Research On Marine Traffic Accident Forecast Based On SARIMA-LSTM

With the rapid development of waterway transportation,the shipping economy as the core industry of the national economy,has been hit and challenges,such as the increasingly serious hidden dangers of maritime traffic and the water transportation situation.Therefore,accurately predicting marine traffic accidents and controlling the trend of accidents in the entire water area is of clear significance for reducing the risk of marine traffic accidents and improving the safety of marine traffic.In this article,the author will combine with the current social environment,on the basis of fully reading domestic and foreign literature,summarize accident prediction methods,introduce domestic and foreign accident prediction research methods and directions,and explain the impact of maritime traffic accident prediction on the development of water transportation.With the rise of computer networks,more and more researchers use deep learning techniques and neural networks to conduct prediction research in various fields.Therefore,in this article,the author will compare the typical methods of predicting marine traffic accidents,clarify the applicable conditions of the model,and establish a scientifically selected marine traffic accident prediction system on this basis.Finally,by clarifying the advantages of the combined prediction method,it is determined to establish a combined prediction model of marine traffic accidents based on the residual optimization model,and verified by examples of marine traffic accident prediction.This paper takes the traditional accident prediction model seasonal autoregressive integrated moving average and the long short-term memory networks as the research objects.The instance data summarized 240 sets of data on maritime traffic accidents in Liaoning jurisdiction from 2000 to 2019,including information such as date of occurrence,collision type,accident type,accident level and other information.After preliminary analysis and analysis,the structure and form of the data are in line with the prerequisites for prediction by the LSTM and SARIMA prediction models.However,given that the time series model is essentially linear prediction,it can only rely on historical data to predict future data trends,and cannot contain a series of non-linear factors,and the prediction accuracy is low;the long short term networks model encompasses the natural complex charm.Therefore,the author considers the idea of complementary advantages,constructs a combined prediction method based on SARIMALSTM neural network residual optimization,selects the root mean square error as an evaluation index to measure the prediction accuracy of the combined prediction model,and compares it with the accuracy of the two single prediction models.Building various prediction models through python and conducting empirical analysis with Liaoning waters data,and obtaining three sets of RMSE values.The combined prediction model based on the residual optimization of SARIMA-LSTM neural network fully combines the advantages of the two,and the evaluation index RMSE reaches 0.15.Based on this evaluation index value,the superiority of the residual optimization combination forecasting model is demonstrated.The marine traffic accident prediction method proposed in this paper can provide guidance for relevant personnel’s decision-making.