Multi-dimensional prediction method based on Bi-LSTMC for ship roll

Abstract When ships sail in the sea, they will move irregularly due to the influence of strong wind, waves and other complex marine environment. Among them, the ship roll is very important to the safety of ship navigation. In order to ensure navigation safety, it is necessary to predict the ship roll effectively and accurately in advance to provide a basis for ship advance control. Aiming at the problem that traditional methods and machine learning methods are not accurate for ship roll prediction, a single input single output (SISO) and multiple input single output (MISO) ship roll prediction methods based on deep learning are proposed, and the influence of input variables on the ship roll prediction model is studied. Firstly, a single input Bi-LSTMC ship roll prediction method is proposed. The network takes the advantage of LSTM time series prediction and combines convolution kernel to extract cross time features. Then the analysis and test results of six different single input prediction models are given based on real ship data. Secondly, a deep bidirectional feature network with multiple inputs of ship roll angle, roll angular velocity, relative wind speed, relative wind direction, turning angle and rudder angle is proposed for ship roll prediction. The network uses Bi-LSTM to extract forward and reverse information respectively, and introduces two Bi-LSTMC branch structures to explore the deep features between multiple input data to improve the accuracy of ship roll prediction. Finally, four error indexes were used to evaluate the proposed single input and multiple input ship roll prediction algorithms on real ship data. The applicable conditions of the single input and multiple input ship roll prediction algorithms are obtained, and the effectiveness of the proposed algorithms are also verified.