seyed hesamoddin zegordi

The purpose of this article is to predict the delays of passenger trains in the railway areas of the Islamic Republic of Iran using historical data of passenger trains and weather data with machine learning methods. Forecasting delays in railway areas with weather conditions in that area in the winter season can be effective for decision-making and preventive measures. The data used in this study includes passenger train delay data from 2017 to 2022 and weather data from synoptic stations from 1396 to 1400. which contains 46596 records. Independent variables include year, month, day of the month, day of the week, axis of movement, type of train, railway area, maximum wind speed, minimum horizontal visibility, minimum temperature, maximum temperature, number of frost reports on the ground surface and precipitation. It rains and snows 24 hours a day. The proposed method for solving the existing problem based on CRISP-DM is a superior methodology in the field of implementing machine learning and data mining techniques in research and executive fields. Predictive modeling has been carried out in the form of classification. In order to predict the classification of the delay dependent variable, they are divided into two classes, on time and with delay. Supervised learning methods of the classification type have been used to predict the influence of weather factors on the occurrence of delays in railway areas. To evaluate the prediction results, cross-validation has been used to check the validity of the model. The results show that the influence of weather factors in the winter season during the 5-year period in the year-by-year surveys had a positive, negative or neutral effect on the occurrence of delays in railway areas. At the end of the article, preventive measures are presented to adapt the railway industry to climate threats in the future.

This paper proposes several innovative approaches to model evaluation after obtaining the reinforcement learning model of scheduling with predictive maintenance. To train this model, its reward and loss function must be determined according to the conditions of the workshop environment. One of the innovations of this paper is to provide a definition of the reward function for the issue. This learning model is examined in different modes of work entry into the workshop and the results obtained from other scheduling methods show better outputs. The predictive maintenance model is evaluated by four learning methods and the quality of these models is examined. By selecting and adding the best machine failure model to the scheduling reinforcement learning model, the instant tasks entered into the workshop are assigned to the machines. By comparing the proposed method with the previous ones, the best performance is found and shown.