Research And Optimiza On Soft-Sensor Measurement Model For Aluminum Strip Grain Size Based On Pso-Lssvm

Aluminum electromagnetic casting-rolling technology is a green, efficient aluminum casting technology, has been widely used in recent years. Meanwhile, the aluminum billet grain size is an important evaluation indicators of the quality of aluminum billet. But at the moment due to technical and process limits, there is not an online measurement of the aluminum billet grain size. To address this problem, This paper presents a measurement using soft-sensor technology to predict the aluminum billet grain size. It achieves to measure the aluminum billet grain size on line.This paper detailed analysised the casting factors and electromagnet-tic factors that affect the aluminum billet grain size according to process mechanism of aluminum electromagnetic casting-rolling. Then combined with the actual situation, chose 11 factors as auxiliary variables of the aluminum billet grain size. Proposed of the soft sensor model of the aluminum billet grain size based on LS-SVM. After pretreatment of the raw data, this paper had used the MATLAB tool to simulate the model. First this paper had compared the influences for the accuracy of the model between different kernel functions, then it had compared the influences for the accuracy of the model between regularization parameters, then it had drew the optimal model parameters with the trial and error method. To address this problem of the lack of the parameter selection method, this paper used the PSO algorithm to optimize the super parameters of the model.Simulation results show that the soft sensor model can be established to achieve the online prediction of the aluminum billet grain size. The new model had recuperated the original model's blindness of the selection of the parameters and improved the generalization ability of the model after using the PSO algorithm to optimize the super parameters of the model. This paper had provided a new idea to detect the aluminum billet grain size. It will provide important reference value for the future research.