[4Yin2] Interactive session 2

[4Yin2-24] Spatio-temporal forecasting for railway track degradation detection with exogenous data

〇Katsuya Kosukegawa¹, Kazuhiko Kawamoto¹, Yasukuni Mori¹, Hiroki Suyari¹ (1.Chiba University)

Keywords:Spatio-temporal forecasting, Railroad maintenance, Deep learning

Track degradation is a deformation of railway track. The degradation rate depends on several external factors, including categorical spatial data such as under-track structures and spatio-temporal binary data such as maintenance work records. With these external factors, we propose a spatio-temporal forecasting method for Shinkansen track degradation using convolutional long short-term memory. In the proposed model, we add an embedding layer to acquire feature vectors from the categorical and binary exogenous data. According to the knowledge of experts, we use additional exogenous data such as precipitation, age of equipment, and unit passage of trains, which were not used in the previous work. With real data from track inspection cars, we examine the effect of the added exogenous data to the forecasting performance in terms of root mean square error.