| Electronic Stability Program(ESP)has become a standard part of conventional vehicles.It can greatly improve the stability of vehicles and ensure safe driving.ESP relies on on-board sensors to accurately collect the driving state of the vehicle.Once the sensors are attacked and feedback false vehicle attitude information to the controller,system failure or even serious traffic accident will occur.It is of practical significance to carry out acoustic attack test and attack detection research on vehicle sensors in this paper.In this paper,the ESP system is analyzed and modeled,and an attack and defense test system based on ESP is designed.On this basis,the sensor attack and data detection functions are realized,and then the intrusion protection strategy is proposed combined with the sensor injection attack.The specific contents are as follows:Firstly,this paper analyzes the structure principle of Micro Electro Mechanical Systems(MEMS)gyroscope and the cause and harm of the resonant frequency of the gyroscope sensor.At the same time,the modeling of the on-board ESP system was simplified,and the two-degree-of-freedom model of the vehicle was built,and the ESP algorithm based on fuzzy control was designed based on the theoretical yaw velocity.Furthermore,the physical device and upper computer software based on yaw velocity are designed.Finally,a set of semi-physical simulation ESP attack and defense test system is formed.Secondly,the covert attack method is designed on the ESP test system.Using the resonant hole of gyroscope sensor,the high-frequency sound wave is used to carry out non-intrusive injection attack to the sensor,so as to interfere the sensor system.At the same time,the frequency sweep method is designed to solve the different resonant frequencies of different sensors.In order to hide the attack signal of the adjacent ultrasonic or non-ultrasonic frequency,a method of music reencoding was designed,and a kind of imperceptible attack music was made by this method.The experimental evaluation of the test system shows that even if a small number of key sensors are attacked,the safe driving of the vehicle will be affected and the dangerous situation such as tail drift will be caused.Finally,two sensor anomaly data detection strategies based on deep learning,CUSUM-LSTM and TWO-STAGE LSTM,are designed for the hidden injection attack of the system sensor.The former is based on the learning of sensor data,which detects abnormal data by the way of error accumulation and sum,and realizes the rapid recovery of sensor system by the mechanism of replacing false data with predicted value.The latter has two feature learning and detection processes,which help restore the sensor system through multi-sensor working condition learning and detection in the first stage and learning for each working condition in the second stage,combined with error accumulation and abnormal data detection.Through the experimental comparison of the two detection strategies,it is shown that the two strategies can achieve higher prediction accuracy,and the latter has better anti-jamming and prediction ability than the former in the case of attack against small models. |
