Research On Ship Collision Risk Prediction Algorithm Based On LSTM

With the continuous and stable growth of the marine transportation industry,improvements have been made in the number,size,and performance of ships.As a result,the density of ships in key waters and waterways has increased,leading to a rise in issues such as waterway congestion and navigation accidents.It is crucial to efficiently and accurately predict the risk of ship collisions in order to alert ship operators in a timely fashion to avoid collision risks,reduce the incidence of maritime accidents,and enhance the regulatory authorities’ management capabilities.High-quality and efficient prediction of a ship’s track is a fundamental prerequisite for achieving successful ship collision risk prediction.This thesis focuses on ship trajectory extraction and prediction,ship domain construction,and ship collision risk prediction based on ship AIS data.We aim to fulfill practical applications through a combination of theoretical research and experimental analysis,relying on machine and deep learning theories as technical supports.The following research is carried out to achieve this purpose:(1)Aiming at the problems of low generality,low extraction efficiency and low data quality of the current flight path extraction algorithm,the preliminary flight path extraction algorithm and two-stage flight path clustering algorithm are proposed in this paper.The preliminary extraction of track is realized by limiting distance.The forward and backward two-stage Gaussian mixture clustering method is used to realize the reextraction of track.Experiments show that the proposed algorithm reduces the mean value of RMSE and MAPE of forecast results by 10.4% and 12.1% respectively under the same prediction model.The algorithm can effectively extract the track data of 20-40 minutes in the future.(2)Aiming at the problems of poor practical application and low accuracy of existing ship trajectory prediction models in ship collision avoidance,a short-term ship trajectory prediction model based on short and long time memory neural network is proposed,and the model is compared with a variety of common prediction models.The experimental results show that the forecast results of the proposed model are optimal in the next 20-40 minutes.The average RMSE and MAPE of the model are 403 m and 0.62%,respectively,which are 12.6% and 42.5%lower than those of the model based on GRU neural network.The prediction model proposed in this thesis has the characteristics of high precision,high universality and high robustness in predicting ship track.(3)In view of the problem that few studies have been conducted on the time consuming of the algorithm,the research on the time consuming of the algorithm is carried out.The ship track prediction algorithm in this thesis provides services for ship collision avoidance and so on,which has high requirements on the overall time consuming of the algorithm.In order to reduce the overall time of the algorithm and improve the practicability of the algorithm,a gridbased preliminary track extraction algorithm is proposed in this thesis.Experiments show that the time of the grid-based extraction algorithm is more than 70% lower than that of the traditional traversal extraction algorithm.The results show that the average time of the proposed route prediction algorithm based on GMM and LSTM is less than 15 s,which is of high practical value in the short and medium term ship path prediction.(4)To address the issues of short prediction times and insufficient accuracy in predicting ship collision risks,this thesis constructs a fuzzy four-element ship safety field and computes the DCPA and TCPA values of the ships through a prediction model based on LSTM and GMM instead of conventional methods.In addition,the thesis proposes the concept of the most dangerous point in distance and time to enhance prediction accuracy.Results of the experiment demonstrate that the proposed ship collision risk prediction algorithm can realize 94% ship collision risk prediction,with significant practical value in reducing the incidence of marine accidents and ship collision accidents.