Research on parameter optimization of fault classifier based on support vector machine

It is significant to choose kernel function and its optimization parameters for performance of support vector machine. Aiming at parameter optimization of fault classifier based on support vector machine, the principle of parameter optimization of support vector machine by means of minimizing the radius-margin (RM) upper bound as optimum object was discussed. Then a simplified algorithm was proposed. The algorithm does not require computing the gradient and can optimize one parameter of kernel function by adopting constant iterative step length. Based on the algorithm, parameter optimization of binary fault classifier was implemented. The simplified algorithm is applied to fault classifier which classifies steam oscillation fault and bearing bushing looseness fault of turbo-generator set. Testing results show that the classification capability of the fault classifier can be improved by means of the parameter optimization algorithm.