基于机器学习的拉矫延伸率预测模型及数值分析

在冷轧弯曲矫直过程中，针对拉矫机工艺参数设置问题，利用经验公式、有限元仿真建立的延伸率模型预测精度不高.为提高预测精度，基于传统解析模型与机器学习算法进行研究，比较了两种方法预测模型的精度，得到机器学习算法的延伸率预测模型要比数值解析模型的拟合优度高.比较BP神经网络算法和支持向量机(SVM)算法，得到两种机器学习算法的预测模型精度基本一致.为进一步提高预测精度，采用Adam算法对BP神经网络进行优化，采用遗传算法对SVM预测模型的参数进行优化，最终得到最优预测模型的均值绝对百分比误差MAPE以及拟合优度R^○2分别为13.4%和0.953，可以为实际生产提供技术指导.

Prediction Model for Elongation of Tension Leveling Based on Machine Learning Algorithm and Numerical Analysis

In the process of cold rolling bending straightening， aiming at the setting of process parameters of tension straightener， the prediction accuracy of elongation models established by empirical formula and finite element simulation is not high. To improve the accuracy， the traditional analytical models and machine learning algorithms are studied. The accuracies of the two methods are compared. It is found that the elongation prediction model of machine learning algorithm has higher goodness of fit (R²) than the numerical analytical model. Comparing BP neural network algorithm with SVM (support vector machine) algorithm， the prediction model accuracies of the two machine learning algorithms are basically the same. In order to further improve the prediction accuracy， the BP neural network is optimized by Adam algorithm， and the parameters of SVM prediction model are optimized by genetic algorithm. Finally， the MAPE(mean absolute percentage error) and R^○2 of the optimal prediction model are 13.4% and 0.953 respectively， which can provide technical guidance for actual production.