چكيده به لاتين
Railway track maintenance processes must be done according to analysis of TGV (Track Geometry Vehicle) recorded data. Planning and prioritizing inspections and preventive maintenance operations can be considered in designing railway track maintenance systems thanks to prediction of future state of tracks. To this end, in this study, prediction of track geometry irregularities in the Railways of the Islamic Republic of Iran (RAI) is done in a data mining context.
We use four approaches to predict track geometry irregularities: predict each point geometry state, predict average geometry parameters in 10-meter sections of track, predict standard deviation index of geometry parameters in 200-meter sections of track and predict Composite Track Record (CTR) index for each kilometer of track. In each approach we use different types of prediction models including multilayer perceptron, radial basis function network, quick neural network, regression, logistic regression, CHAID decision tree, classification and regression tree, C5 decision tree.
In predicting each point geometry state and predicting average geometry parameters in 10-meter sections of track multilayer perceptron and regression models have the best results in train and test data, respectively. In predicting standard deviation index of geometry parameters of track, CHAID decision tree has the best results in train and test data. In predicting Composite Track Record (CTR) index of track, logistic regression has the best results in train and test data.
