항만 터미널의 컨테이너 트럭 턴타임 예측 시스템 구축

Abstract

In the field of logistics, port congestion can cause a seriously negative impact on the cost and efficiency of port operations, due to increased container processing time. Although numerous logistics companies are trying to make their transportation system more efficient, solving the port congestion problem has seldom been studied. In this paper, we explore methods for turn-time prediction of container trucks for efficient port operations. For the dataset, real-world data containing complex data such as truck license plate number, time, and loading/unloading information accumulated for five years at a port terminal company are used and the turn-time prediction algorithm based-on the LSTM model was constructed. For the implementation of the turn-time prediction algorithm, a given time series data was classified into three types: time, day, and week, and used as the input data for the model. When constructing a prediction algorithm based on the time type, it was found that when the input time interval was 7 hours, the time error was 18.31 minutes, which is about a 27% decrease in the time error compared to the time error of 25.17 minutes at 20 hours, which is the lowest input time interval. In the case of the day type, when the time interval is longer, the higher the prediction accuracy can be obtained. When setting the time interval to 20 days, the time error was the highest at 18.18 minutes and the time error was decreased by 30% compared to the time error of 25.82 minutes at the time interval of the 3-day with the lowest accuracy. For the week type, the time error was the lowest at 32.03 minutes when set to a three-week time interval. On the other hand, when the time interval was set to 7 weeks, the time error was 14.13 minutes, showing the time error reduction of more than 57% and the best performance among the total results. In addition, in order to increase the utilization of the above prediction model, we introduced a system consisting of various components such as data acquisition, processing, and analysis along with a mobile user application.