A network attack detection method for lateral control systems of cloud-controlled intelligent connected vehicles
By using BiLSTM-Attention model and innovative data characteristics in the vehicle-road cloud collaborative system, the problem of difficulty in detecting FDI attacks is solved, the accurate identification and positioning of the vehicle lateral control system is achieved, and the security of intelligent connected vehicles is improved.
Patent Information
- Application Number
- CN202410722923.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-05
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-06-05
AI Technical Summary
In the vehicle-road cloud collaboration system, FDI attacks are difficult to detect, resulting in a threat to vehicle driving safety. Especially due to the random changes in vehicle status information and the obscureness of attack signals, identifying and positioning attack sources is complicated.
An FDI attack detector based on the BiLSTM-Attention model was constructed. By extracting multi-level state information of the vehicle and processing data based on the vehicle dynamics, steering system and motor dynamics models, innovative data features are generated to identify and locate FDI attacks.
It realizes accurate identification of FDI attacks and their locations when the vehicle's driving state changes randomly, improves the driving safety of intelligent connected vehicles, and improves detection accuracy and speed.
Smart Images

Figure CN118540139B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to vehicle-road-cloud collaborative safety control technology, and in particular to a method for detecting network attacks on a lateral control system of a cloud-controlled intelligent networked vehicle. Background Art
[0002] In recent years, with the development of high-tech technologies such as wire-controlled chassis, vehicle networking communications and artificial intelligence, autonomous driving, as a transformative technology in the automotive industry, has become a research hotspot. A series of technologies and methods for autonomous driving have emerged in an endless stream, mainly divided into single-vehicle intelligent solutions that rely on self-vehicle perception, decision-making and control and networked control solutions based on vehicle-road-cloud collaborative systems. In the past few years, researchers have done a lot of research in the field of single-vehicle intelligent control and achieved many important results. However, due to the large number of expensive sensors required in single-vehicle intelligent solutions and the difficulty of the current on-board computing power platform to fully support the computing power requirements of autonomous driving, the commercialization process of single-vehicle intelligent driving solutions has been slow. Therefore, some scholars have proposed a solution to achieve vehicle autonomous driving through vehicle-road-cloud collaborative control. In the vehicle-road-cloud collaborative system, the cloud provides vehicles with weather, map and other information, intelligent roadside equipment provides vehicles with vehicle driving environment information, and the edge cloud platform can provide computing power support for vehicles. Vehicles can communicate with other vehicles, RSUs and the cloud through V2X technology.
[0003] Cloud-based intelligent connected vehicles (CICV) are a typical application of vehicle-road-cloud collaborative solutions. In the CICV architecture, the vehicle perception layer, decision layer, and control layer are placed on the edge cloud. Under normal operation, the vehicle only serves as a physical platform containing sensors and actuators. The vehicle transmits information with the edge cloud through the V2X wireless communication network. The vehicle sends its own vehicle status information to the cloud. The cloud combines the information of intelligent roadside equipment, the status information of all vehicles in the area, and traffic information to make the best decision, plan an optimal driving path for the vehicle, and perform lateral trajectory tracking control calculation based on vehicle dynamics. Finally, the optimal control amount is sent to the vehicle for execution to complete the closed-loop control of CICV. The CICV architecture has obvious advantages on roads equipped with a suitable number of intelligent road test equipment, but since this form of control architecture transmits information through a wireless communication network, there is no relatively closed transmission medium, and it is difficult to monitor the information transmission channel throughout the process, so it is extremely vulnerable to network attacks and interference. In addition, in the vehicle-cloud collaborative network, a large number of vehicle nodes and roadside equipment nodes are connected to the network, and vehicle nodes frequently switch between different regional networks. This networking method also brings more potential network attacks.
[0004] Common network attacks against vehicle systems mainly include denial of service (DOS) attacks and false data injection (FDI) attacks. DoS attacks occupy network channels by sending a large amount of data to the wireless communication network, resulting in the cloud failing to receive the vehicle's status information in a timely manner, and the vehicle failing to receive the cloud's control information in a timely manner. The direct result of DoS attacks is packet loss in the communication network, so it can be easily identified by the waveform change of the transmission signal. FDI attacks use the system information of the attacked to evade intrusion detection, inject attack data into the specified signal, so that the cloud obtains the vehicle status information containing the FDI attack, and the vehicle obtains the control information containing the FDI attack. Since FDI attacks can use the information of the vehicle control system itself to design attack signals, they are extremely hidden, making detection difficult. Moreover, due to the complex and changeable driving conditions of vehicles, the vehicle status information may change randomly, making it difficult to distinguish whether the signal change is caused by the FDI attack or the change in the vehicle's driving state, further increasing the difficulty of detecting FDI attacks. FDI attacks invading the vehicle system will seriously affect the driving safety of the vehicle, disrupt regional traffic order, and pose serious challenges to public safety.
[0005] The present invention designs a network attack detection method for the lateral control system of a cloud-controlled intelligent networked vehicle, in order to accurately identify the attack before the FDI attack causes the vehicle to become unstable, and provide timely decision-making information for the safety control of the vehicle. First, the multi-level state information of the vehicle is extracted, and an original database containing vehicle-level, steering system-level and steering actuator-level signals is established, which is used as the original training data set for the FDI attack detector of the vehicle lateral control system, which can reflect the dynamic characteristics of the vehicle lateral control system at three different levels. Secondly, due to the uncertainty of the vehicle driving conditions, the original data set cannot be directly applied to network training. Therefore, the original data is processed based on the vehicle dynamics model, the steering system dynamics model and the motor dynamics model, and innovative data features that reflect the performance of the vehicle's key systems and the FDI attack characteristics are proposed. Finally, based on the innovative data features, an FDI attack detector based on the BiLSTM-Attention model is developed. Summary of the invention
[0006] In view of the shortcomings in the prior art, the present invention proposes a network attack detection method for the lateral control system of a cloud-controlled intelligent networked vehicle. Considering the situation where the upper trajectory tracking control controller-vehicle channel and the steering controller-execution motor channel in the lateral control system are attacked, by extracting the multi-level state information of the vehicle and processing the data based on the vehicle dynamics model, steering dynamics model and motor dynamics model, and then constructing an FDI attack detector based on the BiLSTM-Attention model, it can accurately and quickly identify FDI attacks and their occurrence locations when the vehicle driving state changes randomly.
[0007] The technical solution adopted by the present invention is as follows:
[0008] A method for detecting network attacks on a lateral control system of a cloud-controlled intelligent connected vehicle, the steps are as follows:
[0009] S1: Build a semi-physical in-the-loop test bench for the lateral control system of a cloud-controlled intelligent connected vehicle, collect vehicle-level signals, steering system-level signals, and steering actuator-level signals under normal working conditions and FDI attack conditions, and construct the original training database of the FDI attack detector network.
[0010] S2: The data in the original training database are classified based on the security of the data itself. The first type of original data has reference values that are not affected by FDI attacks and is defined as direct data. The second type of original data needs to be predicted based on past data and vehicle physical models and is defined as indirect data. The indirect data is processed through the vehicle dynamics model, steering system dynamics model, and motor dynamics model to comprehensively construct innovative data features that include direct data and indirect data.
[0011] S3: A FDI attack detection network based on the BiLSTM-Attention model was constructed and trained with innovative data features to achieve better detection performance.
[0012] Furthermore, the original training database constructed in S1 is as follows:
[0013] X=[X Vehicle ;X System ;X Actuator ]
[0014] Among them, X Vehicle Represents vehicle-level signal, X System Indicates the steering system level signal, X Actuator Represents the steering actuator level signal. Considering that the vehicle is in non-limiting conditions and the tires are in the linear working area, the 2-DOF dynamic model of the vehicle is as follows:
[0015]
[0016] Where m is the vehicle mass, is the lateral acceleration of the vehicle, v x is the vehicle longitudinal velocity, C αf is the front wheel slip angle, C αr is the rear wheel slip angle, δ f is the front wheel turning angle, is the yaw angular acceleration, I z is the moment of inertia of the vehicle about the z-axis, l f and l r is the distance from the center of mass of the vehicle to the front and rear axles, C f is the front wheel cornering stiffness, C r is the front wheel cornering stiffness. From the above formula, we can get the vehicle-level parameters of lateral displacement, lateral velocity, yaw angle, yaw angular velocity and front wheel turning angle. The vehicle-level parameters are used to describe the lateral dynamic characteristics of the vehicle. The existing sensors on the vehicle can accurately measure the corresponding vehicle-level signals. Therefore, the vehicle-level signals are used as the original feature signals for network training.
[0017]
[0018] Where y is the lateral displacement, y ref is the lateral displacement reference value, is the yaw angle, is the yaw angle reference value, is the lateral velocity, is the yaw rate, δ ref is the desired front wheel steering angle.
[0019] The purpose of steering system control is to control the steering motor to accurately and quickly track the desired front wheel angle sent by the upper controller. Steering system parameters i d ,i q ,u d ,u q , w m ,θ m , T m The dynamic characteristics and control performance of the steering motor can be described. Some of the above parameters have fixed reference values, and some can be predicted by the motor dynamics model, which can fully reflect the performance changes of the steering control system under network attacks. The steering system-level signal is established as follows:
[0020] X System =[i d ,i d,ref ,i q ,i q,ref ,u d ,u q ,w m ,wm,ref ,θ m ,θ m,ref ,T m ,T m,ref ]
[0021] Among them, i d is the steering motor d-axis current, i d,ref is the steering motor d-axis reference current, i q is the q-axis current of the steering motor, i q,ref is the q-axis reference current of the steering motor, u d is the d-axis voltage of the steering motor, u q is the q-axis voltage of the steering motor, w m is the steering motor speed, w m,ref is the steering motor speed reference value, θ m is the steering motor angle, θ m,ref is the steering motor speed reference value, T m is the steering motor torque, T m,ref is the steering motor torque reference value.
[0022] The output torque of the steering motor needs to be transmitted to the gear rack through the reducer, and then through the steering rod to push the wheel to steer. The steering column tube angle, speed, gear rack displacement and actual front wheel angle can describe the dynamic performance changes of this force transmission process. Therefore, the steering actuator level signal is established as follows:
[0023]
[0024] where θ p is the steering column tube angle, is the steering column speed, x rack is the gear rack displacement, δ w is the actual front wheel turning angle.
[0025] Furthermore, the direct data in S2 has a reference value that is not affected by FDI attacks. We use the tracking error between the measured value and the reference value to evaluate the security of the system. If the tracking error exceeds a reasonable range, it can be considered that the system has been affected by FDI attacks. The characteristics of the tracking error data are defined as follows:
[0026]
[0027]
[0028] Among them, x err_y is the lateral displacement tracking error, represents the yaw angle tracking error, x err_δ represents the vehicle front wheel angle tracking error, x err_idrepresents the steering motor d-axis current tracking error, x err_iq represents the steering motor q-axis current tracking error, x err_w Represents the steering motor speed tracking error, x err_θ represents the steering motor angle tracking error, x err_Tm Represents the steering motor torque tracking error. Through the electromagnetic torque formula T m,ref =K a i q,ref Obtain the motor torque reference value, K a is the steering motor torque constant.
[0029] For indirect data, the values in the previous control cycle and the system's dynamic model are used to predict the indirect data, and the dynamic residual between the measured value and the predicted value is used to evaluate the security of the system. If the dynamic residual exceeds a reasonable range, it can be considered that the system has been affected by an FDI attack. The characteristics of the dynamic residual data are defined as follows:
[0030]
[0031] in, is the lateral velocity residual, is the yaw rate residual, is the residual voltage of the steering motor d-axis, is the steering motor q-axis voltage residual, is the steering column tube angle residual, is the steering column tube speed residual, is the steering rack displacement residual. pre is the predicted value of lateral displacement, is the predicted value of lateral velocity, is the predicted yaw rate, u d,pre is the predicted value of the steering motor d-axis voltage, u q,pre is the predicted value of the q-axis voltage of the steering motor, θ p,pre is the predicted value of the steering column tube angle, is the predicted value of the steering column tube speed, x rack,pre is the predicted value of steering rack displacement.
[0032] Tracking Error Data Features X err and dynamic residual data features X res The two data reflect the cyber-physical security of CICV lateral control systems from different perspectives. Based on the above two data, an innovative dataset X for FDI attack detector network training is constructed. train ,as follows:
[0033] X train =[X err ,X res ]
[0034] Beneficial effects of the present invention:
[0035] 1. The present invention constructs an original database including vehicle-level, steering system-level and steering actuator-level signals, and uses a vehicle dynamics model, a steering system dynamics model and a motor dynamics model to process the original data in consideration of random changes in vehicle driving status, thereby obtaining innovative data features reflecting the key performance of the vehicle lateral control system and the FDI attack characteristics, which effectively solves the problem that it is difficult to accurately distinguish the impact of random changes in vehicle driving status and FDI attacks on the system.
[0036] 2. The present invention constructs an FDI attack detector based on the BiLSTM-Attention model, which can accurately and quickly identify FDI attacks and accurately locate the channel where the attack occurs, which is of great significance to improving the driving safety of intelligent connected vehicles. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 This is the architecture diagram of the lateral control system of the cloud-controlled intelligent networked vehicle under the FDI attack described in the present invention
[0038] Figure 2 The network training parameter extraction principle diagram of the present invention
[0039] Figure 3 The schematic diagram of the FDI attack detector of the present invention is
[0040] Figure 3 x in n 、w n , They are the input matrix, weight matrix, forward LSTM output, and backward LSTM output.
[0041] Figure 4 The performance comparison diagram of the innovative data features described in the present invention, (a) is the accuracy comparison of the BiLSTM-Attention model driven by the original data and the BiLSTM-Attention model driven by the innovative data features, (b) is the training confusion matrix of the BiLSTM-Attention model driven by the original data, and (c) is the training confusion matrix of the BiLSTM-Attention model driven by the innovative data features
[0042] Figure 5 This is a performance comparison chart of the FDI attack detector based on the BiLSTM-Attention model described in the present invention. DETAILED DESCRIPTION
[0043] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0044] The cloud-controlled intelligent networked vehicle lateral control system described in the present invention is as follows Figure 1 As shown in the figure, in this system, the upper-layer trajectory tracking controller in the cloud transmits information with the vehicle through the wireless communication network. The main control information transmitted is the front wheel angle. The vehicle OBU sends the received front wheel angle information to the CAN network inside the vehicle. The lower-layer steering system controller on the vehicle side obtains the reference front wheel angle value from the CAN network, calculates the current required to track the upper-layer angle, and transmits it to the steering motor through the CAN network to drive the motor to move. In this process, completing a trajectory tracking control requires two stages of information transmission. The first is the wireless channel information transmission between the cloud-based trajectory tracking control controller and the vehicle OBU, which is defined as the cloud-to-vehicle OBU channel (CV); the second is the CAN channel information transmission between the steering system controller and the steering motor, which is defined as the steering controller-to-steering motor (SM). Since the dynamic response of the system is mainly affected by the control information, the situation where the control channels in the two stages are attacked by the network is considered. The reference front wheel angle calculated by the upper-layer trajectory tracking controller in the CV channel is attacked, and the reference current calculated by the lower-layer steering system controller in the SM channel is attacked.
[0045] The original training database includes data of CV channel, SM channel and CV\SM channel under FDI attack in overtaking condition, data of CV channel, SM channel and CV\SM channel under FDI attack in lane changing condition, and data of overtaking and lane changing condition without FDI attack. In order to make the developed FDI attack detector detect the system under network attack and accurately determine in which channel the network attack occurs.
[0046] The innovative data feature extraction principle is as follows Figure 2 As shown, the second type of data needs to be predicted based on the vehicle physical model. The lateral velocity prediction value and the yaw rate prediction value are predicted by the vehicle dynamics model. The vehicle 2-DOF dynamics model is deformed to obtain:
[0047]
[0048] The lateral velocity and yaw rate at the current moment can be predicted based on the lateral velocity and yaw rate at the previous moment, as follows:
[0049]
[0050] In the formula, k represents the current control moment, k-1 represents the previous control moment, and Δt represents the time difference between the current control moment and the previous control moment. The predicted value of the steering motor d-axis voltage and the predicted value of the steering motor q-axis voltage are solved by the motor dynamics model as follows:
[0051] u d,pre (k) = R s i d (k)-w m L q i q (k)
[0052] u q,pre (k) = R s i q (k)+w m (L d i d (k)+Φ)
[0053] Among them, R s is the equivalent winding resistance, Φ is the magnetic flux of the permanent magnet, L d and L q They are the steering motor d-axis and q-axis inductances respectively.
[0054] Solve the steering column tube angle prediction value θ through the steering transmission mechanism dynamics model p,pre , Steering column tube speed prediction value and steering rack displacement prediction value x rack,pre ,as follows:
[0055]
[0056] Among them, r p is the radius of the steering column tube, Can be measured by sensors, Calculated by the following formula:
[0057]
[0058] Where, T m is the steering motor torque, M f is the steering resistance torque, J p is the moment of inertia of the steering column tube, B p It is the damping of the steering column tube.
[0059] The principle of the FDI attack detection network based on the BiLSTM-Attention model is as follows Figure 3As shown in the figure, the innovative data features containing direct data and indirect data are input into the BiLSTM-Attention model for training, and the FDI attacks of the CV channel, SM channel and CV\SM channel under overtaking conditions, the FDI attacks of the CV channel, SM channel and CV\SM channel under lane changing conditions, and normal conditions are classified, so that the established FDI attack detection network can identify FDI attacks and the channels where they occur in the typical turning conditions of the vehicle. The outputs of the FDI attack detection network are: attacks on the CV channel, attacks on the SM channel, attacks on the CV\SM channel, and no attacks on both channels.
[0060] Figure 4 The effectiveness of the data processing method in the network attack detection method of the lateral control system of a cloud-controlled intelligent networked vehicle of the present invention is demonstrated. Figure 4 It can be seen that the training set accuracy of the BiLSTM-Attention network driven directly by the original data is 85.45%, and its confusion matrix shows that there are 5394 data classification errors in 38401 training data. The training set accuracy of the BiLSTM-Attention network driven by the innovative data of the present invention is 92.05%, with only 2445 data classification errors. This proves the effectiveness of the data processing method proposed in the present invention in improving detection accuracy.
[0061] Figure 5 The performance comparison results of the FDI detector based on the BiLSTM-Attention model and the FDI attack detector based on LSTM and SVP are shown. Figure 5 It can be seen that when both are trained with innovative data features, the accuracy of the LSTM-based FDI detector is 82.71%, the accuracy of the SVP-based FDI detector is 86.57%, and the accuracy of the BiLSTM-Attention-based FDI detector of the present invention is 92.05%, and its detection accuracy is better than the two conventional methods. This proves the superiority of the FDI detector based on the BiLSTM-Attention model in the present invention in dealing with the FDI attack detection problem.
[0062] The present invention has many specific application paths. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the principle of the present invention. These improvements should also be regarded as the protection scope of the present invention.
[0063] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0064] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.
Claims
1. A method for detecting network attacks on a lateral control system of a cloud-controlled intelligent connected vehicle, characterized in that: The following steps are involved: S1: Build a semi-physical in-the-loop test bench for the lateral control system of a cloud-controlled intelligent connected vehicle, collect vehicle-level signals, steering system-level signals, and steering actuator-level signals under normal working conditions and FDI attack conditions, and build the original training database of the FDI attack detector network; S2: Classify the data in the original training database based on the security of the data itself. The first type of original data has reference values that are not affected by FDI attacks and is defined as direct data. The second type of original data needs to be predicted based on past data and vehicle physical models and is defined as indirect data. The indirect data is processed through the vehicle dynamics model, steering system dynamics model, and motor dynamics model to comprehensively construct innovative data features that include direct data and indirect data. S3: Build an FDI attack detection network based on the BiLSTM-Attention model and train it with innovative data features to achieve better detection performance.
2. The method according to claim 1, characterized in that The original training database constructed in S1 is as follows: X=[X Vehicle ;X System ;X Actuator ] Among them, X Vehicle Represents vehicle-level signal, X System Indicates the steering system level signal, X Actuator Represents the steering actuator level signal. Considering that the vehicle is in non-limited working conditions and the tires are in the linear working area, the 2-DOF dynamic model of the vehicle is as follows: Where m is the vehicle mass, is the lateral acceleration of the vehicle, v x is the vehicle longitudinal velocity, is the yaw rate, C αf is the front wheel slip angle, δ f is the front wheel turning angle, is the lateral velocity, C αr is the rear wheel slip angle, is the yaw angular acceleration, I z is the moment of inertia of the vehicle about the z-axis, l f and l r is the distance from the center of mass of the vehicle to the front and rear axles, C f is the front wheel cornering stiffness, C r is the front wheel cornering stiffness. From the above formula, the vehicle-level parameters of lateral displacement, lateral velocity, yaw angle, yaw angular velocity and front wheel turning angle can be obtained. The vehicle-level parameters are used to describe the lateral dynamic characteristics of the vehicle. The existing sensors on the vehicle can accurately measure the corresponding vehicle-level signals. Therefore, the vehicle-level signals are used as the original feature signals for network training. Among them, X Vehicle is the vehicle-level signal, y is the lateral displacement, y ref is the lateral displacement reference value, is the yaw angle, is the yaw angle reference value, δ ref is the desired front wheel turning angle; The purpose of steering system control is to control the steering motor to accurately and quickly track the desired front wheel angle sent by the upper controller. The steering system parameter i d ,i q ,u d ,u q , w m ,θ m , T m The dynamic characteristics and control performance of the steering motor can be described. Some of the above parameters have fixed reference values, and some can be predicted by the motor dynamics model. It can fully reflect the performance changes of the steering control system under network attacks. The steering system-level signal is established as follows: X System =[i d ,i d,ref ,i q ,i q,ref ,u d ,u q ,w m ,w m,ref ,θ m ,θ m,ref ,T m ,T m,ref ] Among them, X System is the steering system level signal, i d is the steering motor d-axis current, i d,ref is the steering motor d-axis reference current, i q is the q-axis current of the steering motor, i q,ref is the q-axis reference current of the steering motor, u d is the steering motor d-axis voltage, u q is the q-axis voltage of the steering motor, w m is the steering motor speed, w m,ref is the steering motor speed reference value, θ m is the steering motor angle, θ m,ref is the steering motor speed reference value, T m is the steering motor torque, T m,ref is the steering motor torque reference value; The output torque of the steering motor is first transmitted to the gear rack through the reducer, and then through the steering rod, which pushes the wheel to steer. The dynamic performance changes of this force transmission process are described by the steering column tube angle, speed, gear rack displacement and actual front wheel angle. Therefore, the steering actuator level signal is established as follows: Among them, X Actuator is the steering actuator level signal, θ p is the steering column tube angle, is the steering column speed, x rack is the gear rack displacement, δ w is the actual front wheel turning angle.
3. The method according to claim 2, characterized in that In S2, the original data is classified into direct data and indirect data. The direct data has a reference value that is not affected by FDI attacks. The tracking error between the measured value and the reference value is used to evaluate the security of the system. The characteristics of the tracking error data are defined as follows: x err_δ =|δ w (k)-δ ref (k)|,x err_id =|i d (k)-i d,ref (k)| x err_iq =|i q (k)-i q,ref (k)|,x err_w =|w m (k)-w m,ref (k)| Among them, X err is the tracking error data feature, x err_y is the lateral displacement tracking error, represents the yaw angle tracking error, x err_δ represents the vehicle front wheel angle tracking error, x err_id represents the steering motor d-axis current tracking error, x err_iq represents the steering motor q-axis current tracking error, x err_w Represents the steering motor speed tracking error, x err_θ represents the steering motor angle tracking error, x err_Tm Represents the steering motor torque tracking error, through the electromagnetic torque formula T m,ref =K a i q,ref Obtain the motor torque reference value, K a is the steering motor torque constant; For indirect data, the values in the previous control cycle and the system's dynamic model are used to predict the indirect data. The dynamic residual between the measured value and the predicted value is used to evaluate the safety of the system. The characteristics of the dynamic residual data are defined as follows: Among them, X res is the dynamic residual data feature, is the lateral velocity residual, is the yaw rate residual, is the steering motor d-axis voltage residual, is the steering motor q-axis voltage residual, is the steering column tube angle residual, is the steering column tube speed residual, is the steering rack displacement residual, the k in brackets represents the current control moment, and y pre is the predicted value of lateral displacement, is the predicted value of lateral velocity, is the predicted yaw rate, u d,pre is the predicted value of the steering motor d-axis voltage, u q,pre is the predicted value of the q-axis voltage of the steering motor, θ p,pre is the predicted value of the steering column tube angle, is the predicted value of the steering column tube speed, x rack,pre is the predicted value of steering rack displacement.
4. The method according to claim 3, characterized in that The method for predicting indirect data using the values in the previous control period and the system's dynamic model is as follows: The lateral velocity prediction value and the yaw rate prediction value are predicted by the vehicle dynamics model. The vehicle 2-DOF dynamics model is deformed to obtain: The lateral velocity and yaw rate at the current moment are predicted by the lateral velocity and yaw rate at the previous moment, as follows: In the formula, k represents the current control moment, k-1 represents the previous control moment, Δt represents the time difference between the current control moment and the previous control moment, and the steering motor d-axis voltage prediction value and the steering motor q-axis voltage prediction value are solved by the motor dynamics model as follows: u d,pre (k)=R s i d (k)-w m L q i q (k) u q,pre (k)=R s i q (k)+w m (L d i d (k)+Φ) Among them, R s is the equivalent winding resistance, Φ is the magnetic flux of the permanent magnet, L d and L q They are the steering motor d-axis and q-axis inductances respectively; Solve the steering column tube angle prediction value θ through the steering transmission mechanism dynamics model p,pre , Steering column tube speed prediction value and steering rack displacement prediction value x rack,pre ,as follows: x rack,pre (k)=θ p,pre (k)r p Among them, r p is the radius of the steering column tube, Measured by the sensor, Calculated by the following formula: Where, T m is the steering motor torque, M f is the steering resistance torque, J p is the moment of inertia of the steering column tube, B p It is the damping of the steering column tube.
5. The method according to any one of claims 1 to 4, characterized in that: It also includes constructing the tracking error data feature X err and dynamic residual data features X res The two data reflect the cyber-physical security of the lateral control system of cloud-controlled intelligent connected vehicles from different perspectives, and construct an innovative dataset X for FDI attack detector network training based on the two data. train ,as follows: X train =[X err ,X res ]。
Citation Information
Patent Citations
Vehicle queue event trigger control method under wrong data injection attack
CN116800501A
Internet of vehicles intrusion detection method and system based on attention CNN-BiLSTM algorithm
CN117955704A
Cited By
Intelligent vehicle elastic planning and control method based on large language model guidance under network attack
CN121234752A
