Regional ship collision risk prediction: An approach based on encoder-decoder LSTM neural network model

Ship collision risk prediction is vital for maritime traffic surveillance, which determines if there is a sustainable and efficient development of the shipping industry. In order to solve the problem of ship collision risk prediction, this study proposes a regional ship collision risk prediction model based on the Encoder-decoder LSTM neural network. Firstly, the regional ship collision risk is quantified by aggregation density-based method, and the spatial-temporal collision risk matrix is synthesized by grid processing method. Then, a key aspect of the model involves employing the CNN encoder to capture spatial risk features and compress the collision risk structure scale. The LSTM neural network as the middle layer of the model is adopted to predict the risk with spatialtemporal characteristics. Finally, the CNN decoder processes the prediction results by deconvolution, which restore the original dimension of data. In our study, we conducted numerous experiments using AIS data from Zhoushan Port of Ningbo. The results show that the proposed model has high accuracy in predicting the spatialtemporal collision risk, which the accuracy rate reached 97.9%. The proposed approach of regional ship collision prediction is efficient and accurate, which can provide decision support for intelligent maritime traffic surveillance.