Dempster–Shafer regression for multi-step-ahead time-series prediction towards data-driven machinery prognosis

Abstract

Predicting a sequence of future values of a time series using the descriptors observed in the past can be regarded as the stand-stone of data-driven machinery prognosis. The purpose of this paper is to develop a novel data-driven machinery prognosis strategy for industry application. First, the collected time-series degradation features are reconstructed based on the theorem of Takens, among which the reconstruction parameters, delay time and embedding dimension are selected by the C–C method and the false nearest neighbor method, respectively. Next, the Dempster–Shafer regression technique is developed to perform the task of time-series prediction. Moreover, the strategy of iterated multi-step-ahead prediction is discussed to keep track with the rapid variation of time-series signals during the data monitoring process in an industrial plant. The proposed scheme is validated using condition monitoring data of a methane compressor to predict the degradation trend. Experimental results show that the proposed methods have a low error rate; hence, it can be regarded as an effective tool for data-driven machinery prognosis applications.