A new method of fault prediction based on health analysis was proposed for the problem of the existing fault prediction technology could not response the declining trend of system property as whole. Firstly, in order to achieve multi-step prediction, multiple output Support Vector Machine (SVM) was formatted on the basis of support vector machine regression algorithm, while using the Harmony Search-Ant Colony Algorithm (HSACA) to optimize parameters of SVM to solve the local optimal problem. Then nonlinear mapping Harmony Search-Ant Colony Algorithm-Support Vector Machine (HSACA-SVM) model matching monitoring data and health degree was built with the optimal parameters. Finally, the proposed model was used to evaluate a power supply system. The results indicate that the HSACA-SVM model can predict the downward trend of health degree with 97% accuracy, and then realize fault prediction.