A multiple SVR modeling of hot rolling process combined with kernel clustering and grey relational grade

In view of the complexity of the industrial process, the robustness and the generalization capability are two important criteria to evaluate a model. Aimed to solve the prediction problem of mechanical property in steel hot rolling process, we proposed a new multiple support vector regression (SVR) models approach by combining modified kernel k-means clustering with grey relational grade. In this paper, a whole training sample data set is partitioned into several subsets by using modified kernel k-means clustering algorithm, and the individual support vector regression is trained by each subset to construct the sub-model respectively. An evaluation strategy of the prediction performance of sub-model with sliding time window is further proposed for improving the prediction performance and adaptive ability of model. In order to correct the model, the grey relational grades are used for combining the outputs of multiple sub-models to obtain the final result, which are gained from relationship between a new input sample data and each cluster center. Simulation results in actual application demonstrate that this model has better generalization and prediction accuracy than the other three models.