Abstract: Original Least Square Support Vector Machine(LSSVM) algorithm can not reach desired precision in multi-scale regression. To solve the problem, a multi-scale wavelet LSSVM algorithm is proposed by using a wavelet kernel. Mexican-hat wavelet function is used as the support vector kernel function, and the Least Square Wavelet Support Vector Machine(LS-WSVM) algorithm is presented. On this basis, the global optimum of the multi-scale regression modeling problem can be obtained by solving a quadratic programming problem. As a result, the regression model can effectively approximate multi-scale signals. Simulation results show that LS-WSVM is an efficient modeling method, and has high precision.

Key words: multi-scale, least square, wavelet kernel, Support Vector Machine(SVM), MARR kernel, regression modeling

摘要： 普通最小二乘支持向量机算法用于多尺度回归建模时精度较低。针对该问题，选取墨西哥草帽小波函数作为最小二乘支持向量机的核函数，设计一种基于小波核的多尺度最小二乘小波支持向量机。在此基础上，通过解二次优化问题求出多尺度回归建模问题的全局最优解，最终得出的多尺度回归模型能够有效地逼近多尺度信号。仿真结果表明，该算法具有较高的精度。

关键词: 多尺度, 最小二乘, 小波核, 支持向量机, MARR核, 回归建模

CLC Number: 

• TP391