A New Support Vector Machine Model and Its Application in Time-Varying Signal Classification

Aiming at the problem that conventional methods of support vector machine (SVM) are difficult to solve classification of time-varying signal patterns directly, this paper presents a process support vector machine (PSVM) model. The input of PSVM can be functions with time-varying (or function vector). Through the kernel function transforming, dynamic pattern is mapped into high-dimensional feature space. After learning classification characteristic of the training samples, PSVM can extract process characteristics of time-varying function adaptively and classify time-varying signals directly. Some theoretical problems were proved, such as the equivalence of PSVM´s dynamic pattern classification in function space and SVM´s pattern classification in high-dimensional metric space under a group of orthogonal function basis, the equivalence on two-category ability of PSVM and three-layer feedforward process neural networks, etc. The model of PSVM and its solving algorithm were given. The results of simulation experiments confirmed the efficiency of the model and algorithm.