Intelligent automobile risk slow release control method and system in sensitive message attack scene
By introducing gated loop unit network, fuzzy logic rules, security model prediction control and preset shared key tracking control in the intelligent car autonomous driving system, the information security problems caused by sensitive message attacks are solved and the system protection is improved Capability and safety prevent potential car crashes.
Patent Information
- Application Number
- CN202410875650.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-02
- Publication Date
- 2024-11-12
AI Technical Summary
The intelligent car autonomous driving system has unreliable perception status, unreasonable decision-making, wrong path planning, and unsafe control laws in sensitive message attack scenarios, resulting in insufficient information security protection and may cause serious security threats and accidents.
Adopt the state estimation module based on the gated recurrent unit network, the decision-making module based on fuzzy logic rules, the path planning module based on the security model prediction control, and the tracking control module based on the preset shared key to improve the information security protection capability of the autonomous driving system. , enhance protection capabilities from four aspects: perception, decision-making, planning, and control, and resist illegal attacks from sensor channels and actuator channels.
Effectively improve the information security protection capabilities of smart cars in sensitive message attack scenarios, avoid perception layer abnormalities, decision-making errors and path planning errors, ensure that the actuator can safely receive control instructions, reduce the risk of accidents, and improve the Safety and reliability of autonomous driving functions.
Smart Images

Figure CN118938730A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information security of smart cars, and in particular to a risk mitigation control method and system for smart cars in a sensitive message attack scenario. Background Art
[0002] The global technological development with intelligence and networking as important features is bringing major changes to the automotive industry. Smart cars rely on artificial intelligence, combined with vision, radar and global positioning systems to work together, allowing cars to drive automatically without human active operation. The autonomous driving system is usually implemented in a layered architecture, including the perception layer, decision layer, planning layer and control layer. The autonomous driving system is the core of the smart car. The perception layer is responsible for receiving data from the sensors of the smart car, integrating and processing to understand the environment around the car and the state of the car itself; the decision layer formulates the driving mode and driving strategy of the smart car based on the information output by the perception layer, including: path selection, behavior decision, interactive decision, etc.; the planning layer combines the decision results to further carry out path planning and speed planning, and plans a specific driving path and speed curve; the control layer is responsible for converting the path and speed curve generated by the planning layer into specific control instructions such as car acceleration, braking, steering, etc., and sends them to various actuators for execution. The four core layers are independent of each other and work together to realize the autonomous driving function of smart cars.
[0003] Smart cars are complex heterogeneous network information-physical systems. After the in-car network of smart cars is interconnected, the electronic control units related to the autonomous driving system are at risk of being attacked maliciously. Illegal attacks on sensitive status messages in the sensor channel will seriously interfere with the state perception of the perception layer of the autonomous driving system, further causing incorrect decisions and path planning at the decision-making and planning layers, and inducing serious accidents. Illegal attacks on sensitive control messages such as acceleration, braking, and steering in the actuator channel can directly cause the actuator to fail to correctly receive control instructions, or directly maliciously manipulate the throttle mechanism, steering mechanism, etc. to cause illegal control of the car, causing malicious incidents.
[0004] As the functions of the autonomous driving system of smart cars are gradually improved, cars can well complete autonomous driving tasks in various complex road scenarios. However, there are still many problems in information security protection. Specifically, in-vehicle network security protection measures such as data encryption and decryption, identity authentication, and intrusion detection provide certain protection for the information security of smart cars. However, once hackers break through the in-vehicle network security protection and attack the sensitive messages of the autonomous driving system, the autonomous driving system has almost no information security protection capabilities. The information security risks further induce functional abnormalities of the autonomous driving system, resulting in unreliable perception of the autonomous driving system, unreasonable decisions, wrong path planning, and control laws that cannot be safely received in sensitive message attack scenarios. The autonomous driving system is the core part of the smart car and is directly related to the safe operation and driving of the car. Information security protection cannot be completely guaranteed by the in-vehicle network security. Insufficient information security protection of the autonomous driving system may lead to serious security threats. Summary of the invention
[0005] In order to solve the problems that the existing intelligent vehicle automatic driving system has unreliable perception status, unreasonable decision-making, path planning errors and control laws that cannot be safely received in sensitive message attack scenarios, the present invention provides an intelligent vehicle risk mitigation control method and system in sensitive message attack scenarios.
[0006] The objective of the present invention is achieved through the following technical solutions:
[0007] The present invention is based on the realization of the autonomous driving system of the layered architecture, and improves the information security protection capability of the autonomous driving system of the intelligent vehicle from the four aspects of perception, decision-making, planning and control, thereby improving the safety and reliability of the intelligent vehicle. Specifically, by improving or additionally providing the following parts: a state estimation module based on a gated recurrent unit network, a decision module based on fuzzy logic rules, a path planning module based on security model predictive control, and a tracking control module based on a preset shared key.
[0008] The sensitive state messages of the sensor channels during normal driving of the car are collected to construct a sensitive message data set, and the gated recurrent unit network model is used to establish the state perception prediction under sensor post-fusion. During normal driving, the perception layer uses the multi-sensor fusion module based on the Kalman filter to fuse the information of sensors such as cameras, lidar, and global positioning systems to estimate the state and obtain reliable states such as speed, position, and heading. In the sensitive message attack scenario, the multi-sensor fusion module based on the Kalman filter processes illegal sensitive messages, and the obtained state data is no longer reliable. The gated recurrent unit network has unique advantages in time series data prediction, and the prediction speed is fast and the accuracy is high. The current reliable state estimation output is obtained by using the pre-trained gated recurrent unit network and reliable historical sensitive message prediction, avoiding the perception layer abnormality caused by illegal sensitive message attacks, improving the reliability of the perception layer state data, and improving the ability of the autonomous driving system to resist illegal attacks by sensor channel sensitive messages.
[0009] Furthermore, the decision-making module based on fuzzy logic rules establishes fuzzy logic rules and membership functions according to factors that may affect the driving strategy of the car, such as the degree of danger of the scene and the intensity of the sensitive message attack, and then decides on the driving mode and strategy of the intelligent car. The fuzzy logic system has strong flexibility and has a good ability to deal with uncertainty and ambiguity. The degree of impact of sensitive message attacks on the automatic driving system of intelligent cars cannot be accurately described. The use of fuzzy logic rules can flexibly take sensitive message attack factors into account in the decision-making process, improve the safety and reliability of emergency response decisions of intelligent cars in sensitive attack scenarios, and achieve more reasonable driving strategy decisions.
[0010] Furthermore, the impact of sensitive message attacks on the autonomous driving system of intelligent vehicles is described by network risk factors, and the network risk factors are introduced into the path planning process to realize local path planning based on security model predictive control. The model predictive control method is suitable for processing multi-input and multi-output systems and can effectively handle multiple variables and constraints. By improving the model predictive control method, the network risk factor is introduced into the penalty function for correction, so that the model predictive control method can simultaneously optimize the network risk item in real time during the rolling optimization solution process to correct the erroneous path planning caused by sensitive message attacks.
[0011] Furthermore, by updating encrypted communication, the communication channel between the control layer and the actuator is secured, ensuring that the actuator can safely receive and execute the correct control law. Updating encrypted communication increases the difficulty of the actuator channel being continuously attacked by sensitive messages, ensuring that the smart car will not be continuously attacked for a long time, and the actuator can receive the control instructions of the control layer, alleviating the impact on the car's handling performance, and improving the game confrontation ability of the smart car's automatic driving function under sensitive attacks.
[0012] The beneficial effects of the intelligent vehicle risk mitigation control method and system in a sensitive message attack scenario provided by the present invention are as follows:
[0013] 1. The present invention provides a risk mitigation control method for smart cars in sensitive message attack scenarios. It is implemented based on a typical autonomous driving layered architecture and realizes information security protection by improving or additionally providing information security reinforcement methods. It has good practicality and comprehensively improves the information security protection capabilities of smart car autonomous driving systems from four aspects: perception, decision-making, planning, and control.
[0014] 2. The present invention provides a risk mitigation control system for intelligent vehicles in a sensitive message attack scenario, which provides four core contents: a state estimation module based on a gated cyclic unit network, a decision module based on fuzzy logic rules, a path planning module based on security model predictive control, and a tracking control module based on a preset shared key. The state estimation module based on the gated cyclic unit network avoids the abnormality of the perception module and the unreliable perception state caused by sensitive message attacks; the decision module based on fuzzy logic rules implements the attack mitigation process to improve the safety, reliability and rationality of the emergency response decision of the intelligent vehicle; the path planning module based on security model predictive control realizes real-time optimization of network risks to correct the erroneous path planning caused by sensitive message attacks; the tracking control module based on a preset shared key ensures that the actuator can receive the control instructions of the central computing unit, and improves the game confrontation ability of the automatic driving function of the intelligent vehicle under sensitive attacks.
[0015] 3. The present invention provides a risk mitigation control method and system for intelligent vehicles in sensitive message attack scenarios, which realize information security protection of intelligent vehicle autonomous driving systems in scenarios of attacks by sensitive state messages such as speed, position, and heading, and attacks by sensitive control messages such as acceleration, braking, and steering, and effectively avoids functional abnormalities such as longitudinal risks and lateral instability of the automatic driving system induced by information security risks, thereby preventing possible serious accidents such as car collisions. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art; obviously, the drawings in the following description are merely implementation cases of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without creative work.
[0017] Figure 1 A flow chart of a smart car risk mitigation control method in a sensitive message attack scenario provided by an embodiment of the present invention;
[0018] Figure 2 A schematic diagram of a smart car risk mitigation control system in a sensitive message attack scenario provided by an embodiment of the present invention;
[0019] Figure 3 A schematic diagram of a smart car risk mitigation control method and system in a sensitive message attack scenario provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0020] In order to make the above-mentioned objects, features and advantages of the present invention more clearly understood, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. 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.
[0021] Embodiment 1:
[0022] like Figure 1 As shown, the present embodiment provides a flow chart of a smart car risk mitigation control method in a sensitive message attack scenario, and the specific implementation steps are as follows:
[0023] Step 1: Use the gated recurrent unit network model to realize the state estimation compensation under sensor post-fusion. Specifically, collect the sensitive state messages of the sensor channel when the car is driving normally to build a sensitive message data set, and obtain the gated recurrent unit network model that meets the prediction accuracy requirements by training the model of the gated recurrent unit network. Optionally, a central gateway or data acquisition device can be used to collect sensitive messages.
[0024] Step 2: Establish rules and membership functions based on factors that affect the safe driving strategy of smart cars. In this embodiment, the scene danger level and network attack intensity are selected as factors that affect fuzzy logic decision-making. In this embodiment, the scene danger level is based on specific scenes such as unmanned roads, highways, ordinary roads, intersections, etc., and a one-to-one corresponding danger level is established, which is defined as four situations: non-dangerous, slightly dangerous, moderately dangerous, and highly dangerous. In this embodiment, the intensity of network attacks is defined as four situations: non-dangerous, slightly dangerous, moderately dangerous, and highly dangerous, based on the range of network risk factors expressed quantitatively. In this embodiment, the decision results of fuzzy logic are defined as three driving strategies: normal driving, attack mitigation, and emergency parking. Optionally, the factors that affect decision-making, the division of degree levels, and the results of decision-making in the process of establishing fuzzy logic rules are not limited to those described in this embodiment.
[0025] Step 3: Design path planning based on the security model predictive control, modify the penalty function through the network risk factor, and optimize the network risk in real time. In this embodiment, the network risk factor is quantitatively described by the deviation between the actual state and the predicted state in the three dimensions of horizontal, vertical and heading. Specifically, the network risk factor is calculated through the following steps:
[0026] At time t, the prediction time domain N is obtained through the automobile prediction model p Prediction information of vehicle status pred (t,t+N p ). At time t+1, the real-time vehicle status information ξ input by the sensor is collected sensor (t+1), and calculate the difference between the predicted state quantity and the actual state quantity in real time as the state evaluation matrix: ξ e (t+1)=ξ pred (t+1)-ξ sensor (t+1). The state evaluation matrix ξ e (t+1) According to the vehicle state dimension, it is decomposed into ξ from the three dimensions of lateral, longitudinal and heading. ex , ey and At the same time, the three state evaluation matrices ξ ex , ey and Different weights are assigned to each internal state element and normalized to obtain three network risk factors (Network Risk Index, NRI) in the horizontal, vertical and heading directions. The specific expression is as follows: NRI i =k1ξ ei (1)+k2ξ ei (2)…+k j ξ ei (j) where i represents x, y and etc. three dimensions; j = 1, 2…n is the number of state elements; ξ ei (j) is an element of the state evaluation matrix. In this embodiment, only the horizontal, vertical and heading dimensions are selected, and the vertical evaluation matrix ξ ex Select the element x in e , Horizontal evaluation matrix ξ ey Select the element in y e , etc., heading assessment matrix Select elements in Etc. Optionally, the selection of dimensions in the implementation process is diverse, and may also include other elements in other dimensions, k j The weights of the corresponding elements in risk assessment can be optimized according to actual needs.
[0027] In this embodiment, the network risk factor is introduced into the penalty function, including: obstacle avoidance penalty function, road boundary penalty function and collision mitigation penalty function. Specifically, the obstacle avoidance penalty function is expressed in the following form:
[0028]
[0029] In the formula, Δv i is the speed of the obstacle relative to the vehicle; (X, Y) is the position coordinate of the vehicle in the Cartesian coordinate system; (xi, yi) is the position coordinate of the ith obstacle in the Cartesian coordinate system; D s It is the safe braking distance of the vehicle under extreme conditions.
[0030] Specifically, the road boundary penalty function is expressed as follows:
[0031]
[0032] Where, L s is the lateral safety distance between the vehicle and the road boundary; v is the vehicle speed. Specifically, the collision mitigation penalty function is expressed in the following form:
[0033]
[0034] J PCSI,i =PCSI i (Δv)+PCSI i (θ)+PCSI i (m)
[0035] In the formula, m i is the mass of the obstacle; θ is the direction angle between the vehicle and the obstacle.
[0036] In path planning based on safety model predictive control, factors such as road traffic constraints, actuator performance constraints, and tire friction circle constraints need to be taken into account, and the local path planning problem is converted into a constrained multi-objective optimization problem, which is expressed through a cost function. In the process of solving the rolling optimization, the penalty function is used to guide the optimal solution of the cost function toward the direction of vehicle functional safety and information security.
[0037] Step 4: The communication channel between the autonomous driving system and the actuator is secured by updating the encrypted communication, ensuring that the actuator can safely receive and execute the correct control law. In this embodiment, a symmetric encryption method with dynamically updated keys is used to implement secure communication. Once the decision result enters the attack mitigation state, the communicating parties actively update the keys based on the pre-made shared keys.
[0038] Embodiment 2:
[0039] like Figure 2As shown, it is a schematic diagram of a risk mitigation control system for an intelligent vehicle in a sensitive message attack scenario provided by this embodiment. It is implemented based on the hierarchical architecture of the autonomous driving system, and specifically includes four core parts: a perception layer, a decision layer, a planning layer, and a control layer. The sensor inputs information into the autonomous driving system through the sensor channel of the CAN bus network, and the autonomous driving system outputs control instructions to the sensor through the actuator channel. Different from the typical hierarchical architecture system, the present invention additionally provides or improves the following parts: a state estimation module based on a gated recurrent unit network, a decision module based on fuzzy logic rules, a path planning module based on security model predictive control, and a tracking control module based on a preset shared key. The specific working principle is as follows:
[0040] The sensors deployed in the smart car input the sensor information data stream to the autonomous driving system through the CAN bus network. In this embodiment, the sensors include: lidar, camera, inertial navigation, millimeter wave radar and global positioning system. In the sensor channel of the CAN bus network, there are ordinary messages and sensitive status messages. Attackers may launch sensitive message attacks through the sensor channel. In this embodiment, whether it is attacked by sensitive messages is determined by network risk factors. Optionally, the sensors used in the autonomous driving system are not limited to those described in this embodiment; whether it is attacked or not can be determined by other intrusion detection systems.
[0041] The perception layer obtains the information data stream of the sensor, realizes multi-sensor fusion through the Kalman filtering method, and then obtains sensitive states such as speed, position, heading, and other information through fusion. Obviously, the sensitive state is reliable during normal driving and can be output to the decision-making layer normally and reliably. If the sensor channel is attacked by sensitive messages, it is reasonable to believe that the sensitive state obtained through the sensor channel is not reliable enough. The state estimation module based on the gated recurrent unit provided by the present invention obtains reliable historical data of the sensor channel and uses a pre-trained gated recurrent unit network to achieve compensation output during the attack period as a reliable output of the sensitive state of the perception layer.
[0042] The decision-making layer combines information such as driving scenes, vehicle status, and input from the perception layer to make behavioral decisions and interactive decisions to determine the subsequent driving behavior and driving mode of the smart car. Specifically, it includes but is not limited to: normal driving, overtaking, following, parking, and emergency parking. In this embodiment, the decision-making module based on fuzzy logic rules provided by the present invention determines the subsequent driving behavior and driving mode of the smart car through fuzzy logic rules established by the intensity of network attacks and the degree of danger of the scene. If the decision result is attack mitigation, the subsequent path planning module based on security model predictive control and the tracking control module based on the preset shared key are triggered. If the decision result is normal driving, the normal automatic driving function continues to be executed. Specifically, in this embodiment, the intensity of network attacks and the degree of danger of the scene are obtained according to Example 1.
[0043] The planning layer includes two parts: path planning and speed planning. In this embodiment, the path planning module based on safety model predictive control has the same working principle as the path planning based on safety model predictive control design in step 3 of Example 1. The penalty function is corrected by the network risk factor. In the process of implementing the rolling optimization solution, the penalty function is used to guide the optimal solution of the cost function toward the direction of automobile functional safety and information security.
[0044] The control layer calculates the corresponding acceleration, braking, and steering control laws based on the path planning and speed planning results of the planning layer, and performs lateral and longitudinal control of the smart car. The control law calculated by the control layer sends sensitive control instructions to the corresponding actuator through the CAN bus network. In this embodiment, the actuator includes: a throttle mechanism, a braking mechanism, an emergency braking mechanism, and a steering mechanism. The actuator channel of the CAN network includes ordinary messages and sensitive control messages. Attackers can launch sensitive message attacks through the actuator channel to affect the stable operation of the smart car. In this embodiment, the tracking control module based on the preset shared key provided by the present invention performs symmetric encryption communication by presetting a shared key in the automatic driving system and the actuator. Once it is attacked by a sensitive message, the key is updated to ensure that the actuator can safely receive the control instructions and alleviate the instability of the car caused by illegal operation. Optionally, the actuator used in the automatic driving system is not limited to that described in this embodiment.
[0045] In this article, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0046] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method and system for risk mitigation control of smart cars in sensitive message attack scenarios, It is characterized in that Applied to the autonomous driving system under the layered architecture of intelligent vehicles, the described method includes: The state estimation module based on the gated recurrent unit network uses the compensation output of the neural network to replace the unreliable state obtained under the sensitive message attack, avoiding system abnormalities caused by sensitive message attacks on the sensor channel; The decision-making module based on fuzzy logic rules drives the attack mitigation process by establishing fuzzy logic rules for driving mode decision-making; The path planning module based on security model predictive control corrects the incorrect path planning caused by sensitive message attacks by optimizing the network risk item of the cost function in real time. The tracking control module based on the preset shared key ensures that the actuator can safely receive control instructions by updating the symmetric key for communication between the autonomous driving system and the actuator.
2. The method according to claim 1, It is characterized in that The described gated recurrent unit network-based state estimation module uses a pre-trained gated recurrent unit network and reliable historical sensitive state messages of the sensor channel to make predictions and obtain a current reliable state estimation output.
3. The method according to claim 1, It is characterized in that The described decision module based on fuzzy logic rules establishes fuzzy logic rules and membership functions according to the danger level of the scenario and the attack strength of the sensitive message; Optionally, factors for establishing fuzzy logic decisions include, but are not limited to, the intensity of the cyber attack and the degree of danger of the scenario.
4. The method according to claim 1, It is characterized in that The path planning module based on the safety model predictive control described above introduces the network risk factor to modify the penalty function, so that the model predictive control method can optimize the network risk item in real time during the rolling optimization solution process, alleviate the impact of sensitive message attacks on the driving of smart cars, and realize safe path planning; Optionally, modifications to the penalty function include but are not limited to an obstacle avoidance penalty function, a collision mitigation penalty function, and a road boundary penalty function.
5. The method according to claim 1, It is characterized in that The tracking control module based on the preset shared key ensures the information security of the autonomous driving system through symmetric encryption communication and key update; the preset shared key reduces the additional delay and message increase caused by the key update.
Citation Information
Cited By
Method for constructing safe driving space of automatic driving BEV
CN119659667A
An automatic driving BEV safe driving space construction method
CN119659667B
