Parameter Optimization for SVR Based on Genetic Algorithm and Simplex Method

Support vector machine is a machine learning method which is based on structural risk minimization principle. The traditional parameter optimization methods of support vector regression mainly employ grid search method and so on. These methods have shortcomings of being guided by human experience and time-consuming. In recent years, many intelligent search algorithms are used for SVR parameter optimization problem, which show good results. Simplex is a direct search algorithm for solving unconstrained nonlinear programming problems. To avoid precocity and poor local searching ability of genetic algorithm, a new parameter selection method based on hybrid genetic algorithm is proposed which adopts the results of GA to initialize simplex method, combining with the local search ability of simplex. Simulation results show that the proposed algorithm has better searching efficiency than traditional GA and SVR prediction accuracy have better performance than traditional GA, which proved the effectiveness of the proposed method.