy due to the inappropriate randomly selected weights and thresholds,and to improve the learning ability and generalizability of the model.At the same time,the inverse learning strategy,convergence factor based on cosine function adaptation and regular function were introduced into the grey wolf algorithm to improve the algorithm and improve the solution accuracy of the grey wolf algorithm.The experimental results show that the prediction accuracy of ODCGWO-Elman is 94.8%,which is 11.5%and 22.3%higher than the prediction accuracy of GWO-Elman and traditional Elman network,respectively,and the accuracy of the equipment fault prediction is effectively improved.Due to the complexity and diversity of the equipment itself and the limitation of sensor application range and cost,some fault detection information cannot be obtained completely and accurately.To solve the above problems,an Elman neural network(ODCGWO-Elman)soft measurement model based on the improved grey wolf algorithm was proposed to predict the fault time.In this model,Elman network was used as the infrastructure,and the grey wolf algorithm(GWO)was combined with Elman network to overcome the influence of the network on the prediction accuracy due to the inappropriate randomly selected weights and thresholds,and to improve the learning ability and generalizability of the model.At the same time,the inverse learning strategy,convergence factor based on cosine function adaptation and regular function were introduce

关 键 词：软测量 故障预测 ELMAN神经网络 灰狼算法 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置] TP391[自动化与计算机技术—控制科学与工程]