基于灰狼算法优化深度极限学习机的钢轨热处理性能预测模型  

Optimization of rail heat treatment performance prediction model for deep extreme learning machine based on grey wolf algorithm

作　　者：蔡里批 李硕 丁敬国[2] Cai Lipi;Li Shuo;Ding Jingguo(Northeastern University,School of Materials Science and Engineering,Shenyang 110819,China;Northeastern University,State Key Laboratory of Rolling and Automation,Shenyang 110819,China)

机构地区：[1]东北大学材料科学与工程学院,沈阳110819 [2]东北大学轧制技术及连轧自动化国家重点实验室,沈阳110819

出　　处：《材料与冶金学报》2025年第2期162-170,共9页Journal of Materials and Metallurgy

基　　金：2023年辽宁省大学生创新创业训练计划项目(S202310145024)。

摘　　要：为了研究钢轨的化学成分、入口温度、环境温度,以及风冷时风压、风速等参数对热处理钢轨性能的综合影响,进一步解决钢轨热处理后设定精度低的难题,开发了一种基于灰狼算法优化深度极限学习机(grey wolf optimization deep extreme learning machine,GWO-DELM)的钢轨热处理性能预测模型.先采用深度极限学习机(DELM)构建出工艺模型,而后,针对深度极限学习机中初始权值随机确定而引起的预测结果准确度较低的问题,利用灰狼优化算法(GWO)对初始权值进一步确定.结果表明:该模型在预测不同规格钢轨的抗拉强度时,95.80%以上样本点的预测误差集中在-20~20 MPa,在预测踏面布氏硬度时,95.73%以上样本点的预测误差集中在-8~8;与传统模型相比,GWO-DELM具有更优异的预测精度及泛化能力,可应用在热轧钢轨风冷处理的性能预测上,为热处理参数的选择提供参考.In order to study the comprehensive effects of rail chemical composition,inlet temperature,ambient temperature,wind pressure,wind speed,and other parameters on the performance of heat-treated rails,and to further solve the problem of low setting accuracy of rails after heat treatment,a prediction model of rail heat treatment performance based on grey wolf optimization deep extreme learning machine(GWO-DELM)was developed.Firstly,the process model was constructed by using the deep extreme learning machine(DELM),and then the gray wolf optimization algorithm(GWO)was used to further determine the initial weights in order to solve the problem of low accuracy of the prediction results caused by the random determination of the initial weights of the deep extreme learning machine.The results show that when the model predicts the tensile strength of rails of different specifications,the prediction error of more than 95.80%of the sample points is concentrated in-20~20 MPa,and when predicting the Brinell hardness of the tread,more than 95.73%of the prediction errors of the sample points are concentrated in-8~8.Compared with the traditional model,GWO-DELM obviously has better prediction accuracy and generalization ability and can be applied to the performance prediction of air-cooled treatment of hot-rolled rails,providing a reference for the selection of heat treatment parameters.

关 键 词：钢轨热处理 灰狼优化算法 深度极限学习机 性能参数预测 

分 类 号：TG335[金属学及工艺—金属压力加工] TP18[自动化与计算机技术—控制理论与控制工程]