The Method Of Diagnosing Vehicle Anomalies Driven By Vehicle Sensor Data

In the era of big data and artificial intelligence,vehicle information system is gradually developing towards smart internet of vehicle information system.More and more smart cars provide intelligent and convenient services for users through various onboard softwares,which provides users with a better driving experience.However,the ever-growing cyber components of vehicles incur new risks to reliability/security of vehicles.On the one hand,the intelligent software system of vehicles becomes errorprone with the ever-increasing data volume in the in-vehicle network.On the other hand,due to the network is vulnerable,in-vehicle networks expose new vulnerabilities of cyberattacks.In order to diagnose anomalies in the in-vehicle network information system,the existing research works mainly study the methods of intrusion detection and anomaly diagnosis for the in-vehicle network.However,these diagnosis methods only rely on the internal information of the in-vehicle network,and the in-vehicle network itself is easy to be attacked,so those methods themselves are also unreliable.According to the demand of anomaly diagnosis of vehicle sensory data,this thesis proposes a method to diagnose vehicle anomalies from the perspective of cyber-physical system of vehicles,which taking the external information of in-vehicle network(i.e.,vehicle battery voltage)as the trust-worthy information and using the correlation graph model constructed between the external information and internal information of in-vehicle network to diagnosing anomalies.This method avoids the unreliability of in-vehicle network,covers the diagnosis of key sensory variables of vehicles,and provides a new direction for the research field of vehicle anomaly diagnosis.The main research contents of this thesis are as follows:(1)In this thesis,the graph model describing correlations between external information and internal information of the in-vehicle network is studied.Firstly,combining the physical and numerical correlations between external battery information and internal sensory information of vehicles,the Dynamic Time Warping(DTW)algorithm is used to extract the strong correlations between peaks of them,and then the correlation graph model of sensory information is constructed based on different levels of correlation.In the graph model,the vertex set V is divided based on different correlations between battery voltage and the vehicle sensor variables,the edge set E describing the correlations between variables is defined by DTW distance.In this way,the vehicle correlation graph model provides a theoretical basis for the research of vehicle anomaly diagnosis methods.(2)Based on the graph model of describing the correlations among vehicle variables,this thesis proposes the method of vehicle anomaly diagnosis which is guided by correlation graph model.The method diagnoses vehicle anomalies by establishing the data-driven mapping models for each edge in the edge set,and then using the mapping model to check whether the features of sensor variable conform to the model to diagnose the anomalies.In the model construction,the decision classification tree is used to learn the graph model,then the decision regression tree model is used to model the relationship between the eigenvectors of sensor variables,and the high prediction accuracy of the tree model is verified by the fact of prediction errors not exceeding 11%.Finally,this thesis also designs an algorithm of diagnosing vehicle anomalies and verifies the feasibility of the algorithm with vehicle datasets.(3)In this thesis,the algorithm of diagnosing vehicle anomalies is applied to the actual diagnosis system,and the performance of the vehicle anomaly diagnosis system is evaluated online with the 3-month driving datasets collecting from Subaru Crosstrek.A large number of experiments results show that the vehicle anomaly diagnosis system detects anomalies of vehicle sensor variables with larger than 87%(up to 100%)accuracy on average.Besides,the diagnosis system is highly sensitive to anomalies(more sensitive to abnormal data with drastic changes),and widely applicable to other types of vehicles.