Prediction of Mechanical Properties of Hot Rolled Strips With Generalized RBFNN and Composite Expectile Regression

The prediction of mechanical properties of hot rolled strips can be used for on-line dynamic control of product properties and optimal design of new steel grade. Tensile strength is an important index of the mechanical properties for hot rolled strips. The influencing factors of tensile strength include alloy elements, microstructure, and production process parameters. It is very crucial to establish a reliable prediction model of steel tensile strength based on these factors to improve the mechanical properties. This paper proposes a prediction model which combines a generalized radial basis function neural network and composite expectile regression to solve nonlinear problems and data heterogeneity problems in modeling. At the same time, CS algorithm (Cuckoo Search) was applied to develop an estimation procedure, which overcomes the shortcoming of traditional gradient descent algorithm by avoiding to fall into local optimum. Because expectile regression can describe the conditional distribution, and neural network has strong non-linear interpretation ability, the proposed model enables us to explore potential nonlinear relationships among variables. Based on the measured data collected from a hot rolling production process, the experimental results show that the model proposed in this paper has better prediction accuracy, the mean absolute percentage error (MAPE) and root mean square error (RMSE) are 2.49