An integrated physical-based and parameter learning method for ship energy prediction under varying operating conditions

The efficiency of energy consumption of an engineering system dynamically changes during the its operation when the operational and environmental conditions vary in time. Various methods have been developed to monitor the energy consumption rate and predict the consumption efficiency for a given operating condition. The main challenges to maintain the accuracy of modeling and prediction stem from the great diversity of operational and environmental inputs that affect the energy consumption rate dynamically, as well as the lack of a full understanding of the physical relationship between energy efficiency and operation parameters of the system. Operating condition is a key component in system modeling and state identification in many applications because not only the system parameters, but also the structure and complexity of a model might vary significantly during different operation modes. This paper investigates a novel method that integrates a physics-based hydrodynamic model and dynamic parameter learning and estimation, using energy consumption monitoring data and operating condition data, in purpose of improving the prediction accuracy of energy consumption. By leveraging the strengths of both the physics-based models and data-driven parameter learning methods, the proposed method is advantageous when the complex system physics is not perfectly known and the performance of system is affected by the environmental operating condition, while abundant monitoring data are available. We demonstrate the model on a ship propulsion system for fuel consumption prediction, which achieves higher prediction accuracy compared with models without operating condition adaption and tuning.