Abstract: Aiming at the problem of SVDD parameter optimization in training process,an finite exhaustion-local genetic algorithm is proposed.Based on the analysis of C and σ parameters' different influence on SVDD classfication performance respectively,a conclusion that σ is the main influencing factor is made.Focus on σ optimization issue,the near-optimum solution of σ is determined by enumerating finite integer solutions along with comparing these solutions' performance,and then search is launched locally in nearby domain of near-optimum solution.The precise parameter is get at last.Simulation experiment and circuit fault detection application results show that the above algorithm avoids blind parameter searching and can approach optimum solution in shorter time wastes.