Method for estimating safety state of networked automatic driving vehicle under network attack

Abstract

The invention provides a method for estimating the safety state of a networked automatic driving vehicle under network attack, relates to the field of automatic driving vehicle safety state estimation, and aims to estimate the safety state of the automatic driving vehicle by fusing the information measured by a target vehicle and the information of other vehicles in a communication range and combining a close degree algorithm. A risk isolation scheme is designed to achieve more accurate state estimation performance. The method comprises the following steps: fusing vehicle data in an Internet of Vehicles area, and establishing a networked automatic driving vehicle model; an iterative clustering algorithm is provided, abnormal behaviors of a vehicle are identified and early warning is carried out, abnormal vehicle data are eliminated, and a state estimation data set is updated; and defining a close degree function and a state estimation weight parameter, and carrying out safety state estimation on the vehicles in the Internet of Vehicles area. According to the invention, abnormal vehicle behaviors in the Internet of Vehicles area can be effectively handled, and the safety and adaptability of vehicle autonomous driving can be improved.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving, and in particular to a method for estimating the safety state of a networked autonomous driving vehicle under a network attack. Background Art

[0002] With the continuous development of science and technology, autonomous driving technology has become more mature and has been applied to transportation. However, as a driver, facing various situations, you still cannot completely hand over the control to the autonomous driving system. Therefore, how to ensure the safety of autonomous driving vehicles has become an important issue. One of the solutions is to estimate the state of the vehicle. Autonomous driving state estimation is one of the core technologies of autonomous driving. It can collect information around the vehicle through sensors such as radar, camera, lidar, etc., further identify road lines, obstacles, pedestrians, etc. on the road, and determine the vehicle's position, posture, speed and other information, so as to achieve vehicle control and planning.

[0003] Autonomous vehicles can provide real-time feedback on the estimated safety status of the current moment, allowing drivers to have a clearer understanding of the vehicle's current safety status. Many studies have shown that by analyzing vehicle operation data, a more accurate state estimation model can be obtained. In this way, the vehicle's real-time state data can be used to assist the driver to better understand the entire system, facilitate better decision-making, and ensure the safety and reliability of the vehicle. At the same time, the reliability of autonomous driving state estimation may also be threatened by cyber attacks. Hackers can tamper with sensor data, causing errors in vehicle control and planning, thereby causing traffic accidents.

[0004] Therefore, although there are many studies on state estimation algorithms, it is still a challenging problem if the network is attacked by cyber attacks. In order to ensure the safety and reliability of autonomous vehicles, it is necessary to pay attention to the threat of cyber attacks to autonomous driving state estimation, strengthen security testing and evaluation, identify security vulnerabilities, and solve and patch them in a timely manner. Summary of the invention

[0005] The purpose of the present invention is to provide a method for estimating the safety state of a networked autonomous driving vehicle under a network attack to solve the problems raised by the above-mentioned background technology.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] Integrate vehicle data in the connected vehicle area and establish a connected autonomous driving vehicle model;

[0008] Design an iterative clustering algorithm, propose a malicious vehicle warning scheme, and identify abnormal vehicle behavior;

[0009] Eliminate abnormal vehicle data and update the state estimation data set;

[0010] Define the proximity function and state estimation weight parameters;

[0011] Provide real-time feedback on the estimated safety status at the current moment and estimate the status at the next moment.

[0012] Furthermore, the vehicle data in the connected vehicle area is integrated and a connected autonomous driving vehicle model is established, including:

[0013] In the IoV area, each vehicle can collect and transmit data related to the vehicle and the surrounding environment through on-board equipment, such as vehicle speed, driving path, lane traffic conditions, etc. These data can be collected and used to build a networked autonomous driving vehicle model. At time k, the n vehicles in the IoV communication area are numbered and added to the set N. It is assumed that the number of vehicles communicating with the IoV remains unchanged during the K time interval, and all vehicles are fully connected, that is, any two vehicle communication nodes are connected. The connected vehicle must not only transmit its own sensor measurement information y i (k) Upload to the vehicle cloud and share other vehicle sensor information y between the vehicle and neighboring vehicles i|j (k) Upload to the vehicle cloud to achieve vehicle information redundancy. This paper assumes that the sensor dimensions are the same.

[0014] The vehicle dynamics system model is expressed as:

[0015] x i (k) = A i x i (k-1)+ω i (k-1)

[0016] Where A is the system state matrix, ω represents the system noise and its covariance matrix is ​​represented by Q;

[0017] Step S1.3, the vehicle sensor measurement equation can be expressed as:

[0018] y i (k) = H i x i (k)+v i (k)

[0019] Where H is the sensor measurement matrix, v represents the measurement noise and its covariance matrix is ​​denoted by R;

[0020] Step S1.4: The vehicle sensor measurement model shared between vehicles can be expressed as:

[0021] y j|i (k) = y i (k)

[0022] where y j|i Represents the measurement value shared by vehicle i received by vehicle j;

[0023] Step S1.5: The information about the i-th vehicle received by the vehicle cloud can be expressed as:

[0024] z i (k) = y i (k)+a i (k)

[0025] z j|i (k) = y j|i (k)+a j (k)

[0026] It should be noted that in order to simplify the expression of the formula and improve readability, the time parameter k is omitted in some descriptions.

[0027] where z i represents vehicle i sending its own measurement value to the vehicle cloud, z j|i Represents vehicle j sending the measurement value shared by vehicle i to the vehicle cloud; a i and a j is an item of maliciously manipulated information. Assuming that the communication link between the sensor of vehicle i and the vehicle cloud is working normally, then a i =0; otherwise, a i ≠0.

[0028] Furthermore, an iterative clustering algorithm is designed to propose a malicious vehicle early warning scheme to identify abnormal vehicle behavior, including:

[0029] By designing an iterative clustering algorithm, the vehicle cloud can quickly mark and cluster the vehicle's status information and evaluate the vehicle's abnormal behavior. If abnormal vehicle data is found, an alarm will be issued.

[0030] The iterative clustering algorithm includes parameter initialization, allocation step and update step. The goal is to divide the n sensor measurements into two categories and select the one with higher credibility for state estimation. The specific steps are as follows:

[0031] At time k, the vehicle cloud receives n sensor measurements about vehicle i, including the self-measurement value z sent by sensor node i. i and the measurement values ​​z sent by all neighboring nodes j|i , after fusion, {z il (k), l∈N}, where the set N includes all neighboring nodes of sensor node i and itself, and card(N)=n.

[0032] Initialize and define the category set {G ig|i=1,2,...,n.; g=1,2.}, where i is the sensor node label and g is the category label.

[0033] Initialization, set the mean vector m (g) , and assign a random initial value.

[0034] Initialization, define the distance function between two points:

[0035]

[0036] Initialize and define the matrix represents the clustering result of each iteration,

[0037]

[0038] In which, define u gl Represents the lth test sample z il (k) Is it assigned to category g? If so, then u gl =1; otherwise u gl =0.

[0039] Assignment step, the measured value z il (k) Assignment to category g requires the following conditions to be met:

[0040]

[0041] Update step, in order to match the sample measurement data in the current category, the following mean update is required:

[0042]

[0043] Repeat the assignment and update steps until the matrix U does not change.

[0044] Let k (c) is the number of data points belonging to the cth class, defined as follows:

[0045]

[0046] Select the trusted data set of sensor node i as follows:

[0047]

[0048] The g i The data within the class is placed into the confidence set G i In the set G i The data points in the set are considered to be credible data, and the number of elements in the set is k. (gi) ; Put the rest of the data into the collection At this time, the collection It is considered a suspicious set, and the number of elements in the set is nk (gi) .

[0049] The iterative clustering algorithm first integrates the sensor measurement data, and uses iterative allocation and updating to divide all the data into two groups. The group with higher confidence is selected for subsequent state estimation, and the other group with lower confidence is considered to be potentially risky. Each data in the set has its own unique label, which can be used to determine which vehicle the data is measured from and which vehicle sensor is used to send it to the vehicle cloud.

[0050] At each moment, a total of n vehicles upload their own sensor measurement information and the sensor information of n-1 vehicles shared with other vehicles to the cloud. That is, the vehicle cloud will receive n measurement values ​​of sensor i at a time node, and a total of n 2 At each time point, the cloud will determine the suspicious set corresponding to each vehicle sensor based on the iterative clustering algorithm. Count the sensors in the n suspicious sets within the K time interval, select sensors with a number greater than K / 2, and put them into the set Count the sensors in the n suspicious sets, select sensors with a number greater than n / 2, and put them into the set By prompting the technician that the vehicle may be under cyber attack, the driver is advised to intervene in the operating system as soon as possible and make a manual judgment on the current vehicle status.

[0051] Furthermore, the abnormal vehicle data is excluded and the state estimation data set is updated, including:

[0052] Malicious network attacks may cause vehicle cloud and multi-sensor data to be tampered with or injected with false information, which will seriously affect the accuracy and credibility of state estimation. It is necessary to develop an effective abnormal data identification mechanism to filter out abnormal data. Once an anomaly is detected, the data needs to be excluded to reduce its impact on state estimation. At the same time, it is also necessary to continuously update the state estimation data set so that autonomous driving vehicles can make more accurate road condition judgments. In the K time interval, by adopting a malicious vehicle early warning scheme, detect whether the data is suspicious, and obtain the safety sensor set Φ. At this time, card(Φ) = φ, remove the abnormal sensor data from the data set, and add the remaining data to the safety data set Ψ. At this time, This can effectively prevent abnormal data from misleading autonomous vehicles and improve the safety of autonomous driving systems.

[0053] Furthermore, the proximity function and state estimation weight parameters are defined, including:

[0054] In the vehicle cloud and multi-sensor environment, different data sources (vehicle sensors) may provide different measurement data, which may cause deviations and problems in the state estimation results. Therefore, defining the proximity function and state estimation weight parameters is a very important step. The measurement data provided by different data sources are compared and analyzed to calculate the similarity index between them. Based on these similarity indexes, we can effectively determine the relative reliability of different data sources and thus perform a weighted combination of the measurement data.

[0055] Define the closeness function:

[0056]

[0057] At time k, assuming that there are φ sensor measurements of vehicle i in the security dataset Ψ obtained by vehicle cloud screening, z iβ (k), (β=1,2,...,φ) means that the two-by-two combinations are calculated The closeness matrix between the measurements is as follows:

[0058]

[0059] The βth measurement value z of vehicle i iβ and any other measurement value z in the security dataset ip The consistency measure of the closeness between (p=1,2,...,φ) can be expressed as:

[0060]

[0061] Among them, E iβ Reflects the measured value z iβ How close it is to any other sensor measurement.

[0062] In order to obtain a more reasonable weight value, the consistency measure of the entire interval is considered. In the K time interval, the βth measurement value z iβ The mean consistency of closeness to any other measurement in the safety dataset is:

[0063]

[0064] In the K time interval, the βth measurement value z iβ The variance of the consistency of the closeness to any other measurement is:

[0065]

[0066] At this time, if the mean μ iβ is large, and the variance S iβis smaller, it means that the measured value z iβ It has higher reliability and should have higher weight in the data fusion process. Therefore, this design introduces the intermediate variable α iβ as follows:

[0067]

[0068] In the K time interval, define the sensor measurement value z iβ The weight parameter is ξ iβ as follows:

[0069]

[0070] Then, at time k, the credible measurement value of the i-th vehicle is as follows:

[0071]

[0072] Furthermore, the current security status estimate is fed back in real time and the next security status estimate is estimated, including:

[0073] Real-time feedback of the estimated safety state at the current moment and estimation of the state at the next moment are important steps to achieve stable operation and accurate decision-making of the autonomous driving system, which can be achieved by using the Kalman filter algorithm. Kalman filtering is a commonly used state estimation method that can accurately estimate the future state based on the current sensor measurement data and prediction model. By knowing the estimated safety state at the current moment, appropriate measures can be taken in time to protect the safety of the vehicle. At the same time, the state estimation at the next moment can also help us predict risks in advance and provide reliable predictions for future safety assurance. Based on the measurement value at the current moment, the state estimation is performed by using the Kalman filter algorithm and the vehicle model, as follows.

[0074] Initialize system status and the covariance matrix of the state

[0075] The time update steps are as follows to predict the system status in advance:

[0076]

[0077] Look-ahead forecast error covariance:

[0078]

[0079] The measurement update steps are as follows to calculate the Kalman gain coefficient:

[0080]

[0081] pass Update estimate status:

[0082]

[0083] Update the error covariance:

[0084]

[0085] The current state estimate is fed back to the vehicle operating system in real time, and the state at the next moment is predicted in advance for reference.

[0086] Compared with the prior art, the present invention has the following beneficial effects:

[0087] This method is based on the safety state estimation of networked autonomous driving vehicles and malicious vehicle warning. It improves the ability to deal with network attacks by utilizing multi-sensor data and vehicle cloud data redundancy; designs an iterative clustering algorithm to screen malicious sensors and improve the accuracy and timeliness of malicious vehicle warnings; at the same time, designs a proximity function to assign state estimation parameter weights, and quotes the Kalman filter algorithm to achieve higher accuracy and stability in the state estimation of autonomous driving vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 Schematic diagram of the vehicle network of the present invention

[0089] Figure 2 Flowchart of the present invention DETAILED DESCRIPTION

[0090] The specific implementation examples of the present invention are further described in detail below in conjunction with the accompanying drawings. The following specific implementation examples are used to illustrate and explain the present invention, but do not limit the scope of the present invention.

[0091] The present invention provides a method for estimating the safety state of a networked autonomous driving vehicle under a network attack, such as Figure 2 As shown, the following steps are included:

[0092] Step 1: Collect the vehicle's own driving data and send its own data to other vehicles in its communication area. Also send the collected data of other vehicles and its own data to the vehicle cloud. Number the five vehicles in the vehicle network communication area and add them to the set N. Assume that the number of vehicles communicating with the vehicle network remains unchanged within the K time interval, and that all vehicles are fully connected, that is, any two vehicle communication nodes are connected. Vehicle 1 not only uploads its own sensor measurement information y1(k) to the vehicle cloud, but also shares other vehicle sensor information y1(k) between vehicles. 1|2 (k), y 1|3 (k), y 1|4 (k) and y 1|5(k) Upload to the vehicle cloud to achieve vehicle information redundancy and integrate the data of 5 vehicles in the connected vehicle area. Establish a connected autonomous driving vehicle model. This paper assumes that the sensor dimensions are the same.

[0093] Build a simulation scenario and express the vehicle dynamics system model as:

[0094] x i (k) = A i x i (k-1)+ω i (k-1)

[0095] Where A is the system state matrix, ω represents the system noise and its covariance matrix is ​​represented by Q, and the parameters are set as follows:

[0096]

[0097] The system state is set to the position of the vehicle in the x and y directions.

[0098] The vehicle sensor measurement equation can be expressed as:

[0099] y i (k) = H i x i (k)+v i (k)

[0100] Where H is the sensor measurement matrix, v represents the measurement noise and its covariance matrix is ​​represented by R, and the parameters are set as follows:

[0101]

[0102] The vehicle sensor measurement model shared between vehicles can be expressed as:

[0103] y j|i (k) = y i (k)

[0104] where y j|i Represents the measurement value shared by vehicle i received by vehicle j;

[0105] The information about the i-th vehicle received by the vehicle cloud can be expressed as:

[0106] z i (k) = y i (k)+a i (k)

[0107] z j|i (k) = y j|i (k)+a j (k)

[0108] where zi represents vehicle i sending its own measurement value to the vehicle cloud, z j|i Represents vehicle j sending the measurement value shared by vehicle i to the vehicle cloud; a i and a j is an item of maliciously manipulated information. When the communication link between the vehicle's sensor and the vehicle cloud is working properly, a i = 0, assuming that the communication link of vehicle 5 is maliciously manipulated, assign

[0109] Step 2: Identify abnormal vehicle behavior based on the vehicle warning scheme. Divide the five sensor measurements into two categories, and select the one with higher credibility for state estimation. The specific steps are as follows:

[0110] The vehicle cloud receives five sensor measurements about vehicle i, including the self-measured value z sent by sensor node i. i and the measurement values ​​z sent by all neighboring nodes j|i , after fusion, {z il (k),l∈N}, where the set N includes all neighboring nodes of sensor node i and itself, and card(N)=5.

[0111] Initialize and define the category set {G ig |i=1,2,3,4,5.;g=1,2.}, where i is the sensor node number and g is the category number. Set the mean vector m (g) , and assign a random initial value.

[0112] Define the distance function between two points:

[0113]

[0114] Defining the Matrix represents the clustering result of each iteration,

[0115]

[0116] In which, define u gl Represents the lth test sample z il (k) Whether it is allocated to category g; if so, u gl =1; otherwise u gl =0.

[0117] Assignment step, the measured value z il (k) To be assigned to category g, the following conditions must be met:

[0118]

[0119] Update step, in order to match the sample measurement data in the current category, the following mean update is required:

[0120]

[0121] Repeat the assignment and update steps until the matrix U does not change.

[0122] Let k (c) is the number of data points belonging to the cth class, defined as follows:

[0123]

[0124] Select the trusted data set of sensor node i as follows:

[0125]

[0126] The g i The data within the class is placed into the confidence set G i In the set G i The data points in the set are considered to be credible data, and the number of elements in the set is k. (gi) ; Put the rest of the data into the collection At this time, the collection It is considered a suspicious set, and the number of elements in the set is 5-k (gi) .

[0127] At the k time node, the suspicious set corresponding to each vehicle sensor can be determined Count the sensors in the five suspicious sets within the K time interval, select sensors with a number greater than 5 / 2, and put them into the set Count the sensors in the five suspicious sets, select sensors with a number greater than 5 / 2, and put them into the set By prompting the technician that the vehicle may be under cyber attack, the driver is advised to intervene in the operating system as soon as possible and make a manual judgment on the current vehicle status.

[0128] Step 3: Eliminate abnormal vehicle data and update the state estimation data set. Specifically: During the process of autonomous driving, once abnormal vehicle data is detected, the data needs to be excluded to maintain the accuracy of the data set. At the same time, the state estimation data set needs to be continuously updated so that autonomous driving vehicles can make more accurate road condition judgments. In the K time interval, by adopting a malicious vehicle early warning scheme, detect whether the data is suspicious, and obtain the safety sensor set Φ. At this time, card(Φ) = φ = 4, remove the abnormal sensor data from the data set, and add the remaining data to the safety data set Ψ. At this time This can effectively prevent abnormal data from misleading autonomous vehicles and improve the safety of autonomous driving systems.

[0129] Step 4: Define the proximity function and state estimation weight parameters, specifically: Define the proximity function

[0130]

[0131] At time k, assuming that there are 4 sensor measurements of vehicle i in the security dataset Ψ obtained by vehicle cloud screening, z iβ (k), (β=1,2,3,4.) means that the two-by-two combinations are calculated The closeness matrix between the measurements is as follows:

[0132]

[0133] The βth measurement value z of vehicle i iβ and any other measurement value z in the security dataset ip The consistency measure of the closeness between (p = 1, 2, 3, 4.) can be expressed as:

[0134]

[0135] Among them, E iβ Reflects the measured value z iβ How close it is to any other sensor measurement.

[0136] In order to obtain a more reasonable weight value, the consistency measure of the entire interval is considered. In the K time interval, the βth measurement value z iβ The mean consistency of closeness to any other measurement in the safety dataset is:

[0137]

[0138] In the K time interval, the βth measurement value z iβ The variance of the consistency of the closeness to any other measurement is:

[0139]

[0140] At this time, if the mean μ iβ is large, and the variance S iβ is smaller, it means that the measured value z iβ It has higher reliability and should have higher weight in the data fusion process. Therefore, this design introduces the intermediate variable α iβ as follows:

[0141]

[0142] In the K time interval, define the sensor measurement value z iβThe weight parameter is ξ iβ as follows:

[0143]

[0144] Then, at time k, the credible measurement value of the i-th vehicle is as follows:

[0145]

[0146] Step 5: Update the state estimate, specifically: based on the measurement value at the current moment, perform state estimation by using the Kalman filter algorithm and the vehicle model.

[0147] Initialize system status and the covariance matrix of the state

[0148] Time update, predict system status in advance:

[0149] Forecast error covariance:

[0150] Measurement update, calculation of Kalman gain coefficient:

[0151] pass Update estimate status:

[0152] Update the error covariance:

[0153] The current state estimate is fed back to the vehicle operating system in real time, and the state at the next moment is predicted in advance for reference.

[0154] After the iterative clustering algorithm is introduced, abnormal vehicle behavior can be effectively identified. At the same time, the setting of the proximity function and the estimation weight parameter improves the accuracy of state estimation. All of these can demonstrate the effectiveness of the safety state estimation method for connected autonomous driving vehicles.

[0155] The examples described above are only for illustrating the technical solutions of the present invention, and are not intended to limit the technical solutions. Although the above examples are described in detail, a person skilled in the art should be aware that the technical solutions described in the above examples can still be modified, or some or all of the technical features thereof can be replaced by equivalents. These modifications or replacements will not cause the essence of the corresponding technical solutions to exceed the scope of the technical solutions of the present invention.

Claims

1. A method for estimating the safety state of a connected autonomous driving vehicle under a network attack, characterized in that: The following steps are involved: Step S1, integrating vehicle data in the connected vehicle area and establishing a connected autonomous driving vehicle model; Step S2: design an iterative clustering algorithm, propose a malicious vehicle warning scheme, and identify abnormal vehicle behavior; Step S3: exclude abnormal vehicle data and update the state estimation data set; Step S4, defining a closeness function and state estimation weight parameters; Step S5: Feedback the estimated value of the safety status at the current moment in real time and estimate the state at the next moment.

2. The method for estimating the safety state of a networked autonomous driving vehicle under a network attack according to claim 1, characterized in that: The specific steps of integrating vehicle data in the Internet of Vehicles area in step S1 include: Step S1.1: At time k, number the n vehicles in the IoV communication area and add them to the set N. Assume that the number of vehicles communicating with the IoV remains unchanged during the K time interval, and all vehicles are fully connected, that is, any two vehicle communication nodes are connected. i (k) (sensor measurement information of the i-th vehicle) is uploaded to the vehicle cloud, and the other vehicle sensor information y shared between vehicles is also uploaded to the vehicle cloud. i|j (k) (sensor measurement information of the jth vehicle, uploaded by the i-th vehicle through vehicle sharing) is uploaded to the vehicle cloud to achieve vehicle information redundancy. This paper assumes that the sensors have the same dimensions.

3. The method for estimating the safety state of a connected autonomous driving vehicle under a network attack according to claim 1, characterized in that: The specific steps of establishing the networked autonomous driving vehicle model in step S1 include: Step S1.2, the vehicle dynamics system model is expressed as: x i (k)=A i x i (k-1)+ω i (k-1) Where A is the system state matrix, ω represents the system noise and its covariance matrix is ​​represented by Q; Step S1.3, the vehicle sensor measurement equation can be expressed as: y i (k)=H i x i (k)+v i (k) Where H is the sensor measurement matrix, v represents the measurement noise and its covariance matrix is ​​denoted by R; Step S1.4: The vehicle sensor measurement model shared between vehicles can be expressed as: and j|i (k)=y i (k) where y j|i Represents the measurement value shared by vehicle i received by vehicle j; Step S1.5: The information about the i-th vehicle received by the vehicle cloud can be expressed as: z i (k)=y i (k)+a i (k) z j|i (k)=y j|i (k)+a j (k) It should be noted that in order to simplify the expression of the formula and improve readability, the time parameter k is omitted in some descriptions. where z i represents vehicle i sending its own measurement value to the vehicle cloud, z j|i Represents vehicle j sending the measurement value shared by vehicle i to the vehicle cloud; a i and a j is an item of maliciously manipulated information. Assuming that the communication link between the sensor of vehicle i and the vehicle cloud is working normally, then a i =0; otherwise, a i ≠0.

4. The method for estimating the safety state of a networked autonomous driving vehicle under a network attack according to claim 1, characterized in that: The specific steps of designing the iterative clustering algorithm in step S2 include: The iterative clustering algorithm includes parameter initialization, allocation step and update step. The goal is to divide the n sensor measurements into two categories and select the one with higher credibility for state estimation. The specific steps are as follows: Step S2.1: At time k, the vehicle cloud receives n sensor measurements of vehicle i, including the self-measurement value z sent by sensor node i. i and the measurement values ​​z sent by all neighboring nodes j|i , after fusion, {z il (k),l∈N}, where the set N includes all neighboring nodes of sensor node i and itself, and card(N)=n. Step S2.2, initialization, define the category set {G ig |i=1,2,...,n.; g=1,2.}, where i is the sensor node label and g is the category label. Step S2.3, initialization, setting the mean vector m (g) , and assign a random initial value. Step S2.4, initialization, define the distance function between two points: Step S2.5, initialization, definition matrix represents the clustering result of each iteration, In which, define u gl Represents the lth test sample z il (k) Is it assigned to category g? If so, then u gl =1; otherwise u gl =0. Step S2.6, allocation step, the measured value z il (k) Assignment to category g requires the following conditions to be met: Step S2.7, update step, in order to match the sample measurement data in the current category, the following mean update is required: Step S2.8: Repeat the allocation and updating steps until the matrix U does not change. Step S2.9: Let k (c) is the number of data points belonging to the cth class, defined as follows: Step S2.10: Select the trusted data set of sensor node i, as shown below: The g i The data within the class is placed into the confidence set G i In the set G i The data points within are considered to be credible data. At this time, the number of elements in the set is Put the rest of the data into the collection At this time, the collection is considered a suspicious set, and the number of elements in the set is 5. The method for estimating the safety state of a networked autonomous driving vehicle under a network attack according to claim 1, characterized in that: The malicious vehicle warning scheme proposed in step S2 is specifically as follows: The iterative clustering algorithm first integrates the sensor measurement data, and uses iterative allocation and updating to divide all the data into two groups. The group with higher confidence is selected for subsequent state estimation, and the other group with lower confidence is considered to be potentially risky. Each data in the set has its own unique label, which can be used to determine which vehicle the data is measured from and which vehicle sensor is used to send it to the vehicle cloud. At each moment, a total of n vehicles upload their own sensor measurement information and the sensor information of n-1 vehicles shared with other vehicles to the cloud. That is, the vehicle cloud will receive n measurement values ​​of sensor i at a time node, and a total of n 2 At each time point, the cloud will determine the suspicious set corresponding to each vehicle sensor based on the iterative clustering algorithm. Count the sensors in the n suspicious sets within the K time interval, select sensors with a number greater than K / 2, and put them into the set Count the sensors in the n suspicious sets, select sensors with a number greater than n / 2, and put them into the set By prompting the technician that the vehicle may be under cyber attack, the driver is advised to intervene in the operating system as soon as possible and make a manual judgment on the current vehicle status.

6. The method for estimating the safety state of a networked autonomous driving vehicle under a network attack according to claim 1, characterized in that: In step S3, abnormal vehicle data is excluded and the state estimation data set is updated as follows: During the process of vehicle autonomous driving, once abnormal vehicle data is detected, it is necessary to exclude the data to maintain the accuracy of the data set. At the same time, it is also necessary to continuously update the state estimation data set so that the autonomous driving vehicle can make more accurate judgments on the road conditions. In the K time interval, by adopting the malicious vehicle warning scheme, detect whether the data is suspicious, and obtain the safety sensor set Φ. At this time, card(Φ) = φ, remove the abnormal sensor data from the data set, and add the remaining data to the safety data set Ψ. At this time This can effectively prevent abnormal data from misleading autonomous vehicles and improve the safety of autonomous driving systems.

7. The method for estimating the safety state of a networked autonomous driving vehicle under a network attack according to claim 1, characterized in that: The specific steps of defining the closeness function and the state estimation weight parameter in step S4 include: Step S4.1, define the closeness function: Among them, the closer A and B are, the greater the closeness. Step S4.2: At time k, assume that there are φ sensor measurements of vehicle i in the security data set Ψ obtained by vehicle cloud screening, and z iβ (k), (β=1,2,...,φ) means that the two-by-two combinations are calculated The closeness matrix between the measurements is as follows: Step S4.3: The βth measurement value z of vehicle i iβ and any other measurement value z in the security dataset ip The consistency measure of the closeness between (p=1,2,...,φ) can be expressed as: Among them, E iβ Reflects the measured value z iβ How close it is to any other sensor measurement. Step S4.4: In order to obtain a more reasonable weight value, consider the consistency measure of the entire interval. In the K time interval, the βth measurement value z iβ The mean consistency of closeness to any other measurement in the safety dataset is: Step S4.5: In the K time interval, the βth measurement value z iβ The variance of the consistency of the closeness to any other measurement is: At this time, if the mean μ iβ is large, and the variance S iβ is smaller, it means that the measured value z iβ It has higher reliability and should have higher weight in the data fusion process. Therefore, this design introduces the intermediate variable α iβ as follows: Step S4.6: Define the sensor measurement value z in the K time interval iβ The weight parameter is ξ iβ as follows: Then, at time k, the credible measurement value of the i-th vehicle is as follows:

8. The method for estimating the safety state of a networked autonomous driving vehicle under a network attack according to claim 1, characterized in that: The specific steps of feeding back the estimated value of the safety status at the current moment in real time and estimating the status at the next moment in step S5 include: It is essential to provide real-time feedback on the estimated safety status at the current moment and estimate the status at the next moment. By knowing the estimated safety status at the current moment, we can take appropriate measures in time to protect vehicle safety. At the same time, the estimation of the status at the next moment can also help us predict risks in advance and provide reliable predictions for future safety assurance. Based on the current measurement value, the state estimation is performed by using the Kalman filter algorithm and the vehicle model. Step S5.1: Initialize system status and the covariance matrix of the state Step S5.2, the time update step is as follows, predicting the system state in advance: Look-ahead forecast error covariance: Step S5.3, the measurement update step is as follows, calculating the Kalman gain coefficient: pass Update estimate status: Update the error covariance: The current state estimate is fed back to the vehicle operating system in real time, and the state at the next moment is predicted in advance for reference.

Images