A security-enhanced trust assessment method for Internet of Vehicles combined with transfer learning

Abstract

The present invention discloses a security-enhanced Internet of Vehicles (IoV) trust assessment method combined with transfer learning, which relates to the field of IoV security. The method comprises the following steps: constructing an IoV model for urban roads based on urban road conditions; designing an interactive authentication module based on the elliptic curve Diffie-Hellman protocol and the elliptic curve digital signature algorithm based on the IoV model for urban roads; introducing a true and false message identification method based on a three-factor trust model improved by a transfer learning method based on the interactive authentication module, designing a transfer learning model TranferMegaCRN in speed expectation sub-factor analysis to obtain the predicted speed of the message sender; designing a vehicle trust value calculation method based on the transfer learning model TranferMegaCRN using an improved Bayesian method and a density-based clustering screening algorithm, and constructing an IoV trust management model; and completing vehicle trust assessment based on the IoV trust management model.

Description

Technical Field

[0001] The present invention relates to the field of vehicle networking security, and in particular to a vehicle networking trust assessment method that combines security enhancement with transfer learning. Background Art

[0002] The rapid development of vehicular ad hoc networks (VANETs) in recent years has brought numerous opportunities for intelligent transportation systems, significantly improving road safety and optimizing traffic flow management. In VANET environments, vehicles engage in extensive communication and data exchange. Establishing a reliable trust mechanism can ensure communication security and data integrity. However, VANETs are vulnerable to malicious attacks, data tampering, and information leakage, making traditional security mechanisms ineffective. Furthermore, the high-speed mobility, complex communication environments, and rapidly changing network topologies of VANETs complicate trust management, making inter-vehicle trust management a critical issue that remains unresolved. Therefore, establishing an efficient and accurate VANET trust management mechanism is crucial. This mechanism can ensure system security and stability, helping vehicles effectively identify and mitigate malicious behavior in complex network environments, thereby improving overall safety and reliability.

[0003] To ensure the security of the connected vehicle system, a trust management system can ensure data integrity and accuracy through technologies such as encryption and verification, message digests, and timestamps, thereby enhancing the credibility of inter-vehicle communications. Furthermore, trust management mechanisms can help administrators assess the credibility of each vehicle. By continuously tracking and analyzing vehicle behavior records, service quality, and historical performance, administrators can obtain data-driven assessment results, providing a basis for implementing targeted rewards and penalties. This not only curbs malicious behavior but also incentivizes vehicles to provide high-quality services, further optimizing the overall operational efficiency and stability of the connected vehicle system. Summary of the Invention

[0004] To address the technical problems presented above, this paper proposes a security-enhanced IoV trust management model (TSE-TMM) that incorporates transfer learning. This paper examines a trust management architecture for urban road scenarios, where vehicles travel on roads, collecting and transmitting information. To identify potential attackers and ensure road safety, a clear metric is needed to distinguish between legitimate and malicious vehicles. Due to the limited capabilities of vehicles, roadside units and control centers are required to identify attackers on the road. This IoV trust management model utilizes an improved Bayesian method and a density-based clustering and screening algorithm to calculate vehicle trust values, quantifying the level of trust in vehicles. This approach helps identify potential attackers and achieves IoV system security. The framework also incorporates an interactive authentication module based on the Elliptic Curve Diffie-Hellman protocol and the Elliptic Curve Digital Signature Algorithm to enable anonymous communication, vehicle traceability, and identity authentication between vehicles. An improved trust model based on expectation, risk, and confidence factors is introduced to identify the authenticity of messages transmitted between vehicles. Within the expectation factor analysis, a transfer learning model, TranferMegaCRN, is designed to obtain the message sender's predicted speed to determine the speed expectation sub-factor, enhancing the security and reliability of the framework.

[0005] To achieve the above objectives, the present invention provides a security-enhanced Internet of Vehicles trust assessment method combined with transfer learning, comprising the following steps:

[0006] Based on urban road conditions, build an urban road vehicle network model;

[0007] Based on the urban road vehicle network model, an interactive authentication module based on the elliptic curve Diffie-Hellman protocol and the elliptic curve digital signature algorithm is designed;

[0008] Based on the interactive authentication module, a method for identifying true and false messages based on a three-factor trust model improved by transfer learning is introduced. In the speed expectation sub-factor analysis, a transfer learning model TranferMegaCRN is designed to obtain the predicted speed of the message sender.

[0009] Based on the transfer learning model TranferMegaCRN, an improved Bayesian method and a density-based clustering screening algorithm are used to design a vehicle trust value calculation method and build a vehicle network trust management model;

[0010] Based on the Internet of Vehicles trust management model, vehicle trust assessment is completed.

[0011] Preferably, the constructed urban road vehicle network model includes four main bodies: a vehicle equipped with an on-board unit OBU, a roadside unit RSU, a trusted control center TA and a road speed monitor.

[0012] Preferably, the steps of designing the vehicle trust value calculation method include: in direct trust value calculation, designing a sliding window method with a time decay factor and a variance-based risk adjustment mechanism to optimize the calculation method based on the Bayesian formula; in indirect trust calculation, designing a density-based clustering screening algorithm to extract trust feature vectors from RSU and eliminate potential malicious recommendation nodes in high-density clusters.

[0013] Preferably, the method for constructing the Internet of Vehicles trust management model includes:

[0014] Based on the vehicle trust value calculation method, a management model framework is constructed;

[0015] Introducing an interactive authentication module based on the elliptic curve Diffie-Hellman protocol and the elliptic curve digital signature algorithm into the management model framework;

[0016] The three-factor message recognition method improved based on the transfer learning method is introduced into the management model framework to complete the construction.

[0017] Preferably, the steps of introducing the interactive authentication module of the elliptic curve Diffie-Hellman protocol and the elliptic curve digital signature algorithm include: when registering a vehicle, the elliptic curve Diffie-Hellman protocol is used to negotiate a shared key with the control center TA, encrypting the real ID to achieve anonymous communication, and ensuring that TA can trace its identity; during the communication process, the elliptic curve digital signature algorithm is used to sign and verify all messages, and if the receiver fails to verify, the sender is marked as a malicious node.

[0018] Preferably, the transfer learning method is combined with a time series model based on the meta-graph convolutional recurrent network MegaCRN to construct a transfer learning model TransferMegaCRN to predict the speed of the message sender and obtain the speed expectation sub-factor.

[0019] Preferably, the transfer learning steps based on TransferMegaCRN include: in the input stage, the input original speed data x and the time feature covariate y1 are respectively passed through their respective fully connected layers and then input into the MegaCRN model; in the MegaCRN model, the parameters of all encoder modules in the MegaCRN pre-trained with the traffic flow dataset METR-LA are frozen; a modular progressive unfreezing strategy is adopted to unfreeze the decoder and memory module. During the unfreezing process, the deep layers of the encoder are gradually unfrozen starting from the 5th round, and the middle layers are continued to be unfrozen after the 10th round, with a gradual adjustment from fixing the bottom layer to dynamically releasing the high layer; in the output stage, the output of the MegaCRN model is further processed through a fully connected layer.

[0020] Preferably, based on the TransferMegaCRN speed prediction model, the improved true and false message identification method of the three factors of expectation, risk and confidence includes: the receiving vehicle calculates the expectation value, risk value and confidence value determined by the speed and position of the sending vehicle to comprehensively evaluate the credibility of the message, and when the dynamic balance condition is met, the message is judged to be true, otherwise it is marked as unreliable.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] The security-enhanced IoV trust management model combined with transfer learning, described in this paper, integrates elliptic curve cryptography (ECDH and ECDSA) and an improved three-factor message identification method based on an optimized vehicle trust calculation model. It uses an improved Bayesian approach to calculate direct trust values ​​between vehicles. Real-time interaction data is recorded using a sliding window and historical weights are optimized using a time-decay factor. A variance adjustment mechanism is designed to reduce trust scores in high-uncertainty scenarios. A density-based clustering and screening algorithm is designed. The RSU extracts trust feature vectors and removes potentially malicious recommendation nodes within high-density clusters. Finally, the purified indirect trust and direct trust are weighted and aggregated to generate a local trust value. ECDH is used to establish a secure key between vehicles and the control center, enabling anonymous and private communication between vehicles and vehicle identity tracking by the control center. ECDSA is used to authenticate and protect message integrity in vehicle-to-vehicle communications. The improved three-factor message identification method innovatively introduces the TranferMegaCRN spatiotemporal prediction model, optimized through transfer learning, to replace the traditional speed expectation calculation. A modular progressive unfreezing strategy is employed to improve the accuracy of message authenticity verification in complex traffic scenarios. This ensures IoV system security. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 This is a schematic diagram of a trust model for urban road vehicle networking according to an embodiment of the present invention;

[0025] Figure 2 A TransferMegaCRN model diagram of the predicted speed according to an embodiment of the present invention;

[0026] Figure 3 This is a framework diagram of the security-enhanced Internet of Vehicles trust management model TSE-TMM combined with transfer learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0028] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] Before describing the present invention, the terms used in the present invention will be explained first.

[0030] (1) Vehicle: The vehicle is equipped with an on-board unit (OBU), which has certain computing capabilities, can also perceive the environment, collect data, and help vehicles interact.

[0031] (2) Roadside Unit (RSU): RSU is a device with base station functions deployed on both sides of the road, providing communication support, assisting in calculations, and acting as an interactive intermediary to help spread information.

[0032] (3) Control center TA: mainly provides computing support, can perform a large number of computing tasks and help vehicles or RSUs make further decisions.

[0033] Example:

[0034] This embodiment provides a security-enhanced Internet of Vehicles trust assessment method combined with transfer learning, including the following steps:

[0035] S1. Build an urban road vehicle network model based on urban road conditions.

[0036] like Figure 1 As shown, the urban road vehicle network model constructed in this embodiment mainly has four entities, including: vehicles equipped with on-board units OBU, roadside units RSU, trusted control centers TA and road speed monitors.

[0037] (1) Vehicle mobility model:

[0038] To demonstrate the complexity of urban roads, the Manhattan Mobility Model (MMM) is used as the road mobility model. Vehicle speeds are constrained by the speed of the preceding vehicle, reflecting the continuity of traffic flow. Nodes must follow the road's direction and cannot move freely off-road. Under the standard MMM, mobile nodes (i.e., vehicles) typically move along a grid of horizontal and vertical streets on a map. As the vehicle continues its journey until it reaches the next intersection of a horizontal and vertical street, it can turn left, right, or go straight with a certain probability. The default turning probabilities in the Manhattan model are: a probability of 0.5 for going straight and a probability of 0.25 for turning left or right. Each vehicle makes a road choice at an intersection. Therefore, even if all vehicles depart from the same starting point, after a certain period of time, they will be scattered everywhere, similar to the complex urban environment of reality. However, to prevent collisions, roads are typically designed with at least two lanes to ensure that vehicles can move back and forth. This is consistent with urban road conditions.

[0039] (2) Communication loss model:

[0040] Due to the large number of tall buildings in cities, communication loss needs to be considered. According to the characteristics of wireless signal propagation in the Internet of Vehicles, signal fading mainly includes path loss and shadow fading. Path loss can be calculated by calculating the free space path loss:

[0041] ,

[0042] Where α2 is the path loss exponent in the free space model; is the reference distance, usually defaulted to 1 (unit: meter); is the straight-line distance between the transmitter and the receiver (unit: meter). Shadow fading can be used to calculate the penetration loss of the building to obtain:

[0043] ,

[0044] in, Indicates how many walls the signal passes through. Each time a wall is passed through, it is counted as 1 "cut", so The value directly corresponds to the total number of walls the signal passes through. is the fixed loss of passing through a wall; is the signal penetration distance inside the building (unit: meter), that is, the propagation length of the signal inside the building; is the distance-related loss, that is, the loss per meter. Therefore, based on the above path loss formula and shadow fading formula, the overall signal fading model is:

[0045] ,

[0046] Whether a vehicle can receive messages sent by other entities while driving needs to be determined based on the vehicle's power and noise. First, the vehicle's receiving power and signal-to-noise ratio (SNR) need to be calculated:

[0047] ,

[0048] ,

[0049] in, is the vehicle-mounted transmitting power, and are the transmitting and receiving antenna gains, is the noise power. When the vehicle's receiving power is greater than the receiving sensitivity threshold And the signal-to-noise ratio is greater than the demodulation threshold The vehicle can receive the message correctly only when:

[0050] ,

[0051] .

[0052] S2. Based on the urban road vehicle network model, an interactive authentication module based on the elliptic curve Diffie-Hellman protocol and the elliptic curve digital signature algorithm is designed.

[0053] In the urban road model of this embodiment, vehicles achieve V2V / V2I communication by periodically broadcasting beacon messages. The message content includes core data such as vehicle ID, real-time location, driving speed, and communication data. However, this broadcast-based open communication architecture faces multiple security threats. Therefore, cryptographic technology is introduced. Cryptographic technology is used to achieve data integrity, authentication, confidentiality, and anonymity to effectively prevent unauthorized modification and illegal access of information, thereby ensuring the security and reliability of the communication process. In terms of the selection of cryptographic technology, in the case of resource-constrained Internet of Vehicles scenarios with high real-time requirements, the choice of elliptic curve cryptography (ECC) has obvious advantages over traditional public key cryptography systems (such as asymmetric cryptography RSA). ECC can use shorter keys while providing the same security strength, thereby reducing computational overhead and storage requirements. This is particularly important for high-speed vehicles and large-scale network environments.

[0054] The ECDH key negotiation mechanism is used to achieve a safe and efficient key exchange between the vehicle and the control center, ensuring the confidential transmission of sensitive communication data. Negotiate a shared key with the control center TA through the ECDH protocol to protect sensitive information such as the vehicle's true identity. First, TA will pre-disclose the elliptic curve public base point and its own public key , Generate private random numbers And calculate the relevant public key Submit to TA. Then, the vehicle Based on its secret , TA based on its private key The same shared key can be obtained by calculating separately ,vehicle The shared key is used to encrypt its real ID, and the message is signed using ECDSA and sent to the receiving vehicle. Due to other vehicles Did not participate in the above key negotiation, so this shared key cannot be obtained , it cannot be decrypted and identified The real ID of the user can only obtain the encrypted anonymous identity; and the user, as the key holder, can use Decryption and tracing Identity, achieving a balance between anonymity and regulatory traceability in V2V communications.

[0055] The ECDSA authentication mechanism is used to verify and confirm the identity of each vehicle in communication to ensure the authenticity of the communication data source. To the vehicle When sending message m, the ECDSA signature scheme is used to send the signed message : , , is the common base point of the elliptic curve, For vehicles A randomly selected security parameter; , is the hash value of message m, is a secure hash function, It's a vehicle The signature private key (corresponding to the public key ). When receiving the signed message When , the following verification is performed: Calculate , , , ; then calculate And judge the equation Is it true? If so, the signature is valid.

[0056] S3. Based on the interactive authentication module, a method for identifying true and false messages based on a three-factor trust model improved by transfer learning is introduced. In the speed expectation sub-factor analysis, a transfer learning model TranferMegaCRN is designed to obtain the predicted speed of the message sender.

[0057] Based on the three-factor trust model, this embodiment combines the transfer learning method with the meta-graph convolutional recurrent network time series model MegaCRN to design a transfer learning model TransferMegaCRN for vehicle speed prediction to better analyze the impact of the speed expectation sub-factor among the three factors.

[0058] V2V message results are assumed to follow a binomial distribution. Specifically, each interaction between two vehicles is considered an independent experiment, meaning each vehicle's V2V message has two possible outcomes: true or false. The goal of three-factor message identification is to identify all V2V messages and classify them as authentic or unauthentic.

[0059] The main implementation steps are as follows:

[0060] S301. Expected value evaluation: The receiver calculates the expected value based on the position information and speed information in the received beacon message. The position expectation is recorded as , the distance between the receiver and the sender is recorded as , the V2V communication range is recorded as , then the position expectation The value of is set to:

[0061] ,

[0062] The expected speed is recorded as , which passes the sender speed value The receiver predicts the sender's speed value The absolute value of the difference To calculate:

[0063] ,

[0064] in, , , is the maximum speed, and the parameter α1 is .

[0065] This embodiment builds a transfer learning model based on MegaCRN to predict the speed Make predictions. The network architecture of MegaCRN mainly consists of two core parts: the graph convolutional recurrent unit (GCRU) encoder-decoder and the spatiotemporal meta-graph learner. The GCRU encoder-decoder extracts the spatial dependencies (such as road network topology) and temporal dynamics (such as traffic flow changes) of spatiotemporal data in the encoding stage by fusing the graph convolutional network (GCN) and the gated recurrent unit (GRU), and generates multi-step prediction results in the decoding stage. The spatiotemporal meta-graph learner realizes adaptive graph structure learning through the meta-node library (storing learnable traffic pattern prototypes), the query-attention mechanism (dynamically matching the current state with the prototype to reconstruct node features) and the hypernetwork (generating dynamic graph structure), thereby explicitly distinguishing the heterogeneity of different spatiotemporal scenarios (such as rush hour, traffic accidents). The two parts work together to improve the adaptability and prediction accuracy of the model to complex traffic environments. TransferMegaCRN adds multiple fully connected layers in the entire transfer learning model, mainly in the input and output stages, such as Figure 2 As shown. First, in the input stage, the original speed data x and the time feature covariate y1 pass through their respective fully connected layers for feature adjustment or preprocessing. Secondly, in the output stage, the output of TransferMegaCRN will first pass through a fully connected layer for further processing to adapt to specific tasks or adjust the output dimension. In addition, it is also necessary to pay attention to freezing the pre-trained MegaCRN parameters. The purpose is to prevent the destruction of useful features that have been learned when training new tasks, while reducing the number of training parameters and reducing the risk of overfitting. The parameter freezing training scheme is the main step of transfer learning. In practical applications, it can be dynamically frozen and adjusted according to the requirements of the model task. In the transfer learning model TransferMegaCRN used in this embodiment, a modular progressive unfreezing strategy is adopted. First, freeze the parameters of all encoder modules in MegaCRN pre-trained with the traffic flow dataset METR-LA (to retain the universal spatiotemporal features extracted by it), and then unfreeze the decoder and memory modules to adapt to the new task. To ensure gradual unfreezing, during training, the deep layers of the encoder are gradually unfrozen starting from the fifth epoch, and the intermediate layers continue to be unfrozen after the tenth epoch, gradually adjusting from a fixed bottom layer to a dynamic release of higher layers. This strategy maintains the core feature extraction capabilities of the pre-trained model while enabling efficient adaptation to new tasks through phased unfreezing.

[0066] According to the evaluation of the expected value, the final expected value takes the minimum value of the position expectation and the speed expectation: .

[0067] S302. Risk Value Assessment: This value represents the likelihood of the recipient making an incorrect decision and the consequences of that incorrect decision. The risk value is defined using the definition given by the National Institute of Standards and Technology (NIST) of the United States. .in, The calculation is based on the angle between the velocity directions of the sender and the receiver ,

[0068] .

[0069] Angle It can be calculated that:

[0070] ,

[0071] in, 、 and 、 is the speed of the sender and receiver and The velocity components on the horizontal and vertical axes, and .

[0072] The calculation of is based on the distance between vehicles. When the distance between vehicles is closer, the consequences of wrong decisions may be more serious; and as the distance increases, the impact gradually decreases. The specific form is as follows:

[0073] .

[0074] S303. Confidence value evaluation: Consider the impact of the social role of the vehicle (such as police car, ambulance, etc.) on the credibility of the message. Messages broadcast by vehicles with high role authority are generally considered more credible, and therefore given a higher confidence. Since this embodiment is mainly used to identify malicious vehicles on the road, all vehicle roles are treated equally and are regarded as ordinary vehicles on the road. Therefore, the confidence value of all vehicles is the same, which is 0.5, that is, .

[0075] According to the three-factor trust model, using expected value , risk value , confidence value relationship A trust domain is derived. If the condition is met, the message is considered trustworthy, otherwise it is not trustworthy.

[0076] S4. Based on the transfer learning model TranferMegaCRN, an improved Bayesian method and a density-based clustering screening algorithm are used to design a vehicle trust value calculation method and construct a vehicle network trust management model.

[0077] S401. Design a vehicle trust value calculation method using an improved Bayesian method and a density-based clustering screening algorithm.

[0078] Each message sent, after vehicle identity authentication and authenticity verification, is used to assess the vehicle's trust value. The vehicle's trust value consists of two components: local trust and global trust. Local trust is further composed of direct trust and indirect trust.

[0079] S402. Use the improved Bayesian method to calculate the direct trust of the vehicle. Based on the Bayesian method to calculate the local trust method, first use the sliding window method and time decay factor to reduce the amount of data processed by the vehicle, and then use the normalized variance to design a nonlinear risk adjustment method to make the trust value decrease rapidly when the uncertainty is high.

[0080] In this embodiment, it is assumed that the vehicle With vehicle There is interaction (i.e. vehicle To the vehicle Providing message services), The number of true messages sent is preset in advance , Received Times from News, The proportion of true news is , then this experiment is recorded as , its likelihood function is as follows:

[0081] .

[0082] Since this experiment is similar to a coin toss experiment, to describe this situation, we can use distributed for The prior distribution of , the prior probability density is as follows:

[0083] ,

[0084] Before the interaction begins, that is, when the vehicle just enters the environment Uniform prior.

[0085] According to Bayes' theorem, the posterior distribution It is determined by the product of the prior distribution and the likelihood function: . Thus we get a new distributed ,parameter , At this time, according to Distribution mean formula , variance formula New The posterior mean under the distribution and variance In order to distinguish it from the calculation obtained by the following improved method, the posterior mean is and variance Denoted as and .

[0086] In order to prevent the computational overhead from gradually increasing, which may exceed the capacity of the on-board unit (OBU) and thus affect the real-time performance of the trust evaluation between vehicles, this embodiment uses a sliding window and a time decay factor to reduce the amount of data processed by the vehicle and regulate the trust value calculation. The window size is set to , then there is a queue:

[0087] ,

[0088] ,

[0089] Among them, the queue and The total interaction data and real message data in the two car windows were recorded separately. Indicates the Interaction exists, Indicates the Each interaction is a real message. For each new interaction, data is added to the end of the queue, and older data that exceeds the window length is removed. In other words, when a new interaction occurs, the oldest interaction information stored in the window is removed, the new interaction information is introduced, and both queues are updated simultaneously. The new queue is used to perform Bayesian calculations.

[0090] Therefore, the new direct trust value can be obtained by the following calculation :

[0091] ,

[0092] ,

[0093] ,

[0094] ,

[0095] in, and are the total number of interactions and the real number of interactions within the window, respectively; and for Prior parameter, initially 1; It is a time decay factor used to control the weight of historical trust values. It is initially set to 0.7, giving more weight to historical interactions.

[0096] This embodiment also uses variance to further adjust direct trust. Variance represents the uncertainty of trust value. A large variance indicates that the vehicle For vehicles The trust value of has large fluctuations during calculation, which may indicate attack behavior. This embodiment combines the trust value with its variance to obtain a risk-adjusted trust score, which reduces the trust evaluation under high uncertainty. First, the variance is normalized to establish a unified evaluation benchmark. The variance of the distribution is , its maximum value is 0.25, and normalization operation is performed Then the normalized variance is nonlinearly amplified to enhance the discrimination of small variance. The nonlinear amplification function is ,in is the variance amplification coefficient. Then, the exponential decay method is used to punish the clear high-volatility trust to obtain the final direct trust value. as follows:

[0097] ,

[0098] In order to facilitate the subsequent introduction, the subsequent Recorded as .

[0099] S403. Design a density-based clustering screening algorithm to calculate indirect trust.

[0100] The source of indirect trust is the neighboring vehicles, for example, there are three vehicles 、 、 , if there are two direct trust and ,but You can refer to right Therefore, this embodiment establishes a triple set, for each pair of vehicle nodes , collect all possible recommenders , the corresponding triples are as follows In addition, this triple must satisfy 、 、 、 ,The purpose is to ensure that the neighbors do not contain the source vehicle and destination vehicle Both parties; source vehicle and neighbors Communication exists, neighbors With the destination vehicle Communication exists. However, not all nodes are trustworthy. A malicious recommending node might deliberately discredit the target node's behavior; on the other hand, if the recommending node works well with the target node, it might give the target node a higher rating. In other words, the evaluation of a recommending node can be influenced by its relationship with the target node, so these evaluations cannot be fully relied upon.

[0101] In this embodiment, in order to prevent the subjective consciousness of the vehicle from affecting the indirect trust, the vehicle needs to send the direct trust to the RSU through V2I communication for processing, and a density-based clustering screening scheme is proposed to screen and eliminate some potential malicious recommenders.

[0102] First, extract a pair of cars and All triples of , extract the feature vector Then, randomly select vectors as the initial centroids of the cluster, and ensure that each initial centroid corresponds to a different data point to avoid duplication. , calculate it to each centroid separately Euclidean distance , and assign the data point to the cluster corresponding to the nearest centroid Then, the centroid of each cluster is updated according to the newly divided clusters. For each cluster m1, the mean of the feature vectors of all its members is calculated: ,in , is the set of data points of the th cluster, is the number of members of cluster m1, and then the new cluster centroid is obtained If a cluster has no members (e.g., due to improper initialization) or the centroid changes, a new data point is randomly selected as the new centroid. The clustering process ends when all centroids are no longer updated or the number of iterations reaches the maximum.

[0103] Since the triples constructed during clustering are based on direct trust, the values ​​in all triples will be affected by direct trust. An attacker may tamper with the direct trust value between himself and the normal car, resulting in an extreme trust value of 0. Often, a high-density cluster means that the features between members are highly similar, and a collaborative attacker will produce a dense and similar feature distribution to forge trust values. Therefore, the extreme value distribution may cause the feature vectors of some good cars and the attacker to be mistakenly classified into the same cluster. Therefore, this embodiment will select the cluster with the highest density for elimination (mark this cluster number as ), while retaining the recommenders in other clusters. The density of a cluster is the inverse of the average Euclidean distance between members:

[0104] ,

[0105] in, For the A collection of data points for a cluster; For the The recommenders in the remaining clusters are marked as valid for subsequent calculations. Therefore, for each pair of cars and , the valid recommender set is: In the valid recommender set, you can Source vehicle Design different recommendation weights, i.e. vehicle To recommenders The trust weight of the vehicle is quantified based on the number of historical interactions. To recommenders Trust weight:

[0106] ,

[0107] in, For vehicles Receive the vehicle The number of real messages; For vehicles The number of real messages received from all recommenders, i.e. Then, the source vehicle is obtained by weighted aggregation Target vehicle Indirect trust:

[0108] .

[0109] S403. Calculate local trust by weighted aggregation of direct trust and indirect trust.

[0110] In this embodiment, after calculating the indirect trust, the RSU combines the direct trust with the recommended trust to obtain the local trust. Target vehicle Local trust yes:

[0111]

[0112] in, is the weight factor. It can be adjusted dynamically based on direct trust confidence. The more historical interactions, The higher the value, the more accurate the trust evaluation will be.

[0113] S404. Calculate global trust using the PageRank-based VehicleRank algorithm.

[0114] Although local trust can reflect the trust relationship between every two vehicles, in the complex and large-scale urban Internet of Vehicles environment, this point-to-point trust alone is difficult to meet the interoperability requirements of the Internet of Vehicles. Therefore, global trust must be introduced to uniformly manage and coordinate various trust entities to achieve network-wide security incident response and thus ensure the comprehensive security and reliability of the entire Internet of Vehicles system. Therefore, after the RSU obtains the aggregated local trust, the RSU will periodically send the local trust to the TA, which will calculate the global trust using the VehicleRank method and publish the global trust. The RSU or vehicle can make autonomous decisions based on the global trust. The specific steps of this method include:

[0115] Use all local trust LTVs to normalize them into probability weights and build a trust matrix , whose elements Defined as:

[0116] ,

[0117] in, All vehicles the collection of vehicles for which information was provided;

[0118] If the vehicle Because no other vehicle provides information to it and fails to trust other vehicles, the vehicle is considered a hanging node. To avoid the hanging node vehicle The resulting trust allocation is interrupted, using the global trust vector from the previous round Correct the trust matrix (if the iterative calculation has not yet started, use the initial uniformly distributed global trust to correct the trust matrix. For example, in an environment with 200 cars, initialize the global trust of each car to 0.005 to obtain the initial global trust vector . ), forming a new matrix:

[0119] ,

[0120] Among them, the superscript T represents transposition, a 0 is the suspension vehicle indicator vector, vehicle For suspended vehicles =1, otherwise =0.

[0121] In order to ensure the convergence of global trust calculation, a welfare mechanism is introduced to transform the matrix S into a non-negative, irreducible, and aperiodic matrix W:

[0122] ,

[0123] in, is a scaling parameter to enhance the welfare impact and control the calculation weight, which is set to 0.45 in this embodiment; is a column vector of all ones, indicating that the welfare is evenly distributed to all vehicles.

[0124] Based on the above matrix W, the power method is used to iteratively update the global trust GTV, and the global trust GTV is initialized by even distribution, that is, , iterate as follows:

[0125] ,

[0126] Until the difference between adjacent iteration results Less than threshold , it is considered to be convergent.

[0127] Finally, TA broadcasts the global trust value of the vehicle.

[0128] S5. Use the Internet of Vehicles trust management model to complete vehicle trust assessment.

[0129] In this embodiment, based on the aforementioned interactive authentication module, true and false message identification method and trust value calculation method, a security-enhanced Internet of Vehicles trust management model framework combined with transfer learning is constructed, such as Figure 3 As shown in Figure 2, this framework can be used to complete vehicle trust assessment.

[0130] In this embodiment, vehicles are divided into multiple categories:

[0131] (1) Good Car: A car that behaves completely normally, only sends real messages, strictly maintains data integrity, and broadcasts its own calculated data. If an attacker exhibits obvious incorrect behavior, the good car will remember the attacker and reduce its direct trust in the attacker to zero.

[0132] (2) Bad car: A vehicle with certain abnormal behaviors has a certain probability of sending true messages and a certain probability of sending false messages while driving. Similarly, a bad car also strictly maintains data integrity and broadcasts the data calculated by itself.

[0133] (3) Attacker: A malicious vehicle, usually hidden in a disabled car. While driving, it has a certain probability of sending not only true messages but also false messages. It also destroys data integrity, tampers with its own calculated data, and broadcasts this data, interfering with the judgment of other participants in the environment.

[0134] The effectiveness of this TSE-TMM framework is verified in the following four different situations.

[0135] Scenario A: There are 200 vehicles in the environment, 70 of which are good and 130 are bad (no attackers). All vehicles behave normally, and there are no speeding issues. In this scenario, the vehicle's trust value is determined entirely by the authenticity of the interaction message and is unrelated to driving behavior.

[0136] Scenario B: There are 200 vehicles in the environment, 70 of which are good and 130 are bad (including 50 attackers). The attackers are constantly speeding. These speeding behaviors cause good vehicles to perceive them as attackers, reducing their direct trust in them to 0. To prevent a rapid decline in trust, the attackers collaborate with one another to launch tampering attacks, falsifying their trust in other vehicles. This alters their trust in good vehicles to 0 and their trust in other attackers to 1, disrupting the trust system.

[0137] Scenario C: There are 200 vehicles in the environment, 70 of which are good vehicles and 130 are bad vehicles (including 50 attackers). The attackers sometimes exceed the speed limit and sometimes drive normally. This scenario introduces the concept of speed factor. Each vehicle has a speed factor. For example, if the speed limit is 60 km / h, a speed factor of 0.8 results in a maximum speed of 48 km / h; a speed factor of 1.2 results in a maximum speed of 72 km / h. Since the attackers intermittently exceed the speed limit, when observed by good vehicles, they will identify them as attackers and reduce their direct trust in these vehicles to zero. Simultaneously, the attackers will continue to launch tampering attacks, just as in Scenario B.

[0138] Scenario D: There are 200 vehicles in the environment, 70 of which are good vehicles and 130 are bad vehicles (including 50 attackers). The attackers do not exceed the speed limit. Since the attackers drive normally, the good vehicles will no longer be able to directly detect the attackers. The attackers will be hidden in the environment, but the attackers' tampering attacks will not stop.

[0139] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A security-enhanced Internet of Vehicles trust assessment method combined with transfer learning, characterized in that the steps include: Based on urban road conditions, build an urban road vehicle network model; Based on the urban road vehicle network model, an interactive authentication module based on the elliptic curve Diffie-Hellman protocol and the elliptic curve digital signature algorithm is designed; Based on the interactive authentication module, a method for identifying true and false messages based on the three-factor trust model improved by the transfer learning method is introduced. In the speed expectation sub-factor analysis, a transfer learning model TranferMegaCRN is designed to obtain the predicted speed of the message sender; TransferMegaCRN adds several fully connected layers in the input and output stages of the entire transfer learning model; in the input stage, the original speed data is input x and time characteristic covariates y 1. Each of the data passes through its own fully connected layer and is then input into the MegaCRN model. In the MegaCRN model, the parameters of all encoder modules in the MegaCRN pre-trained on the traffic flow dataset METR-LA are frozen. A modular progressive unfreezing strategy is adopted to unfreeze the decoder and memory modules. During the unfreezing process, the deep layers of the encoder are gradually unfrozen starting from the fifth round, and the middle layers are continued after the tenth round, gradually adjusting from fixing the bottom layer to dynamically releasing the upper layers. At the output stage, the output of the MegaCRN model is further processed through a fully connected layer. Based on the transfer learning model TranferMegaCRN, an improved Bayesian method and a density-based clustering screening algorithm are used to design a vehicle trust value calculation method and build a vehicle network trust management model; Based on the Internet of Vehicles trust management model, vehicle trust assessment is completed.

2. The security-enhanced Internet of Vehicles trust assessment method combined with transfer learning according to claim 1, characterized in that: The constructed urban road vehicle network model includes four main bodies: vehicles equipped with on-board units OBU, roadside units RSU, trusted control centers TA and road speed monitors.

3. The security-enhanced Internet of Vehicles trust assessment method combined with transfer learning according to claim 1, characterized in that: The steps of designing the vehicle trust value calculation method include: in direct trust value calculation, designing a sliding window method with a time decay factor and a variance-based risk adjustment mechanism to optimize the calculation method based on the Bayesian formula; in indirect trust calculation, designing a density-based clustering screening algorithm to extract trust feature vectors from RSU and eliminate potential malicious recommendation nodes in high-density clusters.

4. The security-enhanced Internet of Vehicles trust assessment method combined with transfer learning according to claim 1, characterized in that: The method for constructing the Internet of Vehicles trust management model includes: Based on the vehicle trust value calculation method, a vehicle network trust management model framework is constructed; Introducing an interactive authentication module based on the elliptic curve Diffie-Hellman protocol and the elliptic curve digital signature algorithm into the management model framework; The three-factor message recognition method improved based on the transfer learning method is introduced into the management model framework to complete the construction.

5. The security-enhanced Internet of Vehicles trust assessment method combined with transfer learning according to claim 4, characterized in that: The steps of introducing the interactive authentication module of the elliptic curve Diffie-Hellman protocol and the elliptic curve digital signature algorithm include: when registering a vehicle, the elliptic curve Diffie-Hellman protocol is used to negotiate a shared key with the control center TA, encrypt the real ID to achieve anonymous communication, and ensure that TA can trace its identity; during the communication process, the elliptic curve digital signature algorithm is used to sign and verify all messages. If the receiver fails the verification, the sender is marked as a malicious node.

6. The security-enhanced Internet of Vehicles trust assessment method combined with transfer learning according to claim 4, characterized in that: The transfer learning method is combined with the time series model based on the meta-graph convolutional recurrent network MegaCRN to construct a transfer learning model TransferMegaCRN to predict the speed of the message sender and obtain the speed expectation sub-factors.

7. The security-enhanced Internet of Vehicles trust assessment method combined with transfer learning according to claim 6, characterized in that: Based on the TransferMegaCRN speed prediction model, an improved true-false message identification method based on three factors: expectation, risk, and confidence. The receiving vehicle calculates the expectation value, risk value, and confidence value determined by the speed and position of the sending vehicle to comprehensively evaluate the credibility of the message. When the dynamic balance condition is met, the message is judged to be authentic; otherwise, it is marked as unreliable.

Images