An EMD-SVR method for non-stationary time series prediction

In the area of prognostics and health management, data-driven methods increasingly show the superiority against model-based method due to the complex relationships and learn trends available in the data captured without the need for specific failure models. This paper uses Empirical Mode Decomposition (EMD) and Support Vector Machine (SVM) to build a model for non-stationary time series prediction. And it proves that the EMD-SVR method can solve the problem of few training samples in modeling the path of performance degradation. Then when the threshold is given, we can forecast the lifetime of engineering systems based on the performance degradation data.