LSSVM prediction model based on improved kernel parameter selection method

摘要

With the rapid development of modern science and technology, especially electronic information technology, the application of electronic information technology in communication, aerospace and industrial systems is becoming more and more complex. The intelligence and complexity of the system are constantly improving, and the structure of devices and equipment is becoming more and more complex, and the degree of automation is high. Fault prediction studies the future of equipment, one is to understand the future development trend of equipment health status, prepare for a rainy day in advance. Second, it can make more scientific decisions on equipment management and application through prediction. There are two aspects in the application of fault prediction: one is to predict the unknown data according to the current monitoring data, and infer whether the current equipment or equipment status is normal according to the standard of the unknown data; the other is to estimate the remaining life before the failure may occur in the future according to the previous data experience. For the shortage of electronic circuit fault prediction technology, this paper describes a variety of fault prediction methods (fault prediction based on grey system, fault prediction based on particle filter, fault prediction technology based on support vector machine), and briefly describes their advantages and disadvantages, introduces SVM, nonlinear SVM prediction method and nonlinear LSSVM prediction method in detail. With the further research of LSSVM model, the kernel parameter selection of nonlinear prediction model has gradually become the focus of SVM model application. Based on the existing LSSVM model, this paper proposes a new kernel parameter selection method, which enhances the application ability of LSSVM model. By comparing with the traditional LSSVM model, the effectiveness of the proposed method is proved.