基于机器学习的热轧轧制力预测  

Prediction of rolling force in hot rolling based on machine learning

作　　者：刘明华[1] 张强 刘英华 王文礼[1] Liu Minghua;Zhang Qiang;Liu Yinghua;Wang Wenli(School of Metallurgical Engineering,Xi'an University of Architecture and Technology,Xi'an 710055,China)

机构地区：[1]西安建筑科技大学冶金工程学院,陕西西安710055

出　　处：《锻压技术》2021年第10期233-241,共9页Forging & Stamping Technology

摘　　要：针对传统轧制力数学模型预测精度差的问题,基于板带轧制数据和支持向量回归(SVR)模型对轧制力进行预测。采用主成分分析(PCA)技术来降低输入变量的维数,同时提出了改进粒子群优化(IPSO)算法调节惯性权值和加速因子,并采用IPSO算法对SVR模型中的惩罚因子c、核函数参数g和不敏感损失参数ε进行优化,最终建立PCA-IPSO-SVR轧制力预测模型。与PCA-PSO-SVR、PSO-SVR和Grid-SVR模型相比,PCA-IPSO-SVR模型的3种误差指标处于最低水平,且平均绝对百分比误差(MAPE)为4.8153%。仿真结果表明:与常规PSO算法相比,IPSO算法可以避免陷入局部极小值,从而获得模型最优参数和提高模型预测精度;与其他3种模型相比,PCA-IPSO-SVR模型具有较高的预测精度和较好的泛化性能。In view of the poor prediction accuracy for the traditional rolling force mathematical model,the rolling force was predicted based on the strip rolling data and the support vector regression(SVR)model.Principal component analysis(PCA)technology was employed to reduce the dimension of input variables,and an improved particle swarm optimization(IPSO)algorithm was proposed to regulate the inertia weight and acceleration factors.The penalty factor c,kernel function parameter g and insensitive loss parameterεof the SVR model were optimized by the IPSO algorithm.Finally,PCA-IPSO-SVR rolling force prediction model was established.Compared with the PCA-PSO-SVR,PSO-SVR and Grid-SVR models,the three error indexes of the PCA-IPSO-SVR model were at the lowest level,and the average absolute percentage error(MAPE)value was 4.8153%.The simulation results show that compared with the conventional PSO algorithm,the IPSO algorithm can avoid falling into the local minimums,thereby obtaining the optimal parameters of the model and improving the prediction accuracy of the model.Compared with the other three models,the PCA-IPSO-SVR model has higher prediction accuracy and better generalization performance.

关 键 词：支持向量回归 改进的粒子群算法 主成分分析 轧制力模型 机器学习 

分 类 号：TF16[冶金工程—冶金物理化学]