Research On Ship Motion Attitude Prediction Method Based On Bidirectional LSTM

Ships are easy to produce nonlinear 6-DOF random complex motion under the influence of many factors,which has a great impact on the safety of ships and their crew,the efficiency and safety of maritime operations,especially the safe takeoff and landing of carrier-borne aircraft on aircraft carriers.Therefore,the accurate prediction of ship's motion attitude is of practical value for the effective adjustment and control of ship's motion attitude and the timely adjustment of carrier-borne aircraft's takeoff and landing.On the basis of the existing research,this paper proposes a combined prediction method based on the improved particle swarm optimization algorithm(ADPSO)and Bidirectional Long Short-Term Memory(bidirectional LSTM,BiLSTM)neural network,aiming at the characteristics that ship motion nonlinearity is difficult to predict accurately.Firstly,the classical method of ship motion attitude prediction is studied in this paper.According to the characteristics of ship motion data as a set of nonlinear time series,the recurrent neural network is used to predict ship motion attitude.On the basis of studying the structure principle and application of recurrent neural network,BiLSTM neural network is selected to predict the ship motion attitude,and an appropriate recurrent neural network model is built through experiments for prediction simulationThen,in view of the shortcomings of the random setting of the initial structural hyperparameters of the neural network,particle swarm optimization(PSO)algorithm is introduced to find the optimal structural hyperparameters of neural network,thus saving the workload and time of adjusting parameters,and improving the prediction accuracy of the model.It is found that the original particle swarm optimization algorithm is prone to local extremum.After studying the improvement strategy of the particle swarm optimization algorithm,an adaptive dynamic particle swarm optimization algorithm(ADPSO)is proposed based on the parameter adjustment strategy of the particle swarm optimization algorithm Through the optimization iteration experiments of several common test functions,it is verified that ADPSO algorithm can effectively avoid the problem that PSO algorithm is easy to fall into the local extreme value,and can quickly find the global extreme value and has good optimization performanceFinally,the ADPSO algorithm is applied to find the optimal number of hidden layer nodes of BiLSTM neural network,and the ADPSO-BiLSTM neural network model is obtained,and the model is used to predict the ship's motion attitude.In addition,the algorithm and model proposed in this paper are verified in the same experimental environment,and the measured ship motion data are used in the experiment.The experiment compares the simple recurrent neural network model with the algorithm optimized neural network model,and compares and analyzes the iterative change curve of training loss function,the prediction result curve,the prediction error curve and evaluation index respectively.Experiments results show that the prediction method based on BiLSTM neural network and ADPSO algorithm proposed in this paper is better fitted to the real motion curve and has better prediction performance.