Tank vehicle stopping point type identification method for gasoline road transport

Objective Enhance the safety control of tank vehicles during gasoline transport by classifying and processing information on stopping points. Method This study proposed a method for determining vehicle stopping points threshold, based on the gasoline transport trajectory data. On the basis of calibrating stopping point types, a stopping behavior feature set was constructed, which included three subsets, i.e., stopping point features, point of interest features, and stopping cluster features. Support vector machine and random forest models were established. The grid search algorithm was applied to determine the optimal parameter combination. The stopping behavior feature set was taken as the independent variable. The stopping types were taken as the dependent variables. The vehicle returning points, loading points, unloading points, and other points were distinguished. Four kinds of classification evaluation indexes were compared to determine the optimal classification model by comparing the model' prediction performances. Result The stopping behavior feature set containing three subsets could effectively identify the types of stopping points. Random forest model outperforms support vector machine model in terms of classification performance, with an accuracy over 90%. The most influential features on classification results among three subsets are stopping duration, distance to gas station, and total stopping duration within cluster. Among all features, the most influential feature on classification result is stopping duration. The introduction of two features also contributes significantly to the classification, i.e., whether it is the first or last point, and vehicle entropy within cluster. Conclusion The study result could be used to determine the loading status of vehicles in transit, analyze the vehicle travel characteristics, and provide support for the safety supervision in hazardous goods transportation.