Advanced Track Geometry Forecasting Methods
The Federal Railroad Administration (FRA) contracted ENSCO, Inc., to apply alternative forecasting approaches to foot-by-foot track geometry data. The research team used track geometry data collected weekly by FRA’s DOTX-225 and DOTX-226 autonomous inspection vehicles. This investigation also considered the effect of both seasonality and maintenance activities, which helped enhance the accuracy of the resultant models. The work began in September 2020 and was completed in September 2021.
The research team demonstrated two methods for time series-based forecasting of foot-by-foot track geometry data: 1) point data processing with a SARIMAX model and 2) segment data processing with Facebook’s Prophet model. Both methods produced accurate forecasts. The team concluded that these methods, along with frequent measurements from Autonomous Track Geometry Measurement Systems (ATGMS), can accurately predict future behavior of track geometry and help railroads plan preventive maintenance. Researchers also developed a dashboard to assess the performance of various forecasting models, which graphically showed statistical performance measures for easier interpretation.