A vehicle trust evaluation method and device based on double-potential well field entropy flow annihilation
By constructing a dynamic dual-potential-well coupled field and introducing an entropy annihilation mechanism, the vehicle trust assessment method solves the problem of entropy response lag in vehicle trust assessment, achieves high accuracy and high recall rate for identifying malicious and colluding vehicles, and improves the security of vehicle networks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- RONGTONG RESOURCES ANHUI CO LTD
- Filing Date
- 2026-02-12
- Publication Date
- 2026-06-19
Smart Images
Figure CN122248419A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of vehicle-to-everything (V2X) security technology, and in particular to a dynamic trust assessment method and apparatus for evaluating the trust value of vehicle nodes and the authenticity of messages received by vehicles in a V2X environment. Background Technology
[0002] With the rapid development of the Internet of Vehicles (IoV), vehicles improve travel efficiency and road safety by exchanging traffic event and environmental perception information. However, the IoV's openness, dynamism, and high degree of collaboration make it vulnerable to attackers who may inject false information or manipulate malicious nodes to undermine the authenticity and reliability of information exchange. Therefore, conducting credibility assessments of vehicle-reported events and establishing a scientific vehicle trust assessment mechanism have become crucial for ensuring IoV security.
[0003] Scholars both domestically and internationally have proposed various vehicle trust assessment methods, which have improved the accuracy and robustness of assessments to some extent. However, existing methods still have limitations, and insufficient accuracy and robustness remain core challenges facing the industry. The industry generally believes that trust assessment mechanisms based on potential well models and information entropy can characterize the evolution of vehicle trust and improve assessment accuracy. However, this method has a significant problem: "entropy fatigue." Theoretically, trust entropy can freely increase or decrease with external evidence, but in actual iterations, each change in entropy leaves a small and irreversible "plastic deformation" in the vehicle trust state. As the number of iterations increases, these deformations gradually accumulate, making the trust entropy's response to new evidence increasingly sluggish: the system can quickly reflect behavioral changes in the early stages, but after long-term operation, the entropy response lags, leading to untimely responses to new malicious behaviors or normal corrective behaviors, thus affecting the overall accuracy and robustness of trust assessment. Summary of the Invention
[0004] To address the aforementioned issues, this invention provides a vehicle network trust assessment method and device based on entropy flow annihilation in a dual-potential-well field. The core idea is to construct a vehicle trust assessment mechanism based on the synergy of a dynamic dual-potential-well coupled field and an entropy annihilation mechanism. First, the dynamic dual-potential-well coupled field forms the system's elastic recovery framework. The consensus potential well guides node states towards global consensus by establishing a potential energy landscape based on the distribution of group trust, thus releasing stress from the accumulated deformation of entropy values. The individual stability potential well, centered on the historical trust average of vehicles, effectively suppresses irrational fluctuations in trust values by maintaining the stability of individual behavior patterns. Both provide structural protection against fatigue degradation. Simultaneously, an entropy annihilation mechanism is introduced as an active repair strategy to address the entropy response lag problem. This mechanism, by designing an annihilation term proportional to the entropy inflow, selectively eliminates the most active unstable entropy components in each iteration. Mathematically, this process is equivalent to continuous annealing of the system state, maintaining the stability of entropy elasticity by dissipating the most destructive fatigue potential energy.
[0005] A vehicle trust assessment method based on double-potential-well field entropy flow annihilation includes:
[0006] S1: Establish a trust value probability distribution for each vehicle and calculate the information entropy; S2: Based on the vehicle's historical trust value and the information entropy calculated in S1, a dynamic dual potential well coupling field is constructed, consisting of a consensus potential well field and an individual stability potential well field, to guide the trust value to converge toward the group consensus and suppress irrational fluctuations. S3: Based on the information entropy calculated by S1, the double potential well coupling field constructed by S2, and the historical trust value of the vehicle, calculate the entropy flow intensity between vehicles, and introduce an entropy annihilation mechanism to actively eliminate unstable information entropy components. S4: Based on the information entropy calculated by S1, the entropy flow intensity calculated by S3, the vehicle's trust value in the previous evaluation cycle, the behavior consistency coefficient, the penalty factor, and the recovery acceleration factor, update the vehicle's information entropy value and trust value. S5: Based on the information entropy and trust value obtained in S4, as well as the vehicle's report value about the event, calculate the trust-weighted consensus value and dynamic decision threshold, and then make a group consensus decision and calculate the vehicle's behavior consistency coefficient. S6: Detect malicious vehicles based on the vehicle trust value obtained in S4; S7: Based on the inter-vehicle entropy flow intensity obtained in S3, the vehicle trust value obtained in S4, and the vehicle's report value about the event, identify the colluding group. S8: A dynamic network adaptation mechanism is adopted based on different situations such as vehicles joining or leaving the network or communication interruption timeout.
[0007] Optionally, S1 specifically includes the following steps: For each vehicle source of evidence Establish a probability distribution of trust values based on historical interaction data. And calculate its information entropy. ,in: The probability distribution Beta distribution , and vehicles The number of real and fake messages in the past 50 messages sent; Its information entropy The result is calculated using the formula:
[0008] in for function, for function, Represents the trust value.
[0009] Optionally, S2 specifically includes: Based on the vehicle's historical trust value and the information entropy calculated in S1, a dynamic double-potential-well coupled field is constructed, whose total potential field is... Consensus potential field With individual stability potential well field Coupled together: ; Consensus potential field: Based on normal vehicles in the past Average trust value over the assessment period and variance Constructing a consensus potential field The consensus potential well depth , It is a constant. The average entropy of a normal vehicle; Individual stability potential well field: ,in, and vehicles in the past The historical trust value mean and variance over each assessment period are used to resist irrational fluctuations in the population. This represents the depth coefficient of the individual potential well.
[0010] Optionally, S3 specifically includes: calculating the entropy flow intensity from vehicle i to vehicle j based on the information entropy calculated from S1, the double potential well coupling field calculated from S2, and the vehicle trust value. : ; in For learning rate, and Representing vehicles and vehicles Trust expectations in the previous assessment period; For similarity weights, As a potential well field Gradient magnitude at point, As an isolation factor; and Representing vehicles and vehicles Information entropy.
[0011] Optionally, S4 specifically includes: updating the vehicle's entropy and trust value based on the information entropy calculated by S1, the entropy flow intensity calculated by S3, the vehicle's trust value, behavioral consistency coefficient, penalty factor, and recovery acceleration factor in the previous evaluation cycle.
[0012] Update the entropy value of vehicle j based on entropy flow and annihilation: ; in, This indicates the previous evaluation period. Indicates the current evaluation period; It is an active control quantity. The annihilation coefficient is... Indicates vehicle To the vehicle Entropy flow intensity, For inflowing vehicles Total entropy flow; This represents the entropy flow intensity from vehicle j to vehicle k. Let J be the total entropy flow rate out of vehicle j; subsequently, based on the vehicle... Behavioral consistency, gradually updating its trust expectations; ; in For vehicle-based The behavioral consistency coefficient calculated from the difference between behavioral evidence and group consensus. This represents the trust value update coefficient. For inertial adjustment function, As a penalty factor, To restore the acceleration factor, Indicates the inertia adjustment coefficient; Inertial adjustment function Increase trust value The rate of change slows down as the value approaches extreme values of 0 or 1, thus achieving a gradual and smooth evolution of the trust value.
[0013] Optionally, S5 specifically includes: making a group consensus decision based on the information entropy and trust value obtained in S4, and the vehicle's report value about the event, and calculating the vehicle's behavioral consistency coefficient; calculating the trust-weighted consensus value for different states k of the same event reported by the normal vehicle group. ; in For vehicles Reported values, This represents the set of normal vehicles; next, based on the global average entropy of the current normal vehicles... Calculate the dynamic decision threshold and The calculation method is as follows: ; ; in It is the minimum half-interval constant. It is a scaling constant; if The output event is The decision; if Then the output event is not The decision is made based on the vehicle; otherwise, the decision is output as unknown. Finally, based on the vehicle... The reported value and the trust-weighted consensus value of events judged to be real. The difference is used to calculate the behavioral consistency coefficient. The calculation formula is as follows: ; This is the reported value for vehicle j.
[0014] Optionally, S6 specifically includes: detecting malicious vehicles based on the vehicle trust value obtained in S4; If vehicle Meeting their trust expectations If the value is lower than the average trust expectation of all normal vehicles, it is marked. If the condition is met, it is identified as a malicious vehicle, and an isolation factor is applied to its entropy flow intensity calculation. And apply a penalty factor during trust updates. ; The conditions for malicious vehicle marking described in S6 refer to the vehicle status... Divided into normal state The observation state is O, and the malicious state is M; the state transition rule is that when vehicle j meets the condition that the number of consecutive markers is greater than 1, the state transition occurs. When the state changes from normal N to observation O, vehicle j in state O, if the number of consecutive markers mentioned above is greater than... When the state changes to malicious M; vehicle j in the observation state O, if its behavior consistency coefficient Above the recovery threshold The number of consecutive times exceeded If so, the state returns to normal N.
[0015] Optionally, S7 specifically includes: identifying a colluding group based on the inter-vehicle entropy flow intensity obtained in S3, the vehicle trust value obtained in S4, and the vehicle's report value about the event. Calculate vehicle subset Behavioral coordination The degree of confrontation with group consensus and the entropy flow interaction within the group When the three are in continuous The values were all greater than their respective preset thresholds within each evaluation period. Indicates the number of evaluation periods for continuous testing, and the group size. At that time, determine the subset of vehicles. For the conspiratorial group, renewal The entropy flow isolation factor and penalty factor of the vehicle.
[0016] Optionally, in S8, the dynamic network adaptation mechanism is as follows: when a vehicle node joins or leaves the network, its trust distribution is initialized / removed based on its historical trust state or default value; when vehicle communication is interrupted for more than a preset time, its entropy flow calculation and trust update are suspended, and evaluation is continued based on the state before the interruption after the interruption is restored.
[0017] Optionally, the isolation factor described in S3 and the penalty factor described in S4 Different values are available depending on the vehicle's status: , ; in, Indicates vehicle The states are defined by N, O, and M, which represent the normal state, the observation state, and the malicious state, respectively. The trust assessment mechanism is executed once every 30 messages received by the vehicle to ensure the timeliness of the trust assessment results.
[0018] Compared with the prior art, the present invention has the following significant advantages: This invention employs a dual-potential-well coupled field, which guides the trust value of normal vehicles towards collective consensus while maintaining the stability of individual node behavior. The active entropy annihilation mechanism significantly improves the slow entropy response problem commonly found in entropy-based trust updates. When the proportion of malicious vehicles is 50%, the precision and recall rates of this invention in malicious vehicle identification tasks can both reach over 98%. Even with an attack probability of 50%, the invention still maintains a precision and recall rate of over 98%. In collusion attack scenarios, when the proportion of colluding attack vehicles is 50%, the precision and recall rates of colluding vehicle identification can both reach over 97%. Even with an attack probability of 50%, the precision and recall rates of colluding vehicle identification still remain above 95%. Attached Figure Description
[0019] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a diagram illustrating a vehicle-to-everything (V2X) trust assessment scenario according to an embodiment of the present invention. Figure 2 This is a comparison chart of the accuracy, recall rate and F1 score of malicious vehicle identification under different proportions of vehicles attacked by false messages in the embodiment. Figure 3 This is a comparison chart of the accuracy, recall rate and F1 score of malicious vehicle identification under different probabilities of fake news attacks in the embodiment. Figure 4 This is a comparison chart of the accuracy, recall rate and F1 score of malicious vehicle identification under different proportions of collusive attack vehicles in the embodiment. Figure 5 This is a comparison chart of the accuracy, recall rate, and F1 score of malicious vehicle identification under different collusion attack probabilities in the embodiments. Detailed Implementation
[0020] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.
[0021] The core idea of this invention, a vehicle-to-everything (V2X) trust assessment method and device based on entropy flow annihilation in a dual-potential-well field, is to construct a vehicle trust assessment mechanism based on the synergy of a dynamic dual-potential-well coupled field and an entropy annihilation mechanism. First, the dynamic dual-potential-well coupled field forms the system's elastic recovery framework. The consensus potential well guides node states towards global consensus by establishing a potential energy landscape based on the distribution of group trust, thus releasing stress from the accumulated deformation of entropy values. The individual stability potential well, centered on the historical trust average of vehicles, effectively suppresses irrational fluctuations in trust values by maintaining the stability of individual behavior patterns. Both provide structural protection against fatigue degradation for the system. Simultaneously, an entropy annihilation mechanism is introduced as an active repair strategy to address the entropy response lag problem. This mechanism, by designing an annihilation term proportional to the entropy inflow, selectively eliminates the most active unstable entropy components in each iteration. Mathematically, this process is equivalent to continuous annealing of the system state, maintaining the stability of entropy elasticity by dissipating the most destructive fatigue potential energy.
[0022] The method of this invention aims to solve the "entropy elasticity fatigue" problem in existing trust assessment methods, namely, the deficiency of decreased assessment accuracy and robustness due to the lag in trust entropy response after long-term operation. The core scheme includes: establishing a trust value probability distribution for each vehicle and calculating information entropy; constructing a dynamic dual-potential-well coupling field based on the obtained information entropy and the vehicle's historical trust value, consisting of a consensus potential-well field and an individual stability potential-well field, to guide trust values towards group consensus and suppress irrational fluctuations; calculating the entropy flow intensity between vehicles based on information entropy, the dual-potential-well coupling field, and vehicle trust value, and introducing an entropy annihilation mechanism to actively eliminate unstable entropy components; updating the vehicle's entropy and trust value based on information entropy, entropy flow intensity, vehicle's historical trust value, and behavioral consistency; calculating a trust-weighted consensus value and a dynamic decision threshold based on information entropy, vehicle trust value, and vehicle's reported values regarding events, thereby conducting group consensus decision-making and calculating the vehicle's behavioral consistency coefficient; detecting malicious vehicles based on vehicle trust value, and identifying collusive groups by combining the entropy flow intensity between vehicles and vehicle's reported values regarding events, achieving dynamic network adaptation throughout the process. Experiments have shown that the vehicle trust assessment method of this invention, through the synergistic effect of a dual potential well field and entropy annihilation, can effectively release entropy accumulation deformation and maintain entropy elastic stability. When the proportion of malicious vehicles is 50%, the accuracy and recall rate of malicious vehicle identification both exceed 98%; when the attack probability is 50%, it still maintains an accuracy and recall rate of over 98%. In this paper, "trust value" and "trust expectation value" have the same technical meaning and are not distinguished.
[0023] The method of this invention mainly consists of eight steps: S1 is system initialization: establishing an initial trust value probability distribution based on the Beta distribution for each vehicle node in the vehicle network. And calculate its information entropy. ; Information entropy is expressed by the formula Calculation, where For Beta functions, The digamma function. The parameters in the Beta function. and vehicles The number of real and fake messages in the past 50 messages sent, when the vehicle When there is no historical data, use the default parameters. and .
[0024] S2 is the potential well field construction: based on the vehicle's historical trust value and the information entropy calculated in the first step, a dynamic dual potential well coupled field is constructed, whose total potential field... Consensus potential field With individual stability potential well field Coupled together: According to normal vehicles in the past Average trust value over the assessment period and variance Constructing a consensus potential field The consensus potential well depth , This is a preset constant, with a value range of 0.5 to 2.0. The average entropy of normal vehicles, this potential well field guides the evolution of trust values, causing the trust values of normal vehicles to tend towards group consensus; individual stability potential well field ,in, and vehicles in the past The historical trust value mean and variance over each assessment period are used to maintain vehicle... The stability of an individual's trust state helps resist irrational fluctuations within the group, mitigating the impact of malicious vehicles' trust values on legitimate vehicles and ensuring that malicious vehicles are not pulled back to normal values by the majority of legitimate vehicles. Individual potential well depth coefficient. The value range is 0.2 to 1.0.
[0025] S3 is the entropy flow calculation: Based on the information entropy calculated in S1, the double-potential-well coupling field calculated in S2, and the vehicle trust value, the entropy flow intensity from vehicle i to vehicle j is calculated. ; in For learning rate, Weights are based on the similarity of trust values. For the double potential well coupled field in The gradient magnitude at a given point is calculated using the formula: ; Entropy flows from high-entropy nodes (high uncertainty) to low-entropy nodes, and similarity weights ensure that transmission is only effective between nodes with similar trust values. Isolation factors block the entropy flow of vehicles.
[0026] S4 is Entropy Annihilation and Trust Update: This step includes two parallel processes: entropy update and trust update. Based on the information entropy obtained from S1, the entropy flow intensity obtained from S3, and the vehicle's trust value, behavioral consistency coefficient, penalty factor, and recovery acceleration factor from the previous evaluation cycle, the entropy value and trust value are updated. In the entropy update step... Where t represents the previous evaluation period, Indicates the current evaluation period; It is an active control variable, where γ is the annihilation coefficient, ranging from 0.2 to 0.5. This represents the entropy flow intensity from vehicle i to vehicle j. Let J be the total entropy flow of vehicle j. This represents the entropy flow intensity from vehicle j to vehicle k. Let the total entropy flow out of vehicle j be denoted by entropy annihilation, which actively consumes the entropy flowing in, reducing system uncertainty, resolving the entropy response lag problem, and thus addressing the entropy elasticity fatigue problem. In the trust expectation value update, the trust expectation value is progressively updated based on the vehicle's behavioral consistency. ,in The behavioral consistency coefficient is calculated based on the difference between vehicle j's behavioral evidence and the group consensus. This represents the trust value update coefficient, with a value of 0.3. For inertial adjustment function, As a penalty factor, To restore the acceleration factor, This represents the inertia adjustment coefficient, with a value of 0.3. The technical effect of this function is to increase the confidence value. The rate of change slows down as the value approaches extreme values of 0 or 1, thus achieving a gradual and smooth evolution of the trust value.
[0027] S5 is the group consensus decision: based on the information entropy and trust value obtained from S4, and the vehicle reports about the event, a group consensus decision is made; during the consensus calculation process, for different states k of the same event reported by the normal vehicle group, the trust-weighted consensus value is calculated. ,in For vehicles Reported values, This represents the set of normal vehicles; next, based on the global average entropy of the current normal vehicles... Calculate the dynamic decision threshold and The calculation method is as follows: , ,in The minimum half-interval constant has a value of 0.1. This is a scaling constant with a value of 0.3. If If, then the output event is the decision for event k; if If the output event is not k, then the decision is output as unknown; otherwise, the decision is output as unknown. Finally, based on the vehicle... Reported values Trust-weighted consensus value of events judged to be real Based on the differences, calculate the behavioral consistency coefficient of vehicle j. The calculation formula is as follows: .
[0028] S6 is malicious vehicle detection, which detects malicious vehicles based on the vehicle trust value obtained in S4; if vehicle j meets its trust expectation value... Vehicles with trust values below the average of all normal vehicles are flagged. Based on the vehicle trust state transition mechanism, if certain conditions are met, they are identified as malicious vehicles, and an isolation factor is applied to their entropy flow intensity calculation. And apply a penalty factor during trust value updates. ; S7 is for identifying collusive groups: based on the inter-vehicle entropy flow intensity obtained in S3, the vehicle trust value obtained in S4, and the vehicle's report value about the event, collusive groups are identified; the behavioral coordination degree of the vehicle subset G is calculated. ; in, This represents the reported value of vehicle i in the t-th evaluation period; This is an indicator function, which is 1 when the condition is true; To assess the number of cycles, calculate the degree of adversarial relationship between the vehicle subset G and the group consensus: ; Calculate the entropy flow interaction density within the vehicle subset G. ; When the vehicle subset G is in a continuous Within a time period , , And group size At that time, G is the conspiratorial group, and all vehicle states are switched to the observation state. In the continuous... If the above conditions are met within a certain time period, all vehicles will be converted to a malicious status and added to the malicious vehicle blacklist.
[0029] S8 is a dynamic network adaptation mechanism: when a vehicle node joins or leaves the network, its trust distribution is initialized based on its historical trust state or default value; when vehicle communication is interrupted for more than a preset time, its entropy flow calculation and trust update are suspended, and evaluation is continued based on the state before the interruption after the interruption is restored.
[0030] The vehicle trust state transition mechanism in this invention comprises five parts: vehicle state definition, state transition rules and isolation and penalty factor settings, trust restoration and permission restoration, as detailed below: (1) Definition of vehicle state: The state of vehicle j It is divided into three states: normal state N, observation state O, and malicious state M. When... At that time, the vehicle is trusted and participates in various system calculations; when At that time, the vehicle's behavior became abnormal and it was under monitoring; when At that time, the vehicle was considered malicious and was isolated by the system.
[0031] (2) State transition rule: When the trust value of vehicle j is lower than the average trust expectation value of all normal vehicles for more than a certain number of consecutive times... When the state changes from normal N to observation O, vehicle j in state O, if the number of consecutive times exceeds [a certain threshold], [further changes will occur]. When the state changes to malicious M, the vehicle is added to the attack vehicle blacklist. For vehicle j in malicious state M / observation state, if its behavior consistency coefficient... Above the recovery threshold The number of consecutive times exceeded If so, the state will revert to Observation O / Normal N.
[0032] (3) Isolation factor and penalty factor settings: For vehicles in three states, the isolation factor is set as follows: With penalty factor Set to respectively , ; Vehicles under observation can still gain trust, but their speed will be reduced.
[0033] (4) Trust Recovery: When a vehicle in the observation state exhibits consistently good behavior, i.e., transitions to a normal state, the trust recovery process is initiated. The recovery acceleration factor in the trust value update formula... ,in The recovery acceleration factor is set at 0.1-0.3. This factor allows vehicles with higher behavioral consistency to recover their trust value faster.
[0034] (5) Permission Restoration: When a vehicle returns from observation mode to normal mode, its permissions are not immediately and completely restored, but rather a gradual process; after restoration... Within each cycle, its isolation factor and penalty factor recover smoothly as follows: , ,in This represents the number of cycles experienced by vehicle j since its recovery time, preventing sudden changes in permissions and further defending against malicious attacks. It also represents the subsequent recovery acceleration factor after the vehicle recovers from a malicious state. It will be affected by historical records, that is ,in This represents the total number of times the vehicle has been judged as malicious in its history, which slows down the recovery of trust value for vehicles that have launched multiple attacks.
[0035] Figure 1 This invention illustrates a vehicle-to-everything (V2X) trust assessment scenario, in which vehicle nodes exchange traffic event information, each vehicle has a built-in trust assessment module, and the trust value of each vehicle is assessed and event decisions are made based on the method of this invention.
[0036] The overall process of the method of the present invention includes the following steps: system initialization, construction of a dual potential well field, entropy flow calculation, entropy annihilation and trust update, group consensus decision-making, malicious vehicle detection, and collusion group detection.
[0037] Road distribution Vehicles, recorded as When a traffic incident occurs, vehicle nodes near the incident location are the first to detect it. These vehicles then broadcast the detected event, allowing vehicles farther away to receive the message and become aware of the incident. Malicious vehicles in the network can launch attacks, including disinformation attacks and collusion attacks. All vehicles, upon observing a traffic incident, send relevant event messages to surrounding vehicles. When multiple vehicles send information about the same road event (such as "Has a traffic accident occurred?"), a truth reporting group (...) ) Reported value (Observations are consistent with actual events); Malicious reporting group ( ) Reported value The vehicle receiving the message performs a vehicle trust calculation on the vehicle that sent the message, and designates the vehicle that performed the trust calculation as the evaluation vehicle. The vehicle being evaluated is called the target vehicle. When a vehicle receiving a message performs a trust assessment of the vehicle that sent the message, it uses a periodic trust value calculation method, that is, a vehicle trust assessment is performed every 30 messages received.
[0038] Example 1: A typical scenario without collusion attacks (multiple real vehicle reports + a small number of non-malicious false positives) Step 1: Establish an initial trust value probability distribution for each vehicle based on the Beta distribution. .in, For Beta functions, The gamma function is used; the initial trust mean of the vehicle is calculated from the Beta distribution parameters. , To evaluate the vehicle Received the target vehicle within the past 50 messages The number of correct event messages sent. The number of error event messages, using the default parameter. and .
[0039] Subsequently, the uncertainty of the trust value of each vehicle was quantified using the information entropy formula corresponding to the Beta distribution. .in, The function is digamma. Calculate the initial trust mean for normal vehicles (initially all vehicles are normal vehicles). and the average value of the initial trust information entropy. This serves as the basic input for the subsequent construction of the potential well.
[0040] Step 2: Construct a dynamic double-potential-well coupling field based on the vehicle's historical trust value and the information entropy calculated in Step 1. First, based on the average trust value of all normal vehicles over the past 10 evaluation periods... With variance Constructing a consensus potential well field guides group trust to converge towards consensus; the formula is as follows: Among them, the consensus potential well depth , Secondly, regarding vehicles Historical trust average over the past 10 assessment periods Construct an individual stability potential well field centered on [the target]. .in, Finally, the consensus potential well field and the individual stability potential well field are superimposed to form the total potential well field. .
[0041] Step 3: Calculate the entropy flow between vehicles. Based on the information entropy calculated in Step 1, the double-potential-well coupling field calculated in Step 2, and the vehicle trust value, calculate the entropy flow from each vehicle. To the vehicle Entropy flow intensity ,in, The value is 0.3.
[0042] Step 4: Based on the information entropy obtained in Step 1, the entropy flow intensity obtained in Step 3, and the vehicle's trust value, behavioral consistency coefficient, penalty factor, and recovery acceleration factor from the previous evaluation period, update the entropy and trust values. First, update the entropy value of vehicle j. ,in It is an active control variable, with γ taking a value of 0.2. Then, combining the behavioral consistency of vehicle j with the vehicle's trust state, the trust value is updated. .
[0043] Step 5, for the normal vehicle i's report regarding the event "whether a traffic accident occurred" in two different states "occurred" ( ) and "did not happen" Based on the vehicle trust value calculated in step 4, a trust-weighted consensus value is calculated. The trust consensus value for a traffic accident is... The consensus value for trust that no traffic accident has occurred is [value missing]. ,in Represents the set of normal vehicles; based on the global average entropy of the current normal vehicles. Calculate the dynamic decision threshold and , , ,in The value is 0.1. The value is 0.3. If If so, then the output will show the decision that a traffic accident has occurred; if If the result is positive, the output is a decision that no traffic accident occurred; otherwise, the output is an unknown decision. Finally, based on the vehicle... Reported values The difference between the trust-weighted consensus value and the value determined to be real is used to calculate the vehicle's... Behavioral consistency coefficient .
[0044] Step 6: If the trust value of vehicle j is lower than the average trust value of normal vehicles, it is marked. If the number of consecutive markings exceeds 5, its state changes from normal (N) to observation (O). If the number of consecutive markings exceeds 5 again, its state changes to malicious (M). At this point, vehicle j is identified as a malicious vehicle and added to the blacklist. Simultaneously, the isolation factor and penalty factor for vehicle j are adjusted accordingly.
[0045] If vehicle j is misjudged and remains under observation, a trust recovery process begins when vehicle j exhibits consistently good behavior. The trust value update formula includes a recovery acceleration factor. ,in The value is set to 0.1. Simultaneously, the corresponding isolation factor and penalty factor are restored.
[0046] Each trust assessment yields the current trust value for all vehicles and the received event decision results.
[0047] Example 2: Adversarial scenario involving conspiracy attacks (multiple real vehicle reports + conspiracy to create false reports) The scenario is the same as in Example 1, where a group conspiring to falsely report events sends out false event messages in an attempt to interfere with event decision-making.
[0048] Steps 1 to 7 are the same as in Example 1, wherein The value is 2.0. The value is 1.0. The value is 0.3, and γ is 0.5. After detecting malicious vehicles, the behavioral coordination degree of each vehicle subset G over the past 3 evaluation periods is calculated, along with its adversarial degree against the average trust value of all normal vehicles and its internal entropy flow interaction density. When vehicle subset G has been detected for two consecutive time periods... , , And group size At this point, G is the conspiratorial group, and all vehicles are switched to the observation state. If the above conditions are met for four consecutive time periods, all vehicles are switched to the malicious state and added to the malicious vehicle blacklist.
[0049] Comparative Example 1: Vehicle trust assessment and malicious vehicle detection are performed according to the scenario and steps of Example 1, with the difference being that only a consensus potential well field is constructed, and no individual stability potential well field is constructed.
[0050] Comparative Example 2: Vehicle trust assessment and collusive vehicle detection are performed according to the scenario and steps of Example 2, except that the entropy annihilation mechanism is not used.
[0051] Implementation effect analysis: (1) Comparative Example 1 and Comparative Example 1 Figure 2 and Figure 3 As can be seen, by coupling the consensus potential field and the individual stability potential field, the accuracy and recall rate of malicious vehicle identification can both reach over 98% when the proportion of malicious vehicles is 50%; when the attack probability is 50%, the accuracy and recall rate can reach over 98%. The dual potential field coupling can effectively solve the "entropy elastic fatigue" problem and significantly improve the accuracy and reliability of vehicle trust assessment.
[0052] (2) Comparative Example 2 and Comparative Example 2 Figure 4 and Figure 5 As can be seen, by introducing the entropy annihilation mechanism, when the proportion of collusive attack vehicles is 50%, the accuracy and recall rate of collusive vehicle detection can both reach over 97%; when the attack probability is 50%, the accuracy and recall rate of collusive vehicle identification still remain above 95%. The entropy annihilation mechanism can effectively solve the problem of "entropy elastic fatigue" and significantly improve the accuracy and reliability of collusive attack detection.
[0053] The above description provides examples of the preferred embodiments of the present invention. Parts not detailed herein are common knowledge to those skilled in the art. The scope of protection of the present invention is determined by the claims. Any equivalent modifications based on the technical teachings of the present invention are also within the scope of protection of the present invention.
Claims
1. A vehicle trust assessment method based on double-potential-well field entropy flow annihilation, characterized in that, include: S1: Establish a trust value probability distribution for each vehicle and calculate the information entropy; S2: Based on the vehicle's historical trust value and the information entropy calculated in S1, a dynamic dual potential well coupling field is constructed, consisting of a consensus potential well field and an individual stability potential well field, to guide the trust value to converge toward the group consensus and suppress irrational fluctuations. S3: Based on the information entropy calculated by S1, the double potential well coupling field constructed by S2, and the historical trust value of the vehicle, calculate the entropy flow intensity between vehicles, and introduce an entropy annihilation mechanism to actively eliminate unstable information entropy components. S4: Based on the information entropy calculated by S1, the entropy flow intensity calculated by S3, the vehicle's trust value in the previous evaluation cycle, the behavior consistency coefficient, the penalty factor, and the recovery acceleration factor, update the vehicle's information entropy value and trust value. S5: Based on the information entropy and trust value obtained in S4, as well as the vehicle's report value about the event, calculate the trust-weighted consensus value and dynamic decision threshold, and then make a group consensus decision and calculate the vehicle's behavior consistency coefficient. S6: Detect malicious vehicles based on the vehicle trust value obtained in S4; S7: Based on the inter-vehicle entropy flow intensity obtained in S3, the vehicle trust value obtained in S4, and the vehicle's report value about the event, identify the colluding group. S8: A dynamic network adaptation mechanism is adopted based on different situations such as vehicles joining or leaving the network or communication interruption timeout.
2. The vehicle trust assessment method based on double-potential-well field entropy flow annihilation according to claim 1, characterized in that, S1 specifically includes the following steps: For each vehicle source of evidence Establish a probability distribution of trust values based on historical interaction data. And calculate its information entropy. ,in: The probability distribution Beta distribution , and vehicles The number of real and fake messages in the past 50 messages sent; Its information entropy The result is calculated using the formula: in for function, for function, This represents the trust value.
3. The vehicle trust assessment method based on double-potential-well field entropy flow annihilation according to claim 1 or 2, characterized in that, The S2 specifically includes: Based on the vehicle's historical trust value and the information entropy calculated in S1, a dynamic double-potential-well coupled field is constructed, whose total potential field is... Consensus potential field With individual stability potential well field Coupled together: ; Consensus potential field: Based on normal vehicles in the past Average trust value over the assessment period and variance Constructing a consensus potential well field The consensus potential well depth , It is a constant. The average entropy of a normal vehicle; Individual stability potential well field: ,in, and vehicles in the past The historical trust value mean and variance over each assessment period are used to resist irrational fluctuations in the population. This represents the depth coefficient of the individual potential well.
4. The vehicle trust assessment method based on double-potential-well field entropy flow annihilation according to claim 1 or 2, characterized in that, Specifically, S3 includes: calculating the information entropy obtained from S1, the double-potential-well coupling field obtained from S2, and the vehicle trust value, and calculating the entropy flow intensity from vehicle i to vehicle j. : ； in For learning rate, and Representing vehicles and vehicles Trust expectations in the previous assessment period; For similarity weights, As a potential trap field Gradient magnitude at point, As an isolation factor; and Representing vehicles and vehicles Information entropy.
5. The vehicle trust assessment method based on double-potential-well field entropy flow annihilation according to claim 1 or 2, characterized in that, Specifically, S4 includes updating the vehicle's entropy and trust value based on the information entropy calculated by S1, the entropy flow intensity calculated by S3, the vehicle's trust value, behavioral consistency coefficient, penalty factor, and recovery acceleration factor in the previous evaluation cycle. Update the entropy value of vehicle j based on entropy flow and annihilation: ； in, This indicates the previous evaluation period. Indicates the current evaluation period; It is an active control quantity. The annihilation coefficient is... Indicates vehicle To the vehicle Entropy flow intensity, For inflowing vehicles Total entropy flow; This represents the entropy flow intensity from vehicle j to vehicle k. Let J be the total entropy flow rate out of vehicle j; subsequently, based on the vehicle... Behavioral consistency, gradually updating its trust expectations; ； in For vehicle-based The behavioral consistency coefficient calculated from the difference between behavioral evidence and group consensus. This represents the trust value update coefficient. For inertial adjustment function, As a penalty factor, To restore the acceleration factor, Indicates the inertia adjustment coefficient; Inertial adjustment function Increase trust value The rate of change slows down as the value approaches extreme values of 0 or 1, thus achieving a gradual and smooth evolution of the trust value.
6. The vehicle trust assessment method based on double-potential-well field entropy flow annihilation according to claim 1 or 2, characterized in that, Specifically, S5 includes: making a group consensus decision based on the information entropy and trust value obtained in S4, and the vehicle's report value about the event, and calculating the vehicle's behavioral consistency coefficient; calculating the trust-weighted consensus value for different states k of the same event reported by the normal vehicle group. ； in For vehicles Reported values, This represents the set of normal vehicles; next, based on the global average entropy of the current normal vehicles... Calculate the dynamic decision threshold and The calculation method is as follows: ； ； in It is the minimum half-interval constant. It is a scaling constant; if The output event is The decision; if Then the output event is not The decision is made based on the vehicle; otherwise, the decision is output as unknown. Finally, based on the vehicle... The reported value and the trust-weighted consensus value of events judged to be real. The difference is used to calculate the behavioral consistency coefficient. The calculation formula is as follows: ; This is the reported value for vehicle j.
7. The vehicle trust assessment method based on double-potential-well field entropy flow annihilation according to claim 1 or 2, characterized in that, Specifically, S6 includes: detecting malicious vehicles based on the vehicle trust value obtained in S4; If vehicle Meet their trust expectations If the value is lower than the average trust expectation of all normal vehicles, it is marked. If the condition is met, it is identified as a malicious vehicle, and an isolation factor is applied to its entropy flow intensity calculation. And apply a penalty factor during trust updates. ; The conditions for malicious vehicle marking described in S6 refer to the vehicle status... Divided into normal state The observation state is O, and the malicious state is M; the state transition rule is that when vehicle j meets the condition that the number of consecutive markers is greater than 1, the state transition occurs. When the state changes from normal N to observation O, vehicle j in state O, if the number of consecutive markers mentioned above is greater than... When the state changes to malicious M; vehicle j in the observation state O, if its behavior consistency coefficient Above the recovery threshold The number of consecutive times exceeded If so, the state returns to normal N.
8. The vehicle trust assessment method based on double-potential-well field entropy flow annihilation according to claim 1 or 2, characterized in that, Specifically, S7 includes: identifying collusive groups based on the inter-vehicle entropy flow intensity obtained in S3, the vehicle trust value obtained in S4, and the vehicle's report value about the event. Calculate vehicle subset Behavioral coordination The degree of confrontation with group consensus and entropy flow interaction within the group When the three are in continuous The values were all greater than their respective preset thresholds within each evaluation period. Indicates the number of evaluation periods for continuous testing, and the group size. At that time, determine the subset of vehicles. For the conspiratorial group, renewal The entropy flow isolation factor and penalty factor of the vehicle.
9. The vehicle trust assessment method based on double-potential-well field entropy flow annihilation according to claim 1 or 2, characterized in that, In S8, the dynamic network adaptation mechanism is to initialize / remove the trust distribution based on the vehicle node's historical trust state or default value when the vehicle node joins or leaves the network. If vehicle communication is interrupted for more than a preset time, its entropy flow calculation and trust update will be suspended, and the evaluation will continue based on the state before the interruption after the communication is restored.
10. The vehicle trust assessment method based on double-potential-well field entropy flow annihilation according to claim 1 or 2, characterized in that, The isolation factor described in S3 and the penalty factor described in S4 Different values are available depending on the vehicle's status: ， ； in, Indicates vehicle The states are defined by N, O, and M, which represent the normal state, the observation state, and the malicious state, respectively. The trust assessment mechanism is executed once every 30 messages received by the vehicle to ensure the timeliness of the trust assessment results.