Journal of Auto-vehicle Safety Association ISSN:2005-9396(Print) 자동차안전학회지

With the expansion of the national commuting zone, the number of KTX and SRT passengers has steadily increased each year, reaching a record high of approximately 119 million in 2025. Although railway reservation, payment, and inquiry services have significantly improved, little progress has been made in responding to train delays and enhancing passenger convenience. According to the Korail Customer Service Charter standard (KTX delays exceeding 5 minutes), the number of KTX delay incidents increased from 21,661 cases in 2022 to 25,690 cases in 2023, while the delay rate rose from 17.6% to 20.8%. Frequent delays are problematic in themselves, but the inadequacy of the train delay information system further aggravates passenger inconvenience. Differences in displayed delay times between station electronic boards and mobile applications, along with inaccurate and continuously changing arrival time information, increase confusion among passengers. To address these issues, this study proposes a railway arrival time prediction model that combines high-precision GNSS correction technology with a time-series deep learning model based on Long Short-Term Memory (LSTM). The proposed model eliminates subjective judgments by train operators and provides deterministic, distance-based arrival time information with second-level precision. This enables passengers to seek alternative transportation options or prepare appropriate responses based on expected waiting times. Ultimately, the proposed system is expected to contribute to reducing the chain of passenger complaints caused by train delays. In addition, the predicted segment travel time is applied to trackside worker safety management. By combining a worker’s location with the position of an approaching train, the remaining time until the train passes the worker’s location is estimated, and a mobile application that issues staged train-approach alerts based on this estimate was implemented.