Statistical estimation and forecast of seasonal variations in measured track geometry

Abstract

This paper describes a method for statistically estimating five components included in the time series of vertical track geometry measured at the same point on a railway track. We have developed a statistical model to represent the change of vertical track geometry. The model decomposes the time series into five components: seasonality, trend, effects of maintenance works, short-term variation, and noise. The vertical geometry in the near future is also forecasted with the components. Two features of the model are as follows: first, the model extracts the effects of seasonal variations. We actually identified track geometry that changes seasonally at some fixed points. The seasonal variations are found to be relevant to rather air temperature than train loads. This finding is contrary to a general belief that track geometry changes monotonically due to daily train loads. Second, the Kalman filter algorithm is applicable to this model. The parameters included in the model are thus estimated effectively according to the maximum likelihood method. In addition, the five components are obtained according to the fixed-interval smoothing algorithm. The goals of this study are to help the scheduling of railway track maintenance with forecasted track geometry and to obtain new information on changes of track geometry.

Keywords

• state space representation
• long-term prediction
• track geometry car
• track geometry recording
• asymmetrical chord offset