Rolling force prediction based on multiple support vector machines

Accurate rolling force setting is very important for hot strip rolling, but it is difficult to obtain accurate mathematical models for it. A rolling force prediction method based on multiple support machines is proposed in this paper. In order to classify the sample data, the input space of the model is divided into several subspaces utilizing the subtractive clustering method firstly, and several sub support vector machine models are established according to the number of the subspace. The sub models are trained using the actual sampled data, then the output of the sub models are synthesized utilizing the principle component analysis method. Experiment results show that the proposed method can achieve promising performance. The prediction average error rate decreases from 8.19% by BP-NN to 3.76% by the proposed method.