Research On Multi-Output Support Vector Machine And Application

Support vector machine based on statistical learning theory has become a new and generalized method for machine learning. It develops rapidly and has been widely used in recent years. Traditional support vector regression (SVR) algorithm is only for single output system. Several SVR models were evaluated to identify one appropriate for multi-output systems, which has low training speed and poor accuracy, so how to establish multi-output support vector regression algorithm with high training speed and accuracy, and apply it to system modeling and control has very important value in the area of theoretical research and application.In order to improve the training speed of multi-output SVR, this paper is firstly based on sparseness of the solution and establishes multi-output sparse SVR algorithm. Then it proposes a multi-output SVR based on heuristic training algorithm. By introducing the concept of similarity, the training samples whose similarities are bigger than given threshold are removed so as to reduce the number of samples and increase the training speed. Aiming at the simulation example of mathematical function model, these two different SVR algorithms not just guarantee high accuracy of the model, but also increase the training speed compared to regular multi-output support vector regression.Finally, the above method is applied to practical process:the batch polymerization reaction process for methyl methacrylate. By establishing Polymerization multivariable output model and optimizing the model, the optimal temperature control curve is given, which meets the requirement of conversion rate and average molecular weight. By comparing the three methods of simulation modeling, it shows that heuristic multi-output SVR has higher accuracy in modeling with less training time compared to other two methods, so it is more suitable for the actual process applications.