Research On Inspection Model Of Freeway Green Channel Vehicle Based On BP Neural Network Ensemble With Bagging

With the implementation of the "green channel" free policy,many truck drivers are lured by the benefits of toll reduction,the vehicles were disguised as green channel vehicle to escape tolls,which caused a huge toll loss on the freeway management.At present,the green channel vehicle inspection mainly relies on human judgment,lacks of green channel vehicle characteristics analysis and early warning as the reference,the inspection is not targeted,timeconsuming and difficult to ensure accurate.The article through the current status analysis of freeway green channel inspection management,proposes the data mining of the historical data of the "fast green channel" management platform,and extracts the data characteristics of cheating green channel vehicle.The integration model of freeway green channel vehicle is built,which is used for auxiliary inspection of green channel vehicle.Firstly,the common classification algorithm and the machine learning method of the unbalanced data set are studied,aiming at the characteristics of green channel vehicle unbalanced data set.The integration model of freeway green channel vehicle is built based on BP neural network ensemble with Bagging.A number of balanced training sets are extracted by an under-sampling method,and each balanced data set is trained as the basis neural network model.Using the Bagging ensemble learning,the base neural network model is integrated into a strong neural network model to increase the generalization ability and stability of the inspection model.Then,through the statistical analysis of the green channel vehicle traffic data,and the indepth analysis of the green channel vehicle cheating behavior,the green channel vehicle inspection model index set is constructed according to the index selection principle.According to the data characteristics of the index attribute,the numerical method of categorical variables,the continuous value discretization method based on ChiMerge,the data reduction method based on Principal Component Analysis,and the data normalization method with Max-Min processing are proposed.Make data more applicable to model processing.Finally,the data set is selected to Instance validation for the green channel vehicle inspection model.The balanced data set is constructed and the green channel vehicle data set is preprocessed by an under-sampling method.A three-layer BP neural network is established,and the number of nodes such as input layer,hidden layer,output layer,activation function and training function are determined.The multiple base classifiers is integrated into strong neural network classifier with Bagging.The recognition rate of the strong classifier to normal green channel vehicle was 80.37%,and the recognition rate of fake green traffic was 79.82%.Compared with the base classifier,the strong classifier with Bagging has a significant improvement in classification accuracy.Through establishing the freeway green channel vehicle inspection model,the effective identification of cheating vehicles is realized to some extent,which has reference significance of the green channel vehicle inspection.