Digital twin for decision making to manage port operation with uncertainties: a knowledge approach based on machine learning

Abstract

With global port traffic having quadrupled in 20 years from 200 million to 800 million containers, controlling the performance of sea freight is proving to be crucial for world trade but delicate. The port, or container terminal, is the basic unit of the global maritime freight network and the hub of interactions where the impact of uncertainties is accrued. The United Nations Conference on Trade and Development has highlighted the diversity of these uncertainties in 2020. The sooner the impact is quantified, the better the reaction. Thus, the work carried out aims at predicting as soon as possible the impact of hazards on the respect of initial objectives. A state of the art on port resource planning has shown the difficulty in formalizing the relationship between the duration of operations and uncertainties. Faced with these limitations, the developed approach based on knowledge engineering proposes first of all an approach to build a digital twin of a container terminal. This digital twin is then exploited to build a prediction model of LSTM time series. The first set of experiments shows that the proposed inference is applicable for learning and predicting port operations. The second set of experiments shows the use of a multi-step LSTM time series prediction model. With periodically renewed predictions, the operations manager of a container terminal will continuously visualize the evolution of the operations schedule, including possible deviations between the initially planned end date and the predicted one based on the actual data at the considered time.