Ship Flooding Time Prediction Based on Composite Neural Network

When a ship sailing on the sea encounters flooding events, quickly predicting the flooding time of the compartments in the damaged area is beneficial to making evacuation decisions and reducing losses. At present, decision-makers obtain flooding data through various sensors arranged on board to predict the time of compartment flooding. These data help with the calculation of the flooding time in emergency situations. This paper proposes a new approach to obtaining the compartment flooding time. Specifically in damage scenarios, based on Convolutional Neural Network and Recurrent Neural Network (CNN-RNN), using a composite neural network framework estimates the time when the compartment's flooding water reaches the target height. The input of the neural network is the flooding images of the damaged compartment. Transfer learning is utilized in the paper. The ResNet18 model in Pytorch is used to extract the spatial information from the flooding images. The Long Short-Term Memory (LSTM) model is then applied to predict when the compartment flooding water reaches the target height. Experimental results show that, for the damaged compartment, the flooding time predicted by the neural network is 85% accurate while the others' accuracy is more than 91%. Intuitively, when it comes to the actual flooding event, the composite neural network's average prediction error for compartment flooding time is approximately 1 min. To summarize, these results suggest that the composite neural network proposed above can provide flooding information to assist decision-makers in emergency situations.