A multi-task analysis and modelling paradigm using LSTM for multi-source monitoring data of inland vessels

The vessel monitoring data provide important information for people to understand the vessel dynamic status in real time and make appropriate decisions in vessel management and operations. However, some of the essential data may be incomplete or unavailable. In order to recover or predict the missing information and best exploit the vessels monitoring data, this paper combines statistical analysis, data mining and neural network methods to propose a multi-task analysis and modelling framework for multi-source monitoring data of inland vessels. Specifically, an advanced neural network, Long Short-Term Memory (LSTM) was tailored and employed to tackle three important tasks, including vessel trajectory repair, engine speed modelling and fuel consumption prediction. The developed models have been validated using the real-life vessel monitoring data and shown to outperform some other widely used modelling methods. In addition, statistics and data technologies were employed for data extraction, classification and cleaning, and an algorithm was designed for identification of the vessel navigational state