Robust multiwavelets support vector regression network

A new model of support vector regression network with multi-resolution and robust multiwavelets is put forward, combining wavelet network using robust estimation as cost function with wavelet support vector machine. When there are outlines, it can overcome the disadvantage that support vector regression has bad robust performance and solve the problem of the determination of network structure and initial parameters of the wavelet network using robust estimation as cost function. Using the multi-resolution approximation character of wavelet network, the choice of the kernel function of multi-resolution support vector machine can be completed and the approximation precision can be improved. Simulation results show that this model has not only excellent robust performance to outlines, but also better generalization performance and multiscaling character. At the same time, the approximation precision of signals can be improved.