关键词: 船舶碳排放, 时空预测, 航行状态辨识, 深度预测学习, ConvLSTM

Abstract:     Maritime transport is the backbone of international trade and global economy, but ocean-going ships are also causing increasing emissions problems. With the advent of the big data era, ship emission governance has also entered a new stage. The traditional measurement and prediction methods of ship carbon emissions have a number of challenges, including poor data quality, single dimension, limited precision and rough resolution. As a result, it is impossible to establish an adequate picture of the actual emissions produced by ships, and it is also difficult to predict the evolution trend of emissions in time and space dimensions, posing significant difficulties in the governance of carbon emissions from ships. At the moment, the International Maritime Organization (IMO) and China attach great importance to the decarbonization of ships, and have successively proposed and restated the 2050 shipping decarbonization target, as well as the "dual carbon" goal. The decarbonization of ships is becoming increasingly severe, and it is urgent to further accelerate the decarbonization process of ships.Based on the above background, in order to further improve China′s carbon emission measurement system and prediction theory, as well as increase the accuracy of ship carbon emission measurement and prediction, this study relies on the massive shipping big data resources, employs big data analysis technology and artificial intelligence theory, and conducts an in-depth research on the identification of navigation status in ship carbon emission measurement and the issue of ignoring spatial correlation in ship carbon emission prediction. The aim is to propose a novel navigation status identification method and a ship carbon emission prediction framework in the context of big data.The innovations of this study are mainly reflected in the following aspects: firstly, the traditional navigation status classification method only takes into account two factors: ship speed over ground and main engine load, and the error rate is considerable. In order to improve the accuracy of navigation status identification, a novel spatiotemporal trajectory probe algorithm is proposed based on a thorough examination of multidimensional motion characteristics, such as the duration of navigation statuses, the geographical coverage of ship activities, and the heading changes. It considers a trajectory segment as the base unit and can achieve high accuracy of navigation status identification. Secondly, existing ship carbon emission prediction methods mainly adopt time series data, which means they can only predict the time evolution pattern of carbon emission, but cannot reflect the spatial correlation. In order to realize multiple prediction of carbon emissions in space-time dimensions, a novel spatiotemporal prediction model for ship carbon emissions is proposed in this study, which applies the ConvLSTM method in the field of deep predictive learning. A spatiotemporal dataset creating approach based on GIS identity analysis is also provided. The model can not only predict the temporal evolution of ship carbon emissions, but also properly account for the spatial dependency of carbon emissions using convolution operations.Taking China′s Bohai Sea as an example, a verification experiment for the proposed novel navigation status identification algorithm and ship carbon emission prediction framework is designed in this study. The results show that, firstly, after incorporating more ship motion features, the spatiotemporal trajectory probe algorithm proposed in this study can accurately identify various navigation status with an accuracy rate of more than 90% when compared to traditional algorithms, which effectively improves the accuracy of ship carbon emission measurement. Secondly, the ship carbon emission prediction framework based on the ConvLSTM model combines the advantages of convolutional neural networks (CNN) in processing spatial relationships and recurrent neural networks (RNN) in dealing with time series problems, which and can converge stably on spatio-temporal slice datasets constructed based on GIS identity analysis. Furthermore, it can accurately predict the hotspots of ship emissions, enriching the theory of ship carbon emission prediction driven by deep learning technology.Based on the "dual carbon" goal of China, this study employs the emerging big data analysis technology and artificial intelligence algorithms to improve the carbon emission measurement system of ships, providing a new solution for the prediction of spatial and temporal multidimensional ship carbon emissions, which aids in the exploration of the characteristics of spatial and temporal distribution of carbon emission and the law of spatial evolution, and provides a theoretical basis for the formulation and adjustment of decarbonization policy.

Key words: ship carbon emission, spatiotemporal prediction, navigation status identification, deep predictive learning, ConvLSTM