Rolling Force Prediction Based On Least Square Support Vector Machine

In the process of hot strip rolling production, the accuracy of the rolling force prediction has directly an effect on product quality, including shape precision, dimensional accuracy and so on. The reduction of rolling mill depends mainly on the value of the rolling force prediction to reasonably allocate, so the prediction precision of the finish rolling force model largely determines the precision of strip head and how to accurately set the roll gap, and has direct influence on the bite stability. The traditional mathematical model (TMM) is simple in structure and low in accuracy. Because it is limited by the structure of the model, it is difficult to obtain the accurate approximation, even though applying the model self-learning. Therefore, we need to continue to explore and establish a new rolling force model, in order to further improve the accuracy of the rolling force prediction.The rolling force prediction model based on least square support vector machine (LSSVM) is established in this paper. It uses a mixed kernel function linearly combining RBF kernel function with polynomial kernel function to improve generalization ability of the model, and optimize model parameters by the particle swarm optimization algorithm. In order to further improve the prediction accuracy and facilitate on-line application of the LSSVM rolling force model, a combined model which combined least square support vector machine and traditional mathematical model is built. By feature selection and extraction, and collecting field data, a model database of Oracle training is established. A simulation study of rolling force prediction model, including traditional mathematical model, BP neural network model, least square support vector machine model and the combined model, has been done by writing programs in Matlab2012. Finally, the rolling force prediction model is added to the process control system of rolling mill. The rolling force on-line prediction system and the off-line model training tool is obtained by writing program in VC2010, so the on-line application of rolling force prediction is used. The result shows that the BPNN model is better than the TMM model, but VC dimension and complexity of the BPNN model will increase when BPNN model improves precision of prediction, so the generalization ability of the BPNN model is reduced. LSSVM model is better than BPNN and TMM model, and the combined model is better than the three single models, which has a great improvement on the prediction accuracy of the model. LSSVM has a stronger learning ability and generalization ability, which can greatly improve the precision of rolling force prediction, and has great potential in practical application.