基于机器学习的0Cr17Ni4Cu4Nb不锈钢流变应力预测研究

Research on prediction accuracy of the flow stress of 0Cr17Ni4Cu4Nb stainless steel based on machine learning

以0Cr17Ni4Cu4Nb不锈钢为例,提出一种基于粒子群优化BP神经网络预测流变应力的新模型。以常温下的准静态(0.001 s^(−1))压缩试验数据、四种温度(25、350、500、300℃)和六种应变率(750、1500、2000、2600、3500、4500 s^(−1))的冲击试验数据为基础,构建了0Cr17Ni4Cu4Nb不锈钢流变应力的随机森林预测模型、粒子群优化随机森林预测模型、Back Propagation(BP)神经网络预测模型以及粒子群优化BP神经网络预测模型,采用统计学的决定系数(R2)、平均绝对误差(MAE)、均方差(MSE)和均方误差平方根(RMSE)四个指标分析评价上述四种模型,得出四种模型预测的综合性能依次是粒子群优化BP神经网络模型、BP神经网络模型、粒子群优化随机森林模型、随机森林模型。粒子群优化BP神经网络模型决定系数R2=0.9997、平均绝对误差MAE=1.5773、均方差MSE=5.5053和均方误差平方根RMSE=2.3463,该模型能够很好预测0Cr17Ni4Cu4Nb不锈钢流变应力。

A new model for predicting rheological stresses based on particle swarm optimization BP neural network is proposed for 0Cr17Ni4Cu4Nb stainless steel as an example.Based on the quasi-static(0.001 s^(−1))compression testing data at room temperature and the impact testing data at four temperatures(25,350,500 and 300℃)and six strain rates(750,1500,2000,2600,3500 and 4500 s^(−1)),a random forest prediction model for rheological stress of 0Cr17Ni4Cu4Nb stainless steel,a Particle Swarm Optimized Random Forest prediction model,a Back Propagation(BP)neural network,and a Particle Swarm Optimized BP neural network are constructed.Four indicators including the statistical coefficient of determination(R2),mean absolute error(MAE),mean square error(MSE)and root mean square error(RMSE)are used to analyze and evaluate the four models mentioned above.The comprehensive performance of the prediction models is in sequence of particle swarm optimization BP neural network model,BP neural network model,particle swarm optimization random forest model,and the random forest model.The coefficient of determination R^(2)=0.9997,mean absolute error MAE=1.5773,mean squared error MSE=5.5053 and root mean squared error RMSE=2.3463 are determined for the particle swarm optimized BP neural network model,which can predict the rheological stress of 0Cr17Ni4Cu4Nb stainless steel very well.