Joint Tugboat and Dock Pilot Capacity Prediction

Abstract

Nautical chain scheduling is a major point-of-action for the deep-sea port authorities, as mismanagement and lack of long-term information about events happening in the port area incur delays and extra costs. The traditional way of working, where multiple stakeholders involved with own planning mechanisms and limited data sharing, shows inefficacy to meet upcoming cargo flow's demand. In this work, we propose to tackle the long-term information deficiency problem on the joint tugboat and dock pilot assignments. Our contributions are three-fold: (1) extensive data analysis on large-scale tugboat and dock pilot related data from a deep-sea port, (2) creation of a time series dataset for tugboat and dock pilot capacity prediction, and (3) benchmarking of various time series prediction methods on our dataset including LSTM, Transformers and NLinear models. As we aimed to predict the capacity of both tugboats and dock pilots on a longer horizon, we position our work in the domain of Long-Term Time Series Forecasting (LTTF) which is a specific case of Time Series Forecasting. By using NLinear on our joint tugboat and dock pilot capacity dataset, we achieve an average of close to 60% accuracy on capacity prediction on tugboat categories and dock pilots over a 12-h horizon.