Intelligent Control and Optimization for onboard support of surface vessels

Abstract

The maritime industry is charting a new course towards digitalization and automation, propelled by the pursuit of enhanced safety, efficiency and a more sustainable footprint. Recent breakthroughs in sensor technology have transformed marine vessels into sophisticated sensor hubs, seamlessly linked to global networks via satellites and the Internet. This real-time data connectivity becomes a crucial driver of autonomy and multiple cutting-edge technologies, e.g., digital twin. A decision support system (DSS) is one of the key components of intelligent ships, which relies on computer-based tools to assist decision-makers with reliable and timely information and alternatives. Several kinds of DSS can be distinguished, in which model-based and data-driven DSS are explored in-depth in this dissertation, in particular, the integration of artificial intelligence. Model-based methods are characterized by high explainability and good generalization while their counterpart data-driven methods are better at modelling highly nonlinear behaviour and revealing hidden patterns without extensive prior knowledge. With considerable sensor data generation and storage, data-driven methods have become more meaningful and reliable. The development of onboard decision support tools is mainly concerned with two aspects: (1) situation awareness and (2) decision making. Situation awareness involves a better understanding of the current status, such as vessel state, geographical information, environmental conditions, and operating conditions. Also, it requires predictive insights into future behaviours and potential risks. Various applications are associated with this aspect, including risk assessment, encounter classification, action prediction, and trajectory prediction. Decision making centres on evaluating available options anddetermining the optimal solution based on the current information, constraints and goals, which is closely related to optimization problems. Typical applications are path planning, collision avoidance, route optimization, dynamic positioning and fuel management. In this dissertation, Five applications are studied due to their significance in enhancing maritime efficiency and safety: parameter identification, encounter identification, trajectory prediction, path planning and collision avoidance. Multiple key technologies are leveraged from two distinct principles. Three case studies are highlighted: model predictive control-based collision avoidance, prediction enabled path planning and multi-ship trajectory forecasting. Path planning is an importantfunction of the intelligent decision support system, which aims at the enhancement of safety and efficiency of navigation. In addition, it plays a fundamental role in the development of remotely operated and fully autonomous vessels. The first two case studies address the path planning problem in two distinct manners, reactive and deliberative, and model-based and data-driven, where the former method focuses on the integration of a ship dynamic model and rule-compliant actions into the optimization, while the latter emphasizes a precautionary step enabled by data-driven predictive capai bilities. As an important measure to enhance situation awareness, trajectory prediction is often regarded as a prerequisite to the subsequent tasks including path planning, collision avoidance and optimal control. Interaction modelling is necessary when it comesto multi-ship encounters in order to provide a more accurate prediction. Experiments are conducted with data collected in the simulators or historical data available in public databases. The results demonstrate the effectiveness of the model-based and data-driven methods for decision support of intelligent ships. Furthermore, the data-driven methods exhibit outstanding performance in certain applications.