| Water transport is an important mode of transportation,carrying 90%of China’s foreign trade transport and nearly 16%of the volume of domestic transport.In recent years,the rapid development of the waterway transport industry,the number of ships has increased significantly,the increasing density of maritime shipping,despite the shipping industry has done a lot of work on the safety of ship navigation,water traffic accidents still occur frequently.This caused casualties,environmental pollution and economic losses and other consequences are unimaginable.In order to strengthen ship shipping safety management,it is necessary to conduct quantitative risk assessment of ships.Based on the research of domestic and foreign scholars on ship navigation safety risk assessment,this thesis analyze the causes of ship risk from multi-dimension and construct a ship risk assessment model.The data in this thesis mainly come from Ningbo Chuangbao Technology Co.,Ltd.Through data cleaning technology to remove duplicate and irrelevant data,missing value processing,coordinate conversion and normalization processing,a total of 8365 complete ship data are obtained.The pre-processed data were combined with features,and the input features were filtered by wraparound-based feature selection method,and 150 features were retained after several iterations to eliminate irrelevant and redundant variables.The data set was divided into training and testing sets according to 7:3,and three machine learning algorithms,namely Random Forest,XGBoost and Light GBM,were used to construct the ship risk assessment models,and the effectiveness of the models was evaluated by MAE,RMSE and R~2.After comparison,it is found that the Light GBM evaluation model is better than the other two models.Moreover,the setting of hyperparameters will have some influence on the performance of the assessment model.In order to improve the performance of the Light GBM evaluation model,this thesis automatically tunes the Light GBM model based on the grid search algorithm and the TPE algorithm,respectively.Both auto-tuning algorithms outperform the empirical method,especially the TPE algorithm,a heuristic algorithm with a strategy,constructs a probabilistic model based on the previous results and predicts the performance of hyperparameters to decide the settings of a new set of hyperparameters,reducing the number of iterations and speeding up the search speed.Therefore,a five-fold cross-validation based on the TPE-Light GBM model is performed to improve the reliability of the model.Finally,the MAE of the model is 0.0173,the RMSE is0.0771,and the R~2 is 0.9884.The experiments prove that the model has a good practical application for ship risk assessment studies. |
