A Trend Prediction Method for Failures Time Series Data by Exploring Singular Spectrum Analysis and Support Vector Machines Regression

Recently, reliability prediction problems have been sufficiently studied by applying various failures time series prediction models. However, these models' performance could easily reach the bottleneck on account of the noise interference of the failures data. This paper focuses on the trend component of failures time series data as we believe its noiseless curve will achieve better prediction results and help to provide scientific guidance for reliability activities. We propose a trend prediction method which mainly involves two steps, trend extraction by singular spectrum analysis (SSA) and noise test, and trend prediction by support vector machines regression (SVR). Moreover, we further develop a multi-layer grid search algorithm to obtain optimized parameters of SSA and SVR models. The performance of the proposed method is measured against other time series processing techniques such as seasonal-trend decomposition procedure based on loess (STL), Holt-Winters, autoregressive moving average (ARMA), multiple linear regression (MLR) and k-nearest neighbors (KNN). The comparison results indicate that the proposed method outperforms other techniques and can be utilized as a promising tool for failures time series trend prediction.