Route Prediction of tugboats using Deep Reinforcement Learning

Abstract

Efficient Port Operations are essential for minimiming delays and ensuring the safe movement of vessels. Tugboats play a critical role in assisting ships during berthing, unberthing, and navigating port waters. This thesis uses DRL4Route, a deep reinforcement learning (DRL) framework aimed at optimising tugboat routes and pick-up locations. By using historical towage data and adapting to the dynamic conditions of the Port of Rotterdam, DRL4Route provides real-time recommendations that streamline tugboat operations, reduce delays, and improve safety.
The core of the DRL4Route framework lies in its ability to continuously learn and adapt to real-world conditions. Unlike traditional static models, DRL4Route leverages spatiotemporal data to predict optimal tugboat routes. This allows for real-time evaluations that can help the predicted route match closely with the actual route which helps tugboats arrive at the right location at the right times. The framework’s focus on optimising both pick-up and drop-off points helps port operators avoid inefficiencies that can arise from poorly coordinated tugboat movements.
By improving the efficiency of tugboat operations, DRL4Route contributes to a more sustainable and resilient port ecosystem. The system’s adaptability and potential for real-time decision-making make it a strong candidate for future automation of tugboat operations. This thesis highlights how advanced machine learning techniques can enhance the performance of ports like Rotterdam, driving economic benefits and reducing the environmental impact of maritime operations.