Recognition Of Abnormal Driving Behavior Of Key Commercial Vehicles Based On Beidou Data

With the increasing number of motor vehicles in China,the supervision and detection of abnormal driving behavior of vehicles has become an urgent problem to be solved.Key commercial vehicles are mainly commercial buses,commercial trucks and dangerous goods transport vehicles.Compared with other types of vehicles,the consequences of traffic accidents caused by abnormal driving behaviors are often more serious for key commercial vehicles.In order to strengthen the supervision of abnormal driving behaviors of key commercial vehicles,this paper conducts an in-depth study on the Beidou data of key commercial vehicles,divides four kinds of abnormal driving behaviors and constructs a sample data set on the basis of data pre-processing and statistical analysis,uses multiple recognition models to recognize the abnormal driving behaviors in the data set,and proposes a practical combination model TSA-MCNN for the recognition of abnormal driving behaviors.The purpose of accurately recognizing abnormal driving behaviors is achieved.The main work of the paper is as follows.(1)Pre-processing is carried out for the abnormal data existing in the collection and storage process of Bei Dou data,and statistical analysis of the pre-processed data is performed.Firstly,corresponding data cleaning rules are formulated according to the abnormal data types to complete the pre-processing of Beidou data,and the abnormal data cleaning flow chart is summarized and drawn.Secondly,statistical analysis of the cleaned data was conducted to clarify the deficiencies of the data in terms of abnormal driving behavior warning.(2)The abnormal driving behaviors in the data are classified and the abnormal sample dataset is constructed.First,the segmented linear interpolation method with sliding window overlap slicing method is used to convert the Bei Dou data into equal length and equal spacing time series.Secondly,the vehicle trajectory parameter matrix is constructed,and the parameters include dimensionality reduction data based on sparse autoencoder,standard deviation,etc.Then,the data are clustered based on the vehicle trajectory parameter matrix with the K-mean clustering algorithm,and then four abnormal driving behaviors are classified to build the sample data set of abnormal driving behaviors.(3)In order to recognize abnormal driving behaviors in Bei Dou data,a DTW-KNN recognition model and a four-layer convolutional neural network recognition model were built to achieve fast and accurate recognition of abnormal driving behaviors.First,the DTW-KNN model was built,the parameters in the model were analyzed and determined,and the accuracy of the model in recognizing abnormal driving behavior was tested at 91%.Second,the four-layer convolutional neural network model was built,the influence of the internal parameters of the model was analyzed,the various parameters of the model were initialized,and the accuracy of the four-layer convolutional neural network model for the recognition of abnormal driving behavior was tested at 88.67%.(4)In order to further improve the accuracy of abnormal driving behavior recognition,this paper proposes a TSA-MCNN model combining time series symbolization algorithm and multi-scale convolutional neural network algorithm to enhance the recognition of abnormal driving behavior.First,the time series symbolization algorithm is introduced,static and dynamic symbolization algorithms are designed,and the TSA-MCNN model is built.Secondly,the parameters of the convolutional layers of the model and the number of channels configured were analyzed and determined.Then,the comparison with the DTW-KNN model,the four-layer convolutional neural network model and the MCNN model shows that the recognition accuracy of the TSA-MCNN model is 97%,which is 6%,8.33% and 11% higher than that of the DTW-KNN model,the four-layer convolutional neural network model and the MCNN model,respectively.