Safety estimation method for driving state of intelligent electric vehicle under network attack

Abstract

The invention discloses an intelligent electric vehicle driving state safety estimation method under network attacks, and relates to automobile intelligent safety and automatic driving. Firstly, a three-degree-of-freedom dynamic model of the intelligent electric vehicle is established, the model comprehensively considers transverse, longitudinal and yaw dynamic characteristics of the vehicle, and accurate description is provided for subsequent state estimation. And designing an attack signal and state security estimation method based on an adaptive tolerance extended Kalman filtering algorithm. According to the method, the real-time and accurate estimation of the vehicle state is realized by fusing the information of various sensors such as GPS, LiDAR, IMU and the like, and malicious data introduced due to network attack can be effectively eliminated. Meanwhile, the method also has the capability of tolerating non-ideal network environments such as sensor data packet loss and the like, and the safety and reliability of the self-driving automobile are further enhanced. The intelligent electric vehicle can still maintain accurate driving state estimation when subjected to network attack, and safe driving of the vehicle is ensured.

Description

Technical Field

[0001] The present invention relates to automobile intelligent safety and automatic driving, and in particular to a method for safely estimating the driving state of an intelligent electric vehicle under a network attack, aiming at how to accurately estimate the driving state of an intelligent electric vehicle when the intelligent electric vehicle is attacked by a network.

[0002] Background Field

[0003] With the development of social economy, intelligent electric vehicles have been rapidly developed and widely used in various fields. Perception is the primary link of intelligent driving. Autonomous vehicles use various sensors such as global positioning system (GPS), light radar (LiDAR), inertial measurement unit (IMU) to locate themselves and perceive the environment. At present, autonomous vehicles are able to share their local sensor information with surrounding vehicles and roadside units to improve traffic efficiency, safety or form vehicle queues. However, these additional sensors make the vehicle system vulnerable to network attacks such as denial of service and injection of false information. Malware interference with intelligent electric vehicles may cause serious safety problems and may lead to accidents. Therefore, it is very critical to shield the attack signal through vehicle state estimation methods such as Kalman filtering to ensure the authenticity and accuracy of vehicle perception signals. Reference 1 (Xia X, Hashemi E, Xiong L, et al. Autonomous vehicle kinematics and dynamics synthesis for sideslip angle estimation based on consensus kalman filter [J]. IEEE Transactions on Control Systems Technology, 2022, 31 (1): 179-192.) proposed a multi-sensor fusion framework for vehicle sideslip estimation based on consensus Kalman filter, which integrates kinematic and dynamic characteristics to accurately estimate the vehicle sideslip angle. Reference 2 (Wang Y, Yan Y, Shen T, et al. An event-triggered scheme for state estimation of preceding vehicles under connected vehicle environment [J]. IEEE Transactions on Intelligent Vehicles, 2022, 8 (1): 583-593.) proposed an event-triggered estimation framework based on a coupled vehicle model, which combines the event trigger mechanism with the embedded volumetric Kalman filter for the state estimation of the preceding vehicle.

[0004] However, existing vehicle driving state estimation methods do not take into account the inaccurate and abnormal characteristics of sensor and communication network transmission data caused by malicious attacks. In an open and complex driving environment, these abnormal factors will seriously affect the safety performance of autonomous driving vehicle driving state estimation and reduce the accuracy and robustness of state estimation. Summary of the invention

[0005] The purpose of the present invention is to solve the control problem of intelligent electric vehicles being interfered with by network attacks when obtaining their own motion status, and to provide a method for safely estimating the driving status of intelligent electric vehicles under network attacks, which can enable the intelligent electric vehicle to still obtain accurate motion status information when the sensor is subjected to network attacks, thereby ensuring the safe driving of the vehicle.

[0006] The technical solution of the present invention is: collecting driving state information of intelligent electric vehicles through a vehicle-mounted sensor system, considering that the sensor is subject to network attacks, and establishing a state estimation method for intelligent electric vehicles under network attacks. The present invention includes the following steps:

[0007] 1) A three-degree-of-freedom dynamic model describing the lateral, longitudinal and yaw dynamic characteristics of intelligent electric vehicles is established. The differential equations of the nonlinear intelligent electric vehicle dynamic model are simplified according to the conversion relationship between the resultant force on the tire in the x-axis and y-axis directions in the vehicle body coordinate system and the longitudinal and lateral forces. The nonlinear state space equations are discretized, the process noise and measurement noise in the Kalman filter are introduced, and the influence of network attack signals and data packet loss on the sensor measurement output signal is added;

[0008] 1.1) Establish a dynamic model that accurately characterizes the behavior mechanism of intelligent electric vehicles; the model needs to take into account the lateral, longitudinal and yaw dynamic characteristics of the vehicle to fully reflect the dynamic behavior of the vehicle during actual driving;

[0009] 1.2) Establish the conversion relationship between the longitudinal and lateral forces on the tire in the x-axis and y-axis directions in the vehicle body coordinate system;

[0010] 1.3) Simplify the nonlinear state space equation of the intelligent electric vehicle;

[0011] 1.4) The above nonlinear state space equation is approximately discretized using the first-order difference quotient method to obtain the nonlinear discrete state equation;

[0012] 1.5) Introduce the process noise and measurement noise of Kalman filter into the discrete state equation to reflect various uncertain factors that may exist in the actual driving process, and establish the state equation and measurement equation of Kalman filter based on the nonlinear discrete state equation;

[0013] 1.6) Considering the impact of cyber attacks on sensor output signals, introduce cyber attack signals to simulate cyber attack scenarios that may be encountered during actual driving, and establish measurement equations for sensors subjected to cyber attacks;

[0014] 1.7) Considering the impact of non-ideal vehicle network environment on sensor data, random data packet loss is introduced to establish the measurement equation for sensor data packet loss;

[0015] 2) Design an attack signal and state security estimation method based on the adaptive tolerant extended Kalman filter algorithm; use the information of GPS and LiDAR sensors to fuse with the information of IMU sensors to estimate the attack signal, and remove the malicious part of the sensor data that affects the state estimation in combination with the estimated attack signal, and output the safe state estimation value;

[0016] 2.1) The vehicle's longitudinal displacement and lateral displacement are acquired in real time through the vehicle's GPS system; the vehicle's longitudinal displacement, lateral displacement and yaw angle are acquired in real time through the vehicle's LiDAR system; the vehicle's longitudinal velocity, lateral velocity and yaw angular velocity are acquired in real time through the IMU system; these sensor information are used as input data for subsequent state estimation;

[0017] 2.2) Design an adaptive tolerant extended Kalman filter to fuse the sensor information of GPS and IMU, LiDAR and IMU, establish an attack signal estimator through the adaptive tolerant extended Kalman filter, and remove the attack estimation signal to obtain the vehicle driving status information;

[0018] 2.3) Based on the multi-sensor switching control rule of average malicious disturbance power, the estimation result of the adaptive tolerant extended Kalman filter with smaller average malicious disturbance power is selected as the final result; by comparing the malicious disturbance power of different sensors, the optimal sensor data is selected for state estimation to improve the accuracy and reliability of the estimation result;

[0019] 2.4) Based on the above process, a multi-sensor attack signal estimation and recovery mechanism is established to solve the colored noise problem caused by non-ideal vehicle network environment (such as data packet loss), further enhance the tolerance of autonomous driving vehicles to sensors, and achieve effective recovery and compensation of sensor data.

[0020] Compared with the prior art, the present invention has the following outstanding technical effects and advantages:

[0021] The present invention integrates the adaptive tolerant extended Kalman filter algorithm with the multi-sensor attack signal estimation and recovery mechanism. By constructing a multi-sensor switching control rule based on the average malicious disturbance power, it can intelligently screen and combine the optimal sensor data, and effectively resist the data anomaly and noise interference caused by network attacks. On this basis, the adaptive tolerant extended Kalman filter algorithm further plays a role, not only can it adjust the filter parameters in real time to cope with the dynamically changing network environment, but also can accurately estimate the vehicle driving state, such as position, speed and acceleration. When the sensor data is distorted by the network attack, the multi-sensor attack signal estimation and recovery mechanism is immediately started, and the disturbed sensor data is restored by using the characteristics of historical data and sensor redundancy, so as to maintain the continuity and accuracy of the state estimation. The present invention not only significantly improves the safety and robustness of intelligent electric vehicles in complex network environments, but also optimizes the utilization efficiency of sensors, reduces energy consumption and costs. The present invention can still accurately estimate the driving state of the vehicle when the sensor is attacked by the network, ensuring the safe driving of the vehicle. The method improves the fault tolerance and anti-attack ability of the vehicle to sensor data, and provides a strong guarantee for the safe driving of intelligent electric vehicles. It has laid a solid foundation for the further development of autonomous driving technology and promoted breakthroughs and innovations in the technology for safe driving of smart electric vehicles under cyber attacks. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a diagram of the distributed drive vehicle dynamics model.

[0023] Figure 2 Flowchart of the multi-sensor attack signal estimation and recovery mechanism. DETAILED DESCRIPTION

[0024] In order to explain the technical content, structural features, and implementation objectives of the present invention in detail, the present invention is fully explained below in conjunction with the accompanying drawings.

[0025] The overall working idea of ​​the present invention is that when a smart electric vehicle suffers a network attack during normal driving, a network attack detection module is used to perform network attack detection. When it is detected that a sensor is under a network attack, the sensor under the network attack is immediately shielded and the state estimation is performed using the information of sensors that are not under the network attack.

[0026] The embodiment of the present invention mainly includes two steps:

[0027] Step 1: Establish a dynamic model of smart electric vehicles under cyber attacks.

[0028] Step 1.1, such as Figure 1 As shown in the figure, a dynamic model that accurately characterizes the behavior mechanism of the intelligent electric vehicle is established. According to Newton's second law, the longitudinal, lateral and yaw force balance equations and the conversion equations between the vehicle body coordinate system and the inertial coordinate system are obtained:

[0029] (1)

[0030] In the formula, , are the distances from the center of mass to the front and rear axles, respectively; m is the vehicle curb weight, is the moment of inertia of the vehicle around the z-axis, is the lateral velocity of the vehicle, is the lateral acceleration of the vehicle, is the longitudinal velocity of the vehicle, is the longitudinal acceleration of the vehicle, is the yaw angle of the vehicle, is the vehicle’s yaw rate, is the vehicle's yaw angular acceleration, is the lateral position in the geodetic coordinate system, is the lateral velocity in the geodetic coordinate system, is the vertical position in the geodetic coordinate system, is the longitudinal velocity in the geodetic coordinate system; is the force on the front wheel in the x-axis direction, is the force on the rear wheel in the x-axis direction, is the force on the front wheel in the y-axis direction, is the force on the rear wheel in the y-axis direction.

[0031] Step 1.2, establish the conversion relationship between the longitudinal and lateral forces on the tire in the x-axis and y-axis directions in the vehicle body coordinate system:

[0032] (2)

[0033] In the formula, is the front wheel longitudinal force, is the rear wheel longitudinal force, is the lateral force on the front wheel, is the rear wheel lateral force, is the front wheel turning angle.

[0034] Assuming that the tire cornering characteristics of smart electric vehicles are within the linear range, the longitudinal force and lateral force of the tire are obtained using the following formula:

[0035] (3)

[0036] In the formula, is the longitudinal stiffness of the front tire; is the longitudinal stiffness of the rear tire; is the lateral stiffness of the front tire; is the lateral stiffness of the rear tire; is the front tire slip rate; is the rear tire slip rate; is the front tire slip angle; is the rear tire slip angle.

[0037] By simplifying the calculation, the calculation formula of the tire side slip angle is obtained:

[0038] (4)

[0039] In the formula, is the front tire slip angle, is the rear tire slip angle, is the lateral velocity of the vehicle, is the vehicle’s yaw rate, is the longitudinal velocity of the vehicle; , are the distances from the center of mass to the front and rear axles respectively; is the front wheel deflection angle.

[0040] Step 1.3, the nonlinear state space equation of the intelligent electric vehicle is simplified, and the differential equation group of the nonlinear intelligent electric vehicle dynamics model is obtained by combining formulas (1)-(4) as follows:

[0041] (5)

[0042] In the formula, is the vehicle curb weight, is the longitudinal acceleration of the vehicle; is the lateral acceleration of the vehicle, is the vehicle's yaw angular acceleration, is the lateral velocity of the vehicle, is the vehicle’s yaw rate, is the longitudinal stiffness of the front tire; is the longitudinal stiffness of the rear tire, is the lateral stiffness of the front tire, is the lateral stiffness of the rear tire, is the front tire slip rate, is the rear tire slip rate, is the moment of inertia of the vehicle around the z-axis, is the longitudinal velocity of the vehicle in the geodetic coordinate system, is the lateral velocity of the vehicle in the geodetic coordinate system.

[0043] Let the state variable be , the control quantity is ; Then the nonlinear dynamic model represented by formula (5) is abstracted into a nonlinear state space equation, as shown in formula (6):

[0044] (6)

[0045] In the formula, represents the state variable at time t, represents the control variable at time t, represents the output variable, represents the first-order differential of the state variable at time t, Represents abstract about and The function of express and function.

[0046] Step 1.4, the nonlinear state space equation is approximately discretized using the first-order difference quotient method to obtain the nonlinear discrete state equation, as follows:

[0047] (7)

[0048] In the formula, for The state quantity at the moment, for The state quantity at the moment, is the control quantity at time k, is the output at time k, is the sampling period.

[0049] Step 1.5, the state equation and measurement equation of Kalman filtering, considering process noise and measurement noise, are integrated into the following equation:

[0050] (8)

[0051] In the formula, , represents the process noise, represents the measurement noise.

[0052] Step 1.6, establish the measurement equation for the sensor under network attack, the sensor data at time k under network attack The equation is described as:

[0053] (9)

[0054] In the formula, It is malicious data injected by the attacker. Sensor attack matrix set up for attackers;

[0055] Step 1.7, establish the measurement equation for sensor data packet loss. Combine equations (8) and (9) to establish the nonlinear discrete state equation and measurement equation for the vehicle under network attack:

[0056] (10)

[0057] In the formula, represents the process noise, represents the measurement noise, is the output variable, It is malicious data injected by the attacker. Sensor attack matrix set up for attackers. express The state quantity at the moment, express The state quantity at a moment.

[0058] Based on formula (10), considering random packet loss, the nonlinear discrete state equation and measurement equation of the autonomous driving vehicle are as follows:

[0059] (11)

[0060] In the formula, , , is the sensor measurement output, is an independent random variable introduced to describe the random measurement delay and random packet loss phenomenon of the system; when the variable Indicates that the vehicle communication network successfully transmits data without packet loss, and the system is in a closed-loop state; when the variable When , it means that the data is not successfully transmitted through the vehicle communication network, packet loss occurs, and the system is in an open-loop state; the state transition probability matrix of the two-state Markov chain is expressed as:

[0061] , , (12)

[0062] In the formula, p ij represents the state transition probability of the Markov chain, for Independent random variables at time, for are independent random variables at time .

[0063] Step 2: Design an attack signal and state security estimation method based on the adaptive tolerant extended Kalman filter algorithm.

[0064] Step 2.1: The smart electric vehicle obtains the longitudinal displacement of the vehicle in real time through the on-board GPS system and lateral displacement ; Obtain the longitudinal displacement of the vehicle in real time through the on-board LiDAR system , lateral displacement and yaw angle ; Obtain the longitudinal acceleration of the vehicle in real time through the IMU system , lateral acceleration and yaw rate .

[0065] Step 2.2, fuse the sensor information of GPS and IMU, and fuse the sensor information of LiDAR and IMU.

[0066] Among them, the sensor information of GPS and IMU is integrated, including:

[0067] The first step is to initialize the process noise covariance matrix , measurement noise covariance matrix and the covariance matrix :

[0068] , ,

[0069] The second step is to measure the vehicle's current longitudinal acceleration, lateral acceleration and yaw rate through the IMU. Make a one-step prediction for the state quantity as follows:

[0070] (13)

[0071] In the formula, Based on Always The prior state at the moment, for The posterior state at time t, for Relative to The state increment, is the sampling period, and the increment of each state is calculated as follows:

[0072]

[0073] The third step is to The posterior covariance matrix at time ,calculate The prior covariance matrix at time :

[0074] (14)

[0075] In the formula , for The transpose of

[0076] Step 4: Calculate Kalman gain at time , the discretized measurement equation of the vehicle is as follows:

[0077] (15)

[0078] In the formula, For vehicles The measurement output at the moment, is the horizontal position, For vertical position.

[0079] The Kalman gain calculation formula is as follows:

[0080] (16)

[0081] In the formula, , is the output matrix.

[0082] Step 5: Use the sensor measurement value to update the state quantity and obtain the state quantity at the next moment ;sensor Measurement value at time As shown below:

[0083] (17)

[0084] In the formula, and The longitudinal and lateral positions are measured by GPS;

[0085] The state update equation is as follows:

[0086] (18)

[0087] In the formula, for The posterior state at time t, For the prior state The output obtained.

[0088] Step 6: Calculate the error vector of the first Kalman filter :

[0089] (19)

[0090] In the formula, The a posteriori state The output obtained.

[0091] Step 7: Update the covariance matrix , the formula is as follows:

[0092] (20)

[0093] In the formula, It is a 6th-order unit array.

[0094] After entering the next cycle, repeat the above steps, and cycle like this to achieve the fusion of GPS and IMU.

[0095] Fusion of LiDAR and IMU sensor information, including:

[0096] The first step is to initialize the process noise covariance matrix , measurement noise covariance matrix and the covariance matrix :

[0097] , ,

[0098] The second step is to measure the vehicle's current longitudinal acceleration, lateral acceleration and yaw rate through the IMU. Make a one-step prediction for the state quantity as follows:

[0099] (twenty one)

[0100] In the formula, the function With function The format is exactly the same.

[0101] The third step is to The posterior covariance matrix at time ,calculate The prior covariance matrix at time :

[0102] (twenty two)

[0103] In the formula, , for The transpose of

[0104] Step 4: Calculate Kalman gain at time , the discretized measurement equation of the vehicle is as follows:

[0105] (twenty three)

[0106] In the formula, For vehicles The measurement output at the moment, is the yaw angle, is the horizontal position, For vertical position.

[0107] The Kalman gain calculation formula is as follows:

[0108] (twenty four)

[0109] In the formula, , is the output matrix.

[0110] Step 5: Use the sensor measurement value to update the state quantity and obtain the state quantity at the next moment ;sensor Measurement value at time As shown below:

[0111] (25)

[0112] In the formula, , and The yaw angle, longitudinal position and lateral position measured by LiDAR;

[0113] The state update equation is as follows:

[0114] (26)

[0115] In the formula, for The posterior state at time t, For the prior state The output obtained.

[0116] Step 6: Calculate the error vector of the second Kalman filter:

[0117] (27)

[0118] In the formula, The a posteriori state The output obtained.

[0119] Step 7: Update the covariance matrix , the formula is as follows:

[0120] (28)

[0121] In the formula, It is a 6th-order unit array.

[0122] After entering the next cycle, repeat the above steps, and cycle like this to achieve the fusion of LiDAR and IMU.

[0123] In step 2.3, attack signal estimators are established for GPS and LiDAR respectively, and the malicious parts that affect the state estimation in the sensor data are removed by combining the estimated attack signals, and the safe state estimation value is output.

[0124] Considering the nonlinear discrete state equation and measurement equation (11) of a vehicle suffering from random packet loss interference in a non-ideal communication network environment, the following filter is constructed to estimate the vehicle state:

[0125] (29)

[0126] In the formula, and for The prior state and the posterior state at the moment, , , , , and They are The posterior state, control variable, process noise, system matrix, Kalman gain and measurement error at time t.

[0127] The optimal filter gain is It can be expressed as:

[0128] (30)

[0129] In the formula, , represents the Hadamard product, is the measurement noise covariance matrix, for The prior covariance matrix at time , is the output matrix, is the random packet loss probability matrix.

[0130] Prior and a posteriori state estimation error covariance matrices and can be written as:

[0131] (31)

[0132] In the formula, is the process noise covariance matrix.

[0133] (32)

[0134] Measuring data with sensors and vehicle control input data (including front wheel steering angle and additional torque) Design an attack signal estimator for the vehicle model (11) considering non-ideal vehicle network environment and network attack interference .

[0135] The attack signal estimator continuously estimates the attack signal and the comprehensive disturbance caused by communication packet loss, and ensure that the estimated attack signal can meet the following conditions:

[0136] (33)

[0137] In the formula, and is the attack signal estimator Estimated attack vectors, To control the amount, The sensor measurement value.

[0138] It can be assumed The initial condition is and Under this initial condition, Input into the adaptive tolerant extended Kalman filter as shown in equation (13). It can be obtained that the attack signal estimator as follows:

[0139] The first step is the prior state estimate:

[0140] (34)

[0141] In the formula, the variable , , and and functions The form of is the same as the state equation in equation (11).

[0142] Step 2: Deviation between prior state and measured state calculate:

[0143] (35)

[0144] In the formula, is the measured value after the sensor removes the attack signal, is the system output matrix.

[0145] Step 3: Posterior State estimate:

[0146] (36)

[0147] In the formula, is the system matrix, is the Kalman gain.

[0148] Step 4: Attack Signal calculate:

[0149] (37)

[0150] In the formula, , ;in ,

[0151] , ,

[0152] , , ,

[0153] , , ,

[0154] , , , ,

[0155] ;

[0156] , , , .

[0157] and They represent the vehicle driving state information obtained after fusing GPS and IMU sensor signals and fusing LiDAR and IMU sensor signals and removing the attack estimation signal. The attack signal injected into the sensor measurement data is estimated by the above method, and then the hidden malicious data in the data is removed, thereby restoring the estimated performance of the vehicle system.

[0158] In step 2.4, the estimated value of the adaptive tolerant extended Kalman filter with a smaller average malicious disturbance power is selected as the final result.

[0159] Define the attacker's The average malicious disturbance power on :

[0160] (38)

[0161] In the formula, They are GPS and LiDAR sensors, For the current moment, is the step length of recording the attack signal forward based on the current moment, For Time to Every cycle of time, is the 2-norm of the attack signal vector, which is used to evaluate the size of the malicious disturbance.

[0162] Once the attack begins, the attacker will continue to launch attacks, that is, for any sensor i, the average malicious disturbance power satisfies The final safety state estimate output for:

[0163] (39)

[0164] In the formula, is the estimated value of GPS status, is the estimated value of the LiDAR state.

[0165] Step 2.5: Based on the above process, establish Figure 2 The multi-sensor attack signal estimation and recovery mechanism shown in the figure. First, the information of GPS and LiDAR sensors is fused with the information of IMU sensors to estimate the attack signal, and the malicious part of the sensor data that affects the state estimation is removed in combination with the estimated attack signal, and the safe state estimation value is output. And based on the multi-sensor switching control rule of the average malicious disturbance power, the estimation result of the adaptive tolerant extended Kalman filter with a smaller average malicious disturbance power is selected as the final result.

[0166] The above contents are further detailed descriptions of the present invention in combination with the preferred technical solutions, and the specific implementation of the invention cannot be limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, simple deductions and substitutions can be made without departing from the concept of the present invention, which should be regarded as the protection scope of the present invention.

Claims

1. A method for safely estimating the driving state of an intelligent electric vehicle under a network attack, characterized in that The following steps are involved: 1) A three-degree-of-freedom dynamic model describing the lateral, longitudinal and yaw dynamic characteristics of intelligent electric vehicles is established. The differential equations of the nonlinear intelligent electric vehicle dynamic model are simplified according to the conversion relationship between the resultant force on the tire in the x-axis and y-axis directions in the vehicle body coordinate system and the longitudinal and lateral forces. The nonlinear state space equations are discretized, the process noise and measurement noise in the Kalman filter are introduced, and the influence of network attack signals and data packet loss on the sensor measurement output signal is added; 1.1) Establish a dynamic model that accurately characterizes the behavior mechanism of intelligent electric vehicles. The model needs to take into account the lateral, longitudinal and yaw dynamic characteristics of the vehicle to fully reflect the dynamic behavior of the vehicle during actual driving; 1.2) Establish the conversion relationship between the longitudinal and lateral forces on the tire in the x-axis and y-axis directions in the vehicle body coordinate system; 1.3) Simplify the nonlinear state space equation of the intelligent electric vehicle; 1.4) The above nonlinear state space equation is approximately discretized using the first-order difference quotient method to obtain the nonlinear discrete state equation; 1.5) Introduce the process noise and measurement noise of Kalman filter into the discrete state equation to reflect various uncertain factors that may exist in the actual driving process, and establish the state equation and measurement equation of Kalman filter based on the nonlinear discrete state equation; 1.6) Considering the impact of cyber attacks on sensor output signals, introduce cyber attack signals to simulate cyber attack scenarios that may be encountered during actual driving, and establish measurement equations for sensors subjected to cyber attacks; 1.7) Considering the impact of non-ideal vehicle network environment on sensor data, random data packet loss is introduced to establish the measurement equation for sensor data packet loss; 2) Design an attack signal and state security estimation method based on the adaptive tolerant extended Kalman filter algorithm; use the information of GPS and LiDAR sensors to fuse with the information of IMU sensors to estimate the attack signal, and remove the malicious part of the sensor data that affects the state estimation in combination with the estimated attack signal, and output the safe state estimation value; 2.1) The vehicle's longitudinal displacement and lateral displacement are acquired in real time through the vehicle's GPS system; the vehicle's longitudinal displacement, lateral displacement and yaw angle are acquired in real time through the vehicle's LiDAR system; the vehicle's longitudinal velocity, lateral velocity and yaw angular velocity are acquired in real time through the IMU system; these sensor information are used as input data for subsequent state estimation; 2.2) Design an adaptive tolerant extended Kalman filter to fuse the sensor information of GPS and IMU, LiDAR and IMU, establish an attack signal estimator through the adaptive tolerant extended Kalman filter, and remove the attack estimation signal to obtain the vehicle driving status information; 2.3) Based on the multi-sensor switching control rule of average malicious disturbance power, the estimation result of the adaptive tolerant extended Kalman filter with smaller average malicious disturbance power is selected as the final result; by comparing the malicious disturbance power of different sensors, the optimal sensor data is selected for state estimation to improve the accuracy and reliability of the estimation result; 2.4) Based on the above content, a multi-sensor attack signal estimation and recovery mechanism is established to solve the colored noise problem caused by the non-ideal vehicle network environment, further enhance the tolerance of autonomous driving vehicles to sensors, and achieve effective recovery and compensation of sensor data.

2. The method for safely estimating the driving state of an intelligent electric vehicle under a network attack as claimed in claim 1 is characterized in that In step 1.3), the nonlinear state space equation of the intelligent electric vehicle is: In the formula, represents the state variable at time t, represents the control variable at time t, represents the output variable, represents the first-order differential of the state variable at time t, Represents abstract about and The function of express and function.

3. The method for safely estimating the driving state of an intelligent electric vehicle under a network attack as claimed in claim 1 is characterized in that In step 1.4), the nonlinear discrete state equation is as follows: In the formula, for The state quantity at the moment, for The state quantity at the moment, is the control quantity at time k, is the output at time k, is the sampling period.

4. The method for safely estimating the driving state of an intelligent electric vehicle under a network attack as claimed in claim 1 is characterized in that In step 1.5), the state equation and measurement equation of the Kalman filter are as follows: In the formula, , represents the process noise, represents the measurement noise.

5. The method for safely estimating the driving state of an intelligent electric vehicle under a network attack as claimed in claim 1 is characterized in that In step 1.6), the measurement equation of the sensor under network attack, the sensor data at time k under network attack The equation is described as: In the formula, It is malicious data injected by the attacker. Sensor attack matrix set up for attackers.

6. The method for safely estimating the driving state of an intelligent electric vehicle under a network attack as claimed in claim 1 is characterized in that In step 1.7), the measurement equation of the sensor when data packet loss occurs is as follows: Establish the nonlinear discrete state equation and measurement equation of the vehicle under network attack: In the formula, represents the process noise, represents the measurement noise, is the output variable, It is malicious data injected by the attacker. Sensor attack matrix set up for attackers; express The state quantity at the moment, express The state quantity at the moment; Based on the above equation, considering random packet loss, the nonlinear discrete state equation and measurement equation of the autonomous driving vehicle are as follows: In the formula, , , is the sensor measurement output, is an independent random variable introduced to describe the random measurement delay and random packet loss phenomenon of the system; when the variable Indicates that the vehicle communication network successfully transmits data without packet loss, and the system is in a closed-loop state; when the variable When , it means that the data is not successfully transmitted through the vehicle communication network, packet loss occurs, and the system is in an open-loop state; the state transition probability matrix of the two-state Markov chain is expressed as: ， ， In the formula, p ij represents the state transition probability of the Markov chain, for Independent random variables at time, for are independent random variables at time .

7. The method for safely estimating the driving state of an intelligent electric vehicle under a network attack as claimed in claim 1 is characterized in that In step 2.2), an adaptive tolerant extended Kalman filter is designed to fuse the sensor information of GPS and IMU, LiDAR and IMU. An attack signal estimator is established through the adaptive tolerant extended Kalman filter, and the attack estimation signal is eliminated to obtain the vehicle driving state information. Specifically: Considering the nonlinear discrete state equation and measurement equation of a vehicle suffering from random packet loss interference in a non-ideal communication network environment, the following filter is constructed to estimate the vehicle state: In the formula, and for The prior state and the posterior state at the moment, , , , , and They are The a posteriori state, control variable, process noise, system matrix, Kalman gain and measurement error at the moment; The optimal filter gain is It is expressed as: In the formula, , represents the Hadamard product, is the measurement noise covariance matrix, for The prior covariance matrix at time , is the output matrix, is the random packet loss probability matrix; Prior and a posteriori state estimation error covariance matrices and Written as: In the formula, is the process noise covariance matrix; Measuring data with sensors and vehicle control input data Design an attack signal estimator for a vehicle model considering non-ideal vehicle network environment and network attack interference ; The attack signal estimator continuously estimates the attack signal and the comprehensive disturbance caused by communication packet loss, and ensure that the estimated attack signal can meet the following conditions: In the formula, and is the attack signal estimator Estimated attack vectors, To control the amount, is the sensor measurement value; assumed The initial condition is and Under this initial condition, Input to the adaptive tolerant extended Kalman filter; So we get the attack signal estimator as follows: The first step is the prior state estimate: In the formula, the variable , , and and functions The form of is the same as the nonlinear discrete state equation of an autonomous vehicle; Step 2: Deviation between prior state and measured state calculate: In the formula, is the measured value of the sensor after removing the attack signal, is the system output matrix; Step 3: Posterior State estimate: In the formula, is the system matrix, is the Kalman gain; Step 4: Attack Signal calculate: In the formula, , ;in , ， ， ， ， ， ， ， ， ， ， ， ， ； ， ， ， ； and They respectively represent the vehicle driving state information obtained after fusing the GPS and IMU sensor signals and fusing the LiDAR and IMU sensor signals and removing the attack estimation signal; the attack signal injected into the sensor measurement data is estimated by the above method, and then the hidden malicious data in the data is removed, thereby restoring the estimated performance of the vehicle system.

8. The method for safely estimating the driving state of an intelligent electric vehicle under a network attack as claimed in claim 1 is characterized in that In step 2.3), based on the multi-sensor switching control rule of the average malicious disturbance power, the estimation result of the adaptive tolerant extended Kalman filter with a smaller average malicious disturbance power is selected as the final result; specifically: Define the attacker's The average malicious disturbance power on : In the formula, They are GPS and LiDAR sensors, For the current moment, is the step length of recording the attack signal forward based on the current moment, For Time to Every cycle of time, is the 2-norm of the attack signal vector, which is used to evaluate the size of the malicious disturbance; Once the attack begins, the attacker will continue to launch attacks, that is, for any sensor i, the average malicious disturbance power satisfies ; Final safety status estimate output for: In the formula, is the estimated value of GPS status, is the estimated value of the LiDAR state.

Images