Deterioration Prediction of Track Geometry Using Periodic Measurement Data and Incremental Support Vector Regression Model

Abstract

Information on the quality of ballasted track is normally collected from a track measurement vehicle operating on a monthly basis or otherwise periodically. Track deterioration in terms of alignment and vertical levels is normally predicted by time-series data collected up to a certain point, and subsequent maintenance work is undertaken based on the predetermined maintenance level of the track geometry classified according to its importance. In this regard, deterioration of track geometry based on time-series measurement data can be efficiently modeled by an online support vector regression (OSVR) scheme, and detailed investigation has been carried out to improve the previous work on batch-type prediction models of track geometry proposed by the authors. For such purposes, an incremental support vector regression (ISVR) model based on a Bayesian optimization scheme as well as an OSVR model are introduced in this paper, and the prediction results are compared with those obtained by a conventional machine learning model. The results show that the accuracy of the proposed model increases by approximately 20% compared with that of the conventional model, and the outcome can be applied to the optimal scheduling of track maintenance work.