Prediction of Ship Structure Response Signal Based on Bo-LSTM Neural Network

Ship structures are inevitably damaged by frequent wave impact during long-term navigation in harsh ocean environment. In order to detect hull potential damage in a timely manner, ship structure monitoring technology is applied. However, current monitoring methods lack the ability to predict realtime structural signal fluctuations, making it difficult to identify sensor damage under the interference of noise signals. Therefore, a signal prediction method based on artificial neural network is developed for ship structure response signals. The prediction method is based on a Bayesian-optimized long short-term memory neural network (Bo-LSTM) and improves the accuracy of ship structure response signal prediction by finding the best hyperparameters in the long short-term neural network automatically. Some comparative analyses are conducted among the random forest neural network, recurrent neural network, and Bo-LSTM model. The Bo-LSTM model is finally demonstrated to have advantages in predicting ship structure response signals. Therefore, this Bo-LSTM method is suitable for ship structure damage identification and risk warning in ship monitoring systems.