Prediction for Working Performance of Air-and-screen Cleaning Unit Based on the epsilon-SVR Method

On the basis of analyzing disadvantages of conventional prediction model of air-and-screen cleaning device, a new regression model based on support vector machine was proposed to predict and control of cleaning process precisely. Parameters of ε-SVR models were determined utilizing non-heuristic Grid Search, heuristic GA and PSO which could avoid the choice of randomness. The effect of samples in different on prediction performance of ε-SVR was analyzed compared with BP. The results indicate that the prediction property of ε-SVR is better than BP especially in small-sample.