基于机器学习和多目标优化的耐候钢性能预测和逆向设计研究

Study on properties prediction and reverse design of weathering steelbased on machine learning and multi-objective optimization

针对传统耐候钢研发实验复杂、时间长和成本高等问题,提出了一种基于机器学习和多目标优化的耐候钢性能预测和逆向设计模型,同时附加物理冶金学参数指导建模过程,构建了优化后的SVR、RF和GPR模型预测耐候钢的力学性能和耐腐蚀性能,基于正向预测模型,利用SPEA2实现了对屈服强度、抗拉强度、伸长率和相对腐蚀速率的协同优化。结果表明,相较于最优SVR和RF模型,GPR模型与物理冶金学的耦合可实现耐候钢力学性能的高精度预测,显著降低模型过拟合程度,从而提高泛化性,在抗拉强度测试集中有95.19%的样本相对误差在10%以内,最优模型中决定系数达97.67%。针对耐腐蚀性能预测提出的回归和分类模型均表现出较高的预测准确率,GPR模型中有80.4%的样本相对误差在8%以内,GA-SVM模型的平均分类准确率达到80.93%。利用SPEA2可实现对Q700NH高耐候钢的成分和工艺设计,在该成分与工艺体系下,实验钢具备优异的力学性能和耐腐蚀性能,实现了耐候钢低成本高效率的研发设计。

Aiming at the problems of complexity,long time and high cost in traditional research and development experiments of weathering steel,a properties prediction and reverse design model of weathering steel based on machine learning and multi-objective optimization was proposed.At the same time,physical metallurgical parameters were added to guide the modeling process.The optimized support vector regression(SVR),random forest(RF)and Gaussian process regression(GPR)models were constructed to predict the mechanical properties and corrosion resistance properties of weathering steel.Based on the forward prediction model,SPEA2 was used to realize the collaborative optimization of yield strength,tensile strength,elongation and relative corrosion rate.The results show that compared with the optimal SVR and RF models,the coupling of GPR model and physical metallurgy can achieve the high-precision prediction of mechanical properties of weathering steel,and significantly reduce the over-fitting degree of model,thus the generalization can be improved.In the tensile strength test set,the relative error of 95.19%samples is within 10%,and the determination coefficient in the optimal model reaches 97.67%.The regression and classification models proposed for corrosion resistance prediction show high prediction accuracy.The relative error of 80.4%samples in GPR model is within 8%,and the average classification accuracy of the GA-SVM model reaches 80.93%.The composition and process design of Q700NH high weathering steel can be realized using SPEA2.Under this composition and process system,the experimental steel has excellent mechanical properties and corrosion resistance properties,which realizes the low cost and high efficiency research and development design of weathering steel.