A node security trust evaluation method suitable for vehicle-mounted and sensing edge cloud MANET

Abstract

The application provides an energy game multi-component node security trust evaluation method suitable for a vehicle-mounted and sensing edge cloud MANET, and belongs to the technical field of mobile self-organizing network security. The method acquires one-hop neighbor topology, energy attributes and historical interaction records of an evaluation node; divides node energy into self-transaction reserved energy and forwarding service distributed energy, constructs a two-dimensional Bayesian game utility function and solves Nash equilibrium to obtain a game trust component; calculates an observation trust component based on the interaction records; dynamically determines a weighting factor according to the interaction frequency, fuses the two types of components to obtain a comprehensive trust degree, and dynamically updates the comprehensive trust degree with topology changes and new interactions. The application can solve the problem of new node cold start evaluation distortion, adapt to high dynamic topology, and improve network attack resistance and transmission reliability.

Description

Technical Field

[0001] This invention belongs to the field of mobile self-organizing network security technology, and relates to a node security trust assessment method applicable to vehicle-mounted and sensor edge cloud MANET. Background Technology

[0002] With the deep integration of cloud computing and mobile ad hoc networks, a large number of lightweight, autonomous distributed nodes are deployed in scenarios such as field monitoring, vehicle networking, and edge cloud collaboration. In environments with no central control, highly dynamic topology, limited resources, and anonymous autonomy, they undertake data forwarding, distributed computing, and collaborative cloud service tasks, becoming an important support for the efficient operation of cloud services.

[0003] Unlike traditional centralized server architectures, MANET distributed nodes possess four key characteristics. First, decentralized autonomy: there is no unified, trusted management node across the network. Nodes interact peer-to-peer, autonomously completing routing, forwarding, and collaborative services, lacking a global authority to uniformly oversee node behavior. Second, strong resource constraints: most distributed nodes rely on battery power, with limited storage, computing power, and bandwidth resources, naturally leading to a self-interested tendency to conserve energy and reduce forwarding overhead. Third, highly dynamic topology: nodes are mobile, frequently experiencing disconnections and reconstructions of neighbor links due to location changes, signal obstruction, and power depletion, resulting in continuously changing distributed collaborative relationships. Fourth, uncontrollable anonymity: there is no mandatory identity verification mechanism for node entry, allowing unfamiliar nodes to access the network at any time, and malicious nodes can easily impersonate legitimate distributed nodes to infiltrate the network environment.

[0004] Based on the above characteristics, two typical security risks arise during distributed node collaboration. Some nodes exhibit clearly selfish behavior to reduce their own overhead, selectively discarding data packets and falsely reporting remaining energy to evade collaboration after receiving forwarding requests. Malicious distributed nodes launch attacks through continuous packet loss, data tampering, falsifying energy parameters, node collusion, and identity spoofing, disrupting distributed data transmission links. This leads to unstable cloud service data transmission, a significant decline in service quality, and even malicious security incidents such as business privacy leaks and edge cloud service paralysis, seriously threatening the trustworthiness and operational security of cloud services in converged scenarios. (See attached manual.) Figure 1 As shown, in the MANET node interaction and neighbor relationship topology, selfish and malicious nodes are mixed among normal nodes, posing a direct challenge to the stability of distributed collaboration.

[0005] Trust assessment is a key technology for identifying the trustworthiness of distributed nodes, restraining speculative and malicious behavior, and preventing network attacks. It is also an important prerequisite for building a secure and reliable distributed collaborative environment. Existing trust assessment methods mostly rely on historical interaction records, calculating trust values ​​by statistically analyzing indicators such as forwarding success rate and interaction frequency. While these methods have some reference value when sufficient interaction data is available, they are clearly insufficient for the inherent characteristics of distributed nodes.

[0006] Pure interactive observation-based trust assessment schemes rely solely on forwarding interaction records acquired by the node itself to calculate trust values, employing a fixed-weight, globally cumulative statistical approach to evaluate node trustworthiness. This type of scheme utilizes promiscuous mode on network interface cards (NICs) to monitor neighbor forwarding behavior, using the proportion of actual forwarded data packets to total requested data packets within a statistical window as the sole basis for trust. New nodes joining the network without prior interaction history are uniformly assigned a neutral, fixed initial trust value. Its inherent flaws include: a prominent cold start problem, lack of objective assessment criteria for unfamiliar new nodes without interaction records, and the vulnerability of the fixed, neutral initial value to malicious attacks; a single assessment dimension, completely ignoring private key attributes such as remaining node energy and resource allocation, making it impossible to predict potential self-serving behavior; the use of static assessment logic without a historical interaction decay mechanism, allowing outdated interaction records to continuously interfere with the current trust result; and a lack of energy misrepresentation and collusion attack identification and verification mechanisms, resulting in distorted assessment results and poor network robustness when facing attacks such as nodes falsifying energy parameters and coordinated malicious packet loss.

[0007] Single-game trust assessment schemes introduce game theory to construct trust models, relying solely on game theory deduction based on node private attributes to output trust values, without integrating actual interaction observation data. These schemes construct game utility functions based on static attributes such as node energy and location, solve for Nash equilibrium to obtain node cooperation payoffs, and directly map these payoffs to trust values, without considering actual data packet forwarding behavior throughout the process. Their inherent drawbacks include: relying solely on static attribute deductions during long-term network operation, they cannot dynamically capture real-time changes in node forwarding behavior, leading to a lag in identifying nodes that transition from normal to selfish or malicious; lacking an adaptive weight adjustment mechanism, the game component weights remain fixed throughout, failing to amplify the assessment proportion of real interaction behavior as interactions increase, resulting in consistently low assessment accuracy; and lacking sliding window and behavior penalty mechanisms, they cannot implement trust penalties for nodes with high packet loss or passive forwarding, making it difficult to constrain node speculative behavior.

[0008] The fixed-weighted fusion trust scheme, which integrates game theory and observation, attempts to combine two types of information: game theory deduction and interactive observation. However, it uses manually preset fixed weights to complete the trust weighting fusion, and the weight values ​​cannot be changed throughout the process. This type of scheme calculates the game theory trust component and the observation trust component separately, and then uses a constant fixed weight to sum them to obtain a comprehensive trust value. Its inherent defects are that the weights cannot be adaptively adjusted according to the number of interactions, the quality of interactions, and the node activity. In the cold start phase, the excessively high weight of the observation component leads to the failure of trust assessment for new nodes, and in the stable interaction phase, the excessively high weight of the game theory component weakens the reference value of real behavior. It does not distinguish between the node's own maintenance energy and forwarding service energy, and the energy game utility function model is crude, making it impossible to accurately quantify the node's forwarding cost and cooperation benefits. The trust update mechanism is simplistic, only supporting timed batch updates, lacking real-time update logic triggered by topology changes and malicious abnormal behavior, resulting in insufficient timeliness of trust values ​​when the network topology changes rapidly. It also lacks supporting mechanisms such as energy cross-validation and collusion behavior identification, and cannot suppress the behavior of nodes falsely reporting remaining energy to evade forwarding cooperation.

[0009] In summary, existing node trust assessment technologies cannot simultaneously solve a series of technical problems in MANET cloud convergence scenarios, such as distorted assessment of new node cold start, single assessment dimensions, static and fixed weights, weak resistance to energy false alarms and collusion attacks, and poor adaptability to dynamic topology. Therefore, they are unable to meet the secure and reliable forwarding requirements of highly dynamic and resource-constrained distributed networks. Summary of the Invention

[0010] To address the shortcomings and deficiencies of existing technologies, this invention provides a node security trust assessment method applicable to vehicle-mounted and sensor edge cloud MANETs. This method constructs a dual-component trust assessment framework including game theory inference components and direct observation components. In the cold start phase, when there are no historical interaction records, node energy is divided into two dimensions: energy reserved for its own transactions and energy allocated for forwarding services. A Bayesian game utility function in a two-dimensional policy space is constructed, and the maximum expected payoff of the node is obtained by solving the Bayesian Nash equilibrium. Based on this, the game theory trust component is calculated as the initial trust basis. As interaction behavior accumulates, the packet forwarding behavior of neighboring nodes is statistically analyzed based on a sliding time window. A punitive deduction is applied to negative forwarding behavior, and the observation trust component is calculated. Furthermore, a weighting factor is dynamically determined based on the interaction frequency between the assessed node and the target neighboring node. This weighting factor increases with the interaction frequency, thus achieving a smooth transition from game theory-based assessment in the cold start phase to direct observation-based assessment in the stable interaction phase. The two types of trust components are weighted and fused using the weighting factor to obtain a comprehensive trust level, which is updated in real time in response to changes in neighbor topology or the addition of new interaction records. The method also provides a multi-neighbor cross-verification mechanism for node energy false reporting behavior, a collusion attack identification mechanism based on multi-dimensional behavior correlation coefficients and two-level threshold hierarchical judgment, and a lightweight equilibrium solution method for resource-constrained nodes, comprehensively improving the accuracy, dynamic adaptability and anti-attack capability of node trust assessment in mobile ad hoc network environments.

[0011] The specific technical solution adopted by this invention to solve its technical problem is as follows:

[0012] This invention aims to solve the technical problems existing in the current mobile ad hoc network node trust assessment technology, such as cold start trust assessment distortion, single assessment dimension, static and fixed weights, weak anti-attack capability, and insufficient dynamic topology adaptability. It provides a node security trust assessment method and device suitable for vehicle-mounted and sensor edge cloud MANET.

[0013] The node security trust assessment method provided by this invention specifically includes the following steps:

[0014] The evaluation node acquires its one-hop neighbor topology information, the energy attribute information of each neighbor node, and historical interaction records. In a decentralized MANET environment without centralized control, nodes can only obtain local topology and neighbor status information within their one-hop communication range. These three types of information constitute the basic data input for trust evaluation, adapting to the perception capability boundaries of distributed nodes.

[0015] For each neighboring node, based on the energy attribute information, the node's energy is divided into two dimensions: energy reserved for its own transactions and energy allocated for forwarding services. A Bayesian game utility function in a two-dimensional policy space is constructed, and the maximum expected payoff of the node is obtained by solving the Bayesian Nash equilibrium. Based on the maximum expected payoff, the game trust component of the neighboring node is calculated. A node's total energy resources are limited; it needs to reserve some energy to support its own computing, sensing, and other basic transactions, and also allocate some energy to forward data from other nodes to maintain network cooperation. The two-dimensional energy partitioning can more accurately depict the trade-off between resource conservation and cooperation benefits. In the cold start phase without historical interactions, the Bayesian game model can infer the node's willingness to cooperate based on the probability distribution of the node's energy attributes under conditions of information asymmetry, outputting an objective initial trust estimate without the need for manually preset fixed neutral initial values.

[0016] Based on the historical interaction records, the packet forwarding behavior of neighboring nodes is statistically analyzed, and the observed trust component of that neighboring node is calculated. The forwarding behavior data obtained through direct observation serves as empirical evidence for evaluating the trustworthiness of nodes and can truly reflect the actual collaborative performance of nodes.

[0017] The weighting factor is dynamically determined based on the interaction frequency between the evaluation node and its neighboring nodes. This weighting factor corresponds to the weight of the observation trust component, and its value increases with the interaction frequency. A higher interaction frequency means more accumulated empirical evidence, and the reference value of the observation component is correspondingly improved. This dynamic adjustment mechanism enables a smooth transition from game theory-based deduction in the cold start phase to actual observation-based interaction in the stable interaction phase.

[0018] The game-theoretic trust component and the observed trust component are weighted and fused using the weighting factor to obtain the comprehensive trust level of the neighbor node.

[0019] In response to changes in neighbor topology or the addition of new interaction records, the overall trust level is updated to ensure that the trust assessment results remain synchronized with the real-time network status.

[0020] Furthermore, the specific method for calculating the game trust component of the neighbor node based on the maximum expected return is as follows: calculate the maximum expected return when the evaluation node participates in the game with all neighbor nodes, and the maximum expected return after removing the target neighbor node. The difference between the two is taken as the marginal contribution of the target neighbor node. When the marginal contribution is greater than 0, the game trust component is the ratio of the marginal contribution to the maximum expected return; when the marginal contribution is less than or equal to 0, the game trust component is 0. The marginal contribution can quantify the net impact of a single neighbor node on the overall cooperation benefit. A positive contribution indicates that the node's participation can improve the overall cooperation benefit, corresponding to a higher level of trust. A non-positive contribution indicates that the node has no positive effect on the cooperation benefit, corresponding to a low level of trust. This mapping of trust values ​​conforms to the inherent logic of trust assessment.

[0021] Furthermore, when the neighbor set corresponding to the one-hop neighbor topology information does not satisfy the fully connected condition, i.e., there are nodes within the neighbor set that are not within each other's wireless communication range, the game computation space is pruned and adapted: a fully connected subgraph in the neighbor set is selected as the game participant set for multi-player game solving; if no two nodes in the neighbor set have a direct link, the multi-player game is decoupled into multiple independent bilateral games, and the game trust component is calculated based on the maximum expected payoff of each bilateral game. The effective implementation of multi-player Bayesian games presupposes that all participating nodes are reachable. This pruning and adaptation mechanism can ensure the computational effectiveness of the game model under different topologies, avoiding model failure and wasted computing power caused by topological disconnection.

[0022] Furthermore, the observed trust component is calculated statistically based on a sliding time window, and its value is the ratio of the number of successfully forwarded data packets to the weighted total. The weighted total is the sum of the number of successfully forwarded data packets and the penalty term. The penalty term is the product of a preset penalty coefficient and the number of packet losses. The number of packet losses is the difference between the total number of forwarding requests and the number of successfully forwarded data packets, where the penalty coefficient is not less than 1. The sliding time window can gradually weaken the influence of long-term interaction records, making the evaluation results more consistent with the recent behavior of the node. The penalty term reduces trust by amplifying the denominator. The more packet losses, the lower the observed trust component, which can effectively constrain passive forwarding and malicious packet loss behavior. At the same time, this calculation method can ensure that the value of the observed trust component is always within the range of 0 to 1, and the value converges stably.

[0023] Furthermore, the weighting factor is determined as follows: the proportion of the interaction frequency of the target neighbor node within the sliding time window to the total interaction frequency of all neighbor nodes of the evaluation node is used as the weighting factor. Two types of boundary handling rules are also set: when the evaluation node has no historical interaction records within the sliding time window, the weighting factor is 0, and the overall trust level is entirely determined by the game-theoretic trust component; when the evaluation node has only one neighbor node, the weighting factor is 1, and the overall trust level is entirely determined by the observed trust component. These two types of boundary rules correspond to zero-interaction cold start and single-neighbor extreme topology scenarios, respectively, ensuring the logical consistency of the solution under various network conditions.

[0024] Furthermore, when a neighbor topology change is triggered, the game computation space is adjusted synchronously: when a new neighbor node is added, the set of game participants is expanded and the equilibrium is resolved; when a neighbor node leaves, the set of game participants is reduced and the maximum expected payoff is recalculated; when the topology change leads to a change in the full connectivity condition, the corresponding game space pruning or decoupling operation is performed. The addition or removal of the neighbor set changes the composition of the game participants and the strategy space. Synchronously adjusting the game computation space can maintain the matching degree between the trust assessment results and the current network topology, ensuring the timeliness and accuracy of the assessment results.

[0025] Furthermore, the method also includes an energy false alarm verification step: receiving the remaining energy value reported by neighboring nodes and comparing it with the locally calculated remaining energy record value; if the reported remaining energy value is greater than or equal to the record value, it is determined to be an energy false alarm behavior; after confirmation through multi-neighbor cross-verification, a trust discount penalty is applied to the node with energy false alarm behavior, and the comprehensive trust level of the corresponding node is updated by multiplying it by a preset discount factor. The locally calculated remaining energy is calculated based on historical forwarding records and the cumulative energy consumption per unit data packet forwarding. Since the node's energy is continuously consumed with forwarding behavior, under normal circumstances, the reported value should not be higher than the calculated record value, and using this as a judgment basis has physical rationality; multi-neighbor cross-verification can further improve the judgment accuracy, and the discount penalty can effectively suppress the selfish behavior of nodes falsely reporting energy to evade forwarding cooperation.

[0026] Furthermore, the method also includes a collusion attack identification step: statistically analyzing the packet loss rate, energy reporting deviation, trust fluctuation amplitude, and interaction frequency density of each neighbor node within a sliding time window; synchronizing the average statistical indicators of third-party nodes among one-hop neighbors; calculating the multi-dimensional behavioral correlation coefficient between any two nodes targeting the same target node; identifying collusion attack nodes through a two-level threshold hierarchical determination, wherein the behavioral correlation coefficient threshold in the two-level threshold is adaptively adjusted based on the ratio of the total number of local one-hop neighbors to the total number of nodes in the entire network; and imposing trust penalties on nodes determined to be colluding attacks. Multi-dimensional indicators can capture the collaborative anomaly characteristics of nodes from multiple levels, including forwarding behavior, energy reporting, and trust changes; the two-level threshold determination can gradually narrow the scope of suspicion and improve the identification accuracy; the threshold is dynamically adjusted with node density, which can adapt to the detection needs under different network deployment scales and balance the false negative rate and false positive rate of detection.

[0027] Furthermore, the process of solving the Bayesian Nash equilibrium to obtain the maximum expected return of a node can employ at least one lightweight processing method adapted to resource-constrained nodes: a numerical iterative method with a maximum number of iterations and a convergence accuracy threshold, terminating the calculation when the change in return is less than the convergence threshold or the iteration limit is reached; a lookup method with a pre-stored benchmark equilibrium strategy, pre-stored benchmark strategies for common neighbor sizes and energy ranges, and obtaining the result after a small correction after table lookup at runtime; and a complexity reduction calculation method using a simplified utility function that compresses nonlinear calculation terms, retaining the core logic of two-dimensional energy allocation, and exchanging a slight decrease in accuracy for a significant reduction in computational overhead. All three processing methods reduce real-time computing power consumption while retaining the core design logic, adapting to the characteristics of limited computing power and energy resources in vehicle-mounted and sensor edge nodes, and improving the engineering feasibility of the solution.

[0028] Accordingly, the present invention also provides a node security trust assessment device suitable for vehicle-mounted and sensor edge cloud MANET, including an information acquisition module, a game calculation module, an observation calculation module, a weighted fusion module and an update execution module.

[0029] The information acquisition module is used to acquire the one-hop neighbor topology information of the evaluation node, the energy attribute information of each neighbor node, and historical interaction records.

[0030] The game calculation module is used to divide the energy of each neighbor node into two dimensions based on the energy attribute information: its own transaction retention energy and forwarding service allocation energy. It constructs a Bayesian game utility function in a two-dimensional policy space, solves the Bayesian Nash equilibrium to obtain the node's maximum expected payoff, and calculates the game trust component of the neighbor node based on the maximum expected payoff.

[0031] The observation calculation module is used to calculate the observation trust component of a neighboring node by statistically analyzing the packet forwarding behavior of neighboring nodes based on the historical interaction records.

[0032] The weighted fusion module is used to dynamically determine a weighting factor based on the interaction frequency between the evaluation node and its neighboring node. The weighting factor corresponds to the weight of the observed trust component, and the value of the weighting factor increases with the increase of the interaction frequency. The game trust component and the observed trust component are weighted and fused using the weighting factor to obtain the comprehensive trust level of the neighboring node.

[0033] The update execution module is used to update the overall trust level in response to changes in neighbor topology or the addition of new interaction records.

[0034] Compared with the prior art, the present invention and its preferred embodiments have at least the following beneficial effects:

[0035] In the cold start phase without historical interaction records, Bayesian game theory is performed based on the node's private two-dimensional energy attributes to output an objective initial trust value, thus eliminating the security vulnerability of unfamiliar malicious nodes infiltrating the network by using neutral initial values. As interaction data accumulates, a dynamic weighting factor that monotonically increases with the frequency of interaction smoothly transitions to an evaluation mode dominated by directly observed components, avoiding evaluation distortion caused by fixed weights and adapting to highly dynamic topology network environments.

[0036] By distinguishing between the energy reserved for its own transactions and the energy allocated for forwarding services, a game utility function is constructed, which accurately quantifies the forwarding cost and long-term cooperation benefits of nodes and can effectively predict the potential self-interest of nodes. The observation component adopts a penalty mechanism that amplifies the denominator, which effectively deducts the trust value of negative forwarding nodes while ensuring that the trust degree always converges within a reasonable range, thus improving the accuracy and robustness of the assessment.

[0037] It supports game space pruning and bilateral decoupling based on topology changes, ensuring the effectiveness and timeliness of multiplayer game computation in highly mobile environments; at the same time, it introduces lightweight solution methods, which significantly reduces the computing power overhead of resource-constrained nodes and enhances the feasibility of engineering implementation.

[0038] By cross-verifying energy reporting data with multiple neighbors and applying discount penalties, false energy reporting behavior is effectively suppressed. Based on a collusion identification mechanism with multi-dimensional behavior correlation coefficients and adaptive thresholds for node density, collusion attack groups that coordinate packet loss and forge forwarding records can be accurately detected, significantly improving the overall anti-attack capability and operational security of the fusion network. Attached Figure Description

[0039] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0040] Figure 1 This is a topological diagram illustrating the interaction and neighbor relationships between MANET nodes in the existing technology.

[0041] Figure 2 This is a diagram of the Bayesian energy game multi-component node trust evaluation model architecture according to an embodiment of the present invention.

[0042] Figure 3 This is a graph showing the change in the overall trust level of nodes with the number of evaluation rounds in an embodiment of the present invention.

[0043] Figure 4 This is a graph showing the change of the node game trust component with the evaluation rounds in an embodiment of the present invention.

[0044] Figure 5 This is a graph showing the change in overall trust level after the node packet loss rate is reduced in an embodiment of the present invention.

[0045] Figure 6 This is a graph showing the change in the game trust component after the node packet loss rate is reduced in an embodiment of the present invention.

[0046] Figure 7 This is a graph showing the impact of the self-transaction energy utilization rate on node trust when the forwarding energy utilization rate is 0.4 in an embodiment of the present invention.

[0047] Figure 8 The graph shows the impact of the self-transaction energy utilization rate on the node trust level when the forwarding energy utilization rate is 0.9 in this embodiment of the invention.

[0048] Figure 9 The graph shows the impact of forwarding energy utilization on node trust when the self-transaction energy utilization rate is 0.5 in this embodiment of the invention.

[0049] Figure 10 The graph shows the impact of forwarding energy utilization rate on node trust when the self-transaction energy utilization rate is 0.9 in this embodiment of the invention. Detailed Implementation

[0050] To make the features and advantages of the present invention more apparent and understandable, specific embodiments are described below in detail:

[0051] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0052] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0053] I. Overview of the Plan

[0054] This invention proposes a node security trust assessment method that integrates energy-based Bayesian game theory and interactive observation. The overall framework employs a trust calculation framework that fuses game theory deduction and direct observation. A Bayesian game utility function is constructed around the two-dimensional energy allocation of nodes, combined with a dynamically weighted factor that adaptively adjusts with interaction frequency, achieving a smooth transition in trust assessment from the cold start phase to the stable interaction phase. The solution also includes a multi-event-driven trust lifecycle update mechanism and a dual anti-attack system for energy false alarm cross-verification and joint identification of collusion attacks. This enables real-time identification of selfish and malicious nodes in decentralized, resource-constrained, and highly dynamic MANET environments, effectively solving the challenge of trust initialization for new nodes, enhancing network anti-attack capabilities, and providing reliable security trust guarantees for distributed cloud services and networking environments. Figure 2 As shown, the node security trust assessment model in this scheme is a closed-loop iterative architecture. It uses node attributes and network topology as its underlying foundation, node energy parameters as input for game theory calculations, and a neighbor discovery mechanism to complete topology awareness and neighbor set construction. The core of the model employs a dual-component parallel computing framework: on one hand, based on node attributes and topology information, it outputs a game-theoretic trust component through Bayesian energy game theory; on the other hand, based on actual forwarding behavior records, it outputs an observation trust component through interactive observation. Both trust components are input into a dynamic weighted fusion module. The adaptive weight module dynamically adjusts the weight ratio of the two components according to the interaction frequency of the forwarding behavior records, and outputs the node's comprehensive trust score after fusion calculation. After the comprehensive trust score enters the trust update and application stage, the updated node state and interaction data are fed back to the front-end computing module, realizing dynamic iterative optimization of the trust assessment results.

[0055] II. System Model and Basic Definitions

[0056] In a distributed MANET collaborative forwarding scenario, node forwarding behavior naturally presents a binary opposition between resource consumption and long-term collaborative benefits, making its forwarding decisions essentially a game-theoretic trade-off. From a cost perspective, forwarding cross-node data packets continuously consumes a node's limited private resources, such as power, wireless bandwidth, and computing power. Each forward shortens its own working time and consumes its own business transmission resources; rationally self-interested nodes will instinctively avoid unprofitable forwarding overhead. From a benefit perspective, proactively forwarding data for neighbors accumulates network trust and reputation. When a node subsequently needs to transmit data, other nodes are more willing to provide forwarding services, ensuring its communication connectivity. If a node consistently refuses to cooperate, it will be judged as a low-trust entity by surrounding nodes, making subsequent communication requests highly likely to be rejected, leading to isolation and network outages. Nodes cannot unilaterally decide their own communication security; they must make trade-offs between conserving their own resources and reducing immediate losses versus maintaining network cooperation and maximizing long-term communication benefits. All nodes autonomously adjust their forwarding strategies with the goal of maximizing their overall utility, creating a system of checks and balances that perfectly aligns with the core characteristics of game theory.

[0057] To standardize the model description, a standardized definition of the MANET network environment is first established. Node It is a node A neighbor is a neighbor if and only if the former is within the latter's wireless transmission range. The set of all neighbors is The quantity is This invention assumes that the wireless link of the network is bidirectional, that is, if yes The neighbors, then Too The neighbors. Undirected graph. Let V represent a mobile ad hoc network, where V is the set of all mobile nodes. Let be a set of edges. Two nodes that share an edge mean that they are within each other's wireless transmission range, i.e., they are neighbors.

[0058] node The attribute set at time scale t is .in for Energy utilization rate at time t for At time t and its neighbors The interaction history, where interaction refers to the actions of nodes forwarding data packets to each other or other services provided between them. The node at time t The number of neighboring nodes, It is up to time t. Its neighboring nodes Records of data forwarding behavior, among which Up to time t, for The number of data packets forwarded. At time t Require The total number of data packets forwarded. For ease of expression, when the time dimension is not emphasized, it can be... The attribute is abbreviated as The attributes of each node This is private information, known only to the node itself; other nodes only know about it. The probability distribution.

[0059] III. Basic Model of Two-Component Fusion Trust Assessment

[0060] Trust level can be defined as the subjective expectation of the evaluation node that the target node will fulfill its entrusted tasks in future interactions. Direct observation of the target node's actual behavior patterns is a key source of information for forming this expectation. Taking data packet forwarding as an example, direct observation means that the evaluation node entrusts the packets to be forwarded to the target node and continuously records the number of packets actually forwarded, thereby assessing its fulfillment capability. Monitoring adopts a watchdog approach, whereby the evaluation node sets its network interface card to promiscuous mode to listen to the data packets sent by neighboring nodes.

[0061] In the absence of direct observational evidence, existing mechanisms struggle to effectively assess target nodes. A typical scenario is when a new node joins the network without any historical interaction records with its neighbors, making it impossible to establish trust naturally. Current solutions typically assign a neutral initial value to unfamiliar nodes to represent their trustworthiness as being in an ambiguous intermediate state. While this simplifies the computation process, it introduces security vulnerabilities, making them susceptible to attack entry points by selfish or malicious nodes. Therefore, before making a forwarding decision, nodes can conduct game theory analysis based on observable attributes such as the other node's energy, location, and movement speed. This analysis forms a prediction of the other node's behavior, serving as a joint basis for trust assignment and action selection.

[0062] This invention decomposes the overall trust level of nodes into game theory inference components and direct observation components, and obtains the final trust assessment result through dynamic weighted fusion. The calculation formula is as follows:

[0063]

[0064] In the formula, For nodes For neighboring nodes Overall trust level; , is a weighting factor with a value range of [0,1], used to characterize the degree of preference for the two types of trust components; The game-theoretic component of trust level, representing Based on network topology characteristics and node attributes, nodes are analyzed. The game theory and deduction of trust value; The observed component of trust level is the node. By analyzing the nodes Trust assessment values ​​are derived from direct observation of behavior. Of the two components, the observation component, relying on firsthand information acquired by the node itself, holds a fundamental position in trust assessment; the game-theoretic component, as a priori prediction, mainly plays a supplementary role in stages where direct evidence is insufficient. In the cold-start stage, where there is no historical information, the weighting factor... =0, trust assessment relies entirely on game theory; as interaction increases, As the value gradually increases, the weight of the observed component gradually increases.

[0065] IV. Trust Component in Bayesian Game Theory Based on Two-Dimensional Energy Allocation

[0066] 4.1 Basic Framework of Bayesian Game Theory

[0067] Bayesian game models have significant advantages in information asymmetry, uncertainty decision-making, and private type inference. They can characterize the uncertainty of private attributes of distributed nodes through probability distribution, infer the true type of nodes through Bayesian inference when there is a lack of direct interaction evidence, solve the optimal cooperation strategy by combining Nash equilibrium, and give a reasonable initial value of trust under the condition of no historical interaction.

[0068] Set nodes and nodes They are neighbors, without any historical interaction information ( ),like Need to Conduct a trust assessment, then Will be based on its own type And to type The inference is used to conduct game theory analysis to solve for the maximum expected payoff under the Bayesian Nash equilibrium. This gain is used as the basis for calculating the trust component of the game.

[0069] The Bayesian game between nodes can be represented as , Where N is the set of participating nodes; Depicting nodes The private type space, specifically taking the value ; for The space for action. The actions it actually performs; for Utility function; According to its own type and the type of opponent They choose their actions based on their inferences, therefore strategy It is defined in Mapping on type space : , , Right now : .

[0070] Utility function The value is random, and the node The conditional expected return is:

[0071]

[0072] In the formula, ,express In its own type When the opponent type is Subjective probability judgment. When the node types are independent of each other, there is .

[0073] For the above game, the strategy combination A Bayesian Nash equilibrium is formed if and only if for any node and any type and except Any deviation strategy other than The following inequalities hold:

[0074]

[0075] Therefore, node Adopting a dominant strategy This allows for the attainment of the maximum expected return, which is then used as the basis for trust assessment.

[0076] When the strategy of a node in a game is a combination of multiple action vectors, the game expands into a multidimensional Bayesian game, where the action space of the node is multidimensional. The policy output is an action vector rather than a single action choice, and the global policy combination is also in the form of a multi-dimensional vector. The utility function is determined by a combination of multiple strategies.

[0077] 4.2 Two-Dimensional Energy Distribution Utility Function Modeling

[0078] In the MANET environment, the mathematical form of the utility function not only shapes the incentive structure for node strategy selection but also fundamentally determines the rationality and discriminative power of trust assessment results. Since each node's total energy is finite, besides the energy required to forward data for other nodes, nodes also need to reserve some energy for their own tasks, such as numerical computation, data generation, and maintenance. This invention divides node energy into two categories of private variables: energy reserved for their own tasks and energy allocated for forwarding services. A two-dimensional Bayesian game utility function is constructed to achieve precise quantification of node forwarding costs and collaborative benefits.

[0079] Let the nodes in the energy allocation game be... The policy vector is ,in It is about conserving energy for one's own affairs. It is a forwarding service that allocates energy. , This represents the space of possible energy allocation strategies. Clearly, and Meet the conditions ,in For nodes The total energy, and It changes dynamically over time and with the increase of interactive behaviors.

[0080] Different nodes utilize their own two types of energy at different rates; this is a private characteristic of each node. Let... This refers to the utilization rate of the node's own transaction energy. The utilization rate of energy for node forwarding services, and When a node allocates more forwarding service energy to transmit data for other nodes, it is more likely to receive more forwarding services from other nodes, and its trust level is also higher. In the game of energy allocation, the task of a node is to find an advantageous strategy to maximize its expected utility.

[0081] The constructed utility function takes the following form:

[0082]

[0083] In the formula, and The preference weights for the two types of energy gains satisfy the following conditions: ; and It is a constant, which can be understood as a node. The baseline preference weights for energy allocation are subject to the following constraints: .

[0084] When node The opponent's energy utilization rate is unknown, only its distribution is known. At that time, its expected utility is:

[0085]

[0086] In the formula, For type vectors , The joint probability density function.

[0087] Therefore, the energy allocation problem among nodes constitutes a Bayesian two-dimensional game. First, the type vector of nodes is naturally selected. All nodes simultaneously choose advantageous strategies in terms of both self-reserved energy and forwarding service energy. Finally, each node obtains the final payoff under the corresponding strategy combination.

[0088] 4.3 Calculation of Trust Components in Game Theory

[0089] This invention selects Bayesian game theory based on energy allocation for trust assessment and calculates the trust component of the game. Since the expected payoff function of a node is continuously and smoothly differentiable with respect to the policy variable, the equilibrium policy can be obtained through the first-order partial derivative condition, thereby obtaining the maximum payoff.

[0090] Assuming nodes Participate in a large-scale event at a certain moment In a multi-party game, the maximum overall payoff is denoted as... Furthermore, define the nodes. The marginal contribution is the change in overall revenue after removing that node, i.e.:

[0091]

[0092] in Indicates will After being removed from the game theory, The maximum gain obtained by playing against the other neighbors.

[0093] Therefore, the game-theoretic component of trust assessment can be quantified as follows:

[0094]

[0095] As shown in the above formula, when the marginal contribution is greater than 0, it indicates that the node... Participation helps to improve If the node's return is high, the node's credibility is relatively high; otherwise, it indicates that the node has a negative impact on the return and can be regarded as a low-credibility node, with its game component set to 0.

[0096] The aforementioned multi-player game model assumes the complete connectivity of the neighbor set. If this condition is not met, a fully connected subset of the neighbor set needs to be introduced to prune the game space. Specifically, taking... such that any two belonging to All nodes are neighbors of each other, thus restricting the game to... The process is performed on the complete subgraph. If there is no direct link between any two nodes in the neighbor set, the original multi-participant Bayesian game is decoupled as follows: An independent bilateral game As co-participants, they need to interact strategically with each neighbor one by one. In this case, the game components are calculated as follows:

[0097]

[0098] In the formula, For nodes and The maximum expected payoff when engaging in a bilateral game.

[0099] As a preferred implementation, to address the computational constraints of resource-constrained nodes, three lightweight processing methods can be employed to reduce the computational overhead of game solving. First, a numerical iterative method is used to solve the Nash equilibrium, setting a maximum number of iterations and a convergence accuracy threshold. The calculation terminates when the number of iterations reaches the upper limit or the change in payoff is less than the convergence threshold, avoiding computational waste caused by infinite iteration. Second, a pre-stored equilibrium strategy table for common neighbor sizes is used. At runtime, the baseline strategy is reused by directly looking up the table based on the current number of neighbors and energy range, and then a small adjustment is made to obtain the adapted result, reducing real-time computation. Third, low-computing-power nodes can degenerate into a simplified utility function, retaining the core logic of two-dimensional energy allocation while compressing nonlinear computational terms, trading a slight decrease in accuracy for a significant reduction in computational overhead.

[0100] V. Observing the Trust Component using a Sliding Window

[0101] The observed components are calculated based on direct node interaction records and serve as the core empirical evidence for trust assessment. To fully utilize historical interaction data and mitigate the interference of outdated records, this invention introduces a sliding window mechanism to discretize the time axis. Let the assessment period be... Node every Each time unit performs a trust assessment on its neighbors, defining the observation window duration as... ,in This represents the number of historical interaction rounds included in the statistics within the window.

[0102] The formula for calculating the observed components of the k-th round of trust assessment is:

[0103]

[0104] In the formula, Within the s-th round of evaluation for The number of data packets successfully forwarded; Within the s-th round of evaluation Towards Total number of forwarding requests initiated; This is a penalty coefficient used to impose a trust deduction on negative or inappropriate behavior by nodes.

[0105] When c=k, the window covers all historical interaction records, and the observed components degenerate into a global cumulative form. In real-world network environments, storing and processing all historical information incurs significant computational and storage overhead; therefore, c is typically configured based on specific application scenarios and resource constraints. The sliding window mechanism naturally achieves the decay effect of historical interaction records, with earlier interaction data sliding out of the statistical window first, thereby reducing the interference of old, expired records on the current trust assessment.

[0106] VI. Adaptive Dynamic Weighted Fusion Mechanism

[0107] The weighting factor is used to quantify the node's tendency to weigh game components against observation components, and its reasonable value directly affects the accuracy of the comprehensive trust value. The weighting factor in this invention follows three basic principles: when direct observation data is lacking, trust inference should be primarily based on game theory; as interactive evidence accumulates, the weight of the observation component in trust synthesis should monotonically increase; and the weighting allocation mechanism must take into account the fairness of the evaluation scales among different neighboring nodes.

[0108] The first two principles have intuitive physical meaning, while the third principle aims to eliminate evaluation bias caused by node heterogeneity in the calculation of observed components. Specifically, observed components only reflect the ratio of interaction results and fail to characterize individual differences in the scale of interactions. If two nodes have the same observed component values, but one has only completed one interaction while the other has completed thousands of interactions, the actual trust levels will obviously differ significantly. Therefore, the selection of weighting factors needs to introduce interaction frequency as a moderating variable to correct for such bias.

[0109] set up To observe the area within window c and If the frequency of interaction is given, then the weighting factor for the k-th round of evaluation is defined as:

[0110]

[0111] In the formula, the denominator is not zero. If the denominator is zero, it indicates that the node did not interact with any neighbors during the window period, and this is treated as a boundary case. The quantification of the number of interactions can be achieved using a counter to record the number of interactions within the observation window. Towards The number of data transmission requests sent can also be taken as the total number of forwarded requests or the total number of successful forwards within the window as a proxy variable.

[0112] As shown in the above formula, the interaction frequency is positively correlated with the weighting factor. The higher the interaction frequency, the stronger the coupling between the node and its neighbors, and the more the trust assessment relies on direct observations. In particular, if a node has only one neighbor, the weighting factor is 1. In this case, the trust value of that neighbor is entirely determined by the observed component. This is topologically inevitable because communication between the node and other nodes in the network must be forwarded through this neighbor; otherwise, the node will become topologically isolated. Conversely, when the interaction frequency is 0, the observed component has no effective data, and the trust assessment degenerates into a pure game theory model. This situation also conforms to rational expectations.

[0113] Furthermore, interaction frequency and trust level are bidirectionally coupled in real networks: nodes tend to choose neighbors with higher trust values ​​to establish connections, and frequent interactions will strengthen the trust weight of that neighbor, forming a positive feedback mechanism.

[0114] VII. Anti-attack verification and trust penalty mechanism

[0115] 7.1 Energy false alarm cross-validation mechanism

[0116] The total energy of a node changes dynamically over time and with variations in inter-node interactions. For ease of analysis, we assume energy consumption occurs only in forwarding data packets, and that a node consumes a fixed amount of energy to forward one unit of data packet for its neighbor. After the k-th interaction, the node's remaining energy can be approximated as:

[0117]

[0118] In the formula, Energy consumption per unit of data packet forwarding For nodes Replacement node The number of data packets forwarded.

[0119] Since each node's participation in the game presupposes knowing its own and its opponent's current energy level distribution, if this distribution is unknown, each node must calculate its own or its opponent's remaining energy after each interaction and record it in its local information table. A node may forward data packets to multiple nodes, and the energy consumed by this forwarding cannot be precisely known to any single node. Therefore, during the game, nodes will request their opponents to report their current energy level, which may lead nodes to misreport their energy level in order to gain a greater advantage.

[0120] The impact of such false reporting is limited. Assume a node... The actual energy is It sends to the node Report your energy level as And at this moment, the node Nodes recorded in the information table The energy is Because of the node Knowing that the actual energy will not exceed the recorded value, if the node If a node reports an energy level greater than or equal to the recorded value, it can immediately conclude that the node is lying and refuse further interaction with it. A node can only successfully lie if the reported energy is between the actual energy and the recorded energy.

[0121] In this scenario, nodes can mitigate the negative impact of false reports through two methods. First, they can query neighboring nodes for interaction records related to the target node, performing multi-source cross-verification. Second, they can apply a discount correction to the energy level reported by the node. ,in It is a discount factor.

[0122] As a preferred implementation, energy cross-verification can adopt a dual-trigger mode combining periodic routine verification and verification triggered before forwarding requests. Verification is initiated proactively by the evaluation node, which broadcasts the target node's energy observation request to its one-hop neighbors, using the estimated values ​​from the majority of neighbors as a reference benchmark. When a node is found to be falsely reporting energy, a tiered trust penalty mechanism can be used, with the penalty magnitude positively correlated with the magnitude of the false report. The updated trust value is... The discount factor decreases in stages as the proportion of false reports increases. For nodes that make multiple false reports, the penalty can be cumulatively increased until they are included in the list of low-trust nodes.

[0123] 7.2 Conspiracy Attack Identification Mechanism

[0124] In the node trust assessment framework that integrates energy game theory and interactive observation, collusion attacks refer to multiple malicious nodes forming a coordinated attack group. They bypass the single-point energy false reporting verification mechanism by synchronously manipulating forwarding behavior, colluding to falsify energy reporting data, mutually boosting internal trust, and jointly suppressing normal nodes. Specifically, it can be divided into three attack modes: **Group mutual trust boosting attack:** Colluding nodes forward data packets at high frequency, uniformly falsely report high remaining energy, synchronously increasing each other's game theory and observation components to seize network forwarding resources; **Collaborative node defamation attack:** Group nodes synchronously experience high packet loss against the same target node, uniformly report low energy parameters of the target node, and batch-downgrade the target node's overall trust value, blocking its communication link; **Joint energy false reporting attack:** The group uniformly falsifies its own energy parameters to external neighbors, relying on cross-verification from multiple nodes to circumvent single-node energy consistency verification.

[0125] By summarizing the behavioral patterns of collusive groups, four quantifiable identifiable characteristics can be extracted: the packet loss rate, energy reporting deviation, and trust value fluctuation of multiple nodes targeting the same target node are strongly synchronously correlated; the frequency of bidirectional interactions and forwarding success rate within the group are significantly higher than the interaction indicators between the group and external nodes; the game utility function offsets of group nodes are highly similar within the same evaluation period; and the group synchronously adjusts its forwarding strategy and energy reporting parameters when the topology is updated or neighbors change.

[0126] This invention employs a fully distributed collusion behavior identification mechanism, requiring no central node and synchronizing lightweight statistical summaries only among one-hop neighbors, making it suitable for low-power, resource-constrained nodes in MANET. The mechanism extends collusion behavior statistical fields to the node's local information table, establishing periodic indicator collection and local data synchronization logic. Assuming a sliding observation window length of c, four statistical indicators are synchronously updated and stored in the local node information table during each trust assessment cycle: average packet loss rate (the ratio of the number of neighbor-forwarded request packets dropped by a node within the window to the total number of request packets); energy reporting deviation (the difference between the node's reported remaining energy and the estimated actual remaining energy); single-round trust fluctuation amplitude (the absolute value of the increase or decrease in trust value within a single assessment cycle); and interaction frequency density (the ratio of the total number of bidirectional interactions between nodes within the window to the window duration), used to quantify the closeness of node interactions.

[0127] After completing a single round of trust assessment, a node only synchronizes the behavioral statistics summary of the third-party node to its own one-hop neighbor, following three constraints: it only synchronizes the average of the four indicators of the third-party node and does not transmit the complete original data packet log to reduce communication overhead; it attaches a local timestamp and a unique node identifier to the synchronization message and uses a simple hash check to prevent message tampering; and it pauses summary synchronization when there are no new interactions or topology changes to reduce node energy consumption.

[0128] For any pair of neighboring nodes, a multi-dimensional behavioral correlation coefficient is constructed for the same target node to quantify the degree of synchronization between their behaviors. The calculation formula is as follows:

[0129]

[0130] In the formula, Corr(·) is the Pearson correlation coefficient, which takes values ​​in the range of [-1, 1]. The closer the value is to 1, the stronger the synchronicity of the behavior. , , The dimensional weights are set to 0.5, 0.3, and 0.2 respectively, based on the MANET forwarding behavior characteristics, with packet loss behavior serving as the core basis for determining collusion.

[0131] Collusion determination employs a two-tiered threshold mechanism. The first-tier threshold performs coarse screening of suspected colluding node pairs, while the second-tier threshold identifies colluding groups. An adaptive threshold correction strategy is also designed to adapt to dynamic changes in network node density. Preset basic static determination thresholds include a behavioral correlation coefficient threshold of 0.8, an internal interaction density threshold of 0.7, and an energy deviation synchronization threshold of 0.4. A pair of nodes is marked as a suspected colluding node pair when all three constraints are met: there is at least one common target node such that the behavioral correlation coefficient is not lower than the corresponding threshold; the bidirectional interaction density of the node pair is not lower than the corresponding threshold; and the energy deviation of the node pair towards the same target node has the same positive or negative sign, and the deviation difference is less than the corresponding threshold. As a supplementary determination rule, if multiple nodes simultaneously exert significant trust suppression on the same node, with a single-round trust decay exceeding 0.3, they can be directly marked as a high-risk suspected colluding pair.

[0132] An undirected suspected association graph is constructed based on all suspected colluding node pairs. In the graph, vertices represent network nodes, and undirected edges represent suspected collusion relationships between nodes. If the number of nodes in a connected subgraph of the association graph is not less than 3, then the connected subgraph is determined to be a formal colluding attack group.

[0133] Static fixed thresholds are prone to false positives and false negatives in high-density or sparse topology scenarios. Therefore, a node density correction term is introduced to dynamically adjust the correlation coefficient threshold. The correction formula is as follows:

[0134]

[0135] In the formula, The threshold for the corrected correlation coefficient; This represents the total number of one-hop neighbors of this node; This refers to the total number of nodes deployed across the entire network. This is a density correction coefficient, the value of which can be determined by debugging according to the network scenario, with a typical range of 0 to 1. When the network node density increases, the judgment threshold is slightly lowered, and the threshold is raised in sparse topology scenarios to balance the false negative rate and the false positive rate of normal nodes.

[0136] As a preferred implementation, nodes identified as part of a conspiracy attack group can be subject to targeted trust penalties. For nodes within the group, their overall trust value can be directly lowered to the low trust range. In severe attack scenarios, their trust value can be set to zero and they can be included in the forwarding blacklist with a fixed isolation period. During the isolation period, their forwarding services will not be accepted and they will not be provided with forwarding support.

[0137] VIII. Management of the entire lifecycle of trust relationships

[0138] The evolution of trust relationships depends on the continuous maintenance of node attributes and historical interaction data. Therefore, each node needs to reserve local storage resources and organize the above information in a structured form. The update of trust relationships between nodes is driven by two types of triggering events: one is topology change, that is, the addition or deletion of members in the one-hop neighbor set of a node; the other is behavioral interaction, that is, the generation of new data interaction instances between nodes.

[0139] 8.1 Trust Relationship Initialization

[0140] When a new node joins the network, it needs to perform a trust assessment on its potential neighbors within a one-hop radius. This process is called the initialization of trust relationship establishment. Since the node has not yet maintained any neighbor information and its local information table is empty, it first needs to complete neighbor discovery by broadcasting hello request packets within a one-hop radius, and then calculate the initial trust value of each neighbor based on game theory.

[0141] The initialization process consists of three steps. First, the neighbor set is updated. Nodes broadcast a "hello" request packet to their neighbors and set a time-to-live (TTL) of 1. Nodes receiving the "hello" request packet reply with a "reply" message and add the requesting node to their own neighbor set. Within the default waiting time, the requesting node constructs its own neighbor set based on the received "reply" packets. Second, the initial trust value is calculated. Nodes engage in a game with their neighbors and solve for the optimal payoff, calculating the game trust component. Since there are no historical interaction records at this stage, the observation component is 0, and the weight coefficient is also 0. The overall trust value is entirely determined by the game component. Finally, the trust value, attribute parameters, and interaction records of each neighbor are written into a local information table, completing the initialization process. This derivation implicitly assumes that all nodes in the network use isomorphic utility functions to ensure the comparability and consistency of the game theory results.

[0142] As a preferred implementation, the hello and reply signaling can include core fields such as a unique node identifier, timestamp, reported remaining energy, number of neighbors, and hash checksum, which is sufficient to support neighbor discovery and basic energy information interaction, without the need for additional redundant fields. The node's local information table can include fields such as neighbor node identifier, remaining energy record, dual-dimensional energy utilization rate, current comprehensive trust value, interaction history sliding window queue, collusion risk marker, and last update timestamp, covering all the basic data required for trust assessment.

[0143] 8.2 Multi-Event-Driven Trust Update Mechanism

[0144] Trust updates employ three triggering modes: scheduled batch updates, event-driven updates, and exception response updates. The scheduled mode aggregates recalculations at fixed intervals to suppress computational overhead; the event mode evaluates in real time when topology changes or interactions are completed to ensure state synchronization; and the exception mode is triggered when malicious behaviors such as packet loss rate jumps, energy false reporting, or collusion attacks are detected, and a punitive decay is applied.

[0145] Topology changes correspond to scenarios involving the addition or deletion of members in the neighbor set. When a node receives a hello request packet from a new node, if that node is not in its own neighbor set, it is added to the neighbor set, a trust cold start is performed, the initial trust value is calculated, and the information table is updated. For neighbor failure detection, since nodes cannot passively perceive the migration of other nodes, they need to actively probe by periodically broadcasting hello packets; unresponsive nodes are removed from the neighbor set. When the neighbor set changes, the pruning and reconstruction of the game computation space is automatically triggered: adding a neighbor expands the game participant set and resolves the multi-player game equilibrium; removing a neighbor shrinks the game participant set and adjusts the game scale accordingly. If the change in the neighbor set topology causes the fully connected condition to no longer be satisfied, it is automatically adapted according to the complete subgraph pruning or bilateral decoupling rules to ensure the effectiveness of the game computation.

[0146] This interactive update mechanism addresses scenarios where data forwarding interactions are completed between nodes. After a node completes a data forwarding interaction with its neighbor, it continuously calculates the neighbor's actual forwarding volume and total requests. Based on this, it performs trust recalculation and synchronously updates the information table records. The operation process is consistent with the trust calculation steps during initialization. This real-time update mechanism ensures that the trust state is dynamically synchronized with the network topology and node behavior, maintaining the timeliness of trust relationships between neighbors. In high-density, high-mobility scenarios, frequent evaluation calculations and signaling interactions will significantly increase node resource overhead and may delay the trust convergence process. However, because the update operation is constrained within a single-hop local neighborhood, its convergence performance is still superior to the global update strategy.

[0147] As a preferred implementation, to further reduce system overhead, an on-demand update strategy can be adopted. Trust assessment is triggered only when a node initiates a data forwarding request, thus generating an interaction need. In this mode, the node first performs a neighbor table consistency check, and then performs a trust assessment of the target neighbor. The assessment result directly guides the forwarding decision. After the interaction is completed, the observed data is written to the information table as a historical sample, providing prior support for subsequent assessments.

[0148] IX. Simulation Experiments and Result Analysis

[0149] 9.1 Experimental Setup

[0150] Simulation environment: The MANET simulation platform is built based on NS-3, with a network size of t nodes, random topology, node movement speed of 0~10m / s, simulating a high dynamic scenario; the initial energy of each node is 5~10 units, the energy consumption of forwarding a single data packet is 0.005 units, and dynamic energy consumption is supported.

[0151] The simulation parameters are set as follows: the energy consumption of a node forwarding a single data packet is 0.005 energy units; in each evaluation period... Inside, source node The system broadcasts 40 data packets to its neighbors. The actual number of packets received by each neighbor is proportionally allocated based on their current trust level; that is, neighbors with higher trust levels receive more packets. The more data forwarding tasks are acquired, the higher the proportion will be. , , The game utility function adopts the form constructed above. However, in the experiment, to reduce the computational complexity, it is assumed that the energy utilization rate of all nodes is public knowledge, i.e., it degenerates into a game of complete information. This simplification does not change the essential characteristics of the trust evolution law, and the experimental results can still effectively reflect the overall performance trend of the proposed evaluation mechanism.

[0152] Consider nodes After joining the network, it will interact with its neighboring nodes. , The trust evolution process to assess rounds (Corresponding time window is) The horizontal axis represents the level of trust. , The dynamic trajectory of the change. Experimental parameter configuration is shown in Table 1, and the penalty coefficient is taken as... .

[0153] Table 1 Basic Environmental Parameter Settings

[0154]

[0155] First, we analyze the game components, nodes. , , This constitutes a three-player game, and the structure The Lagrange function is:

[0156]

[0157] in, These are Lagrange multipliers. The optimal conditions are solved using the first derivative method, i.e.: The optimal solution can be obtained as follows:

[0158]

[0159] in,

[0160]

[0161]

[0162]

[0163] By substituting the relevant parameters, the nodes in the three-player game can be calculated. Maximum gain Similarly, we can calculate the exclusion of nodes separately. and Time node The maximum gain from the game and Through multiple rounds of iterative calculations, the relationship curve between trust level and behavior changes with the number of evaluation rounds can be obtained.

[0164] 9.2 Analysis of Experimental Results

[0165] Figure 3 This study reveals the dynamic game-theoretic relationship between the observed and game-theoretic components in the evolution of trust. The dominance of direct observation is reflected in the fact that even... During the cold start phase, it gains a higher game value due to its energy reserve advantage. > ), It still relies on its low packet loss rate to quickly accumulate trust capital in actual interactions, achieving right However, this trust advantage based on behavioral observation is conditional; that is, when the frequency of interaction is too high, it surpasses expectations. The energy consumption undermines the basis of its game-theoretic valuation, causing a decline in the trust curve; and Because energy loss is controllable, its game-theoretic weight is relatively strengthened, and the level of trust is restored, such as Figure 4 As shown above, the dynamics demonstrate that the weighting ratio between static attributes and dynamic behaviors adaptively adjusts with the network's operating state, rather than remaining fixed.

[0166] In the experiment, nodes were set. exist The observation window period reduced the packet loss rate from 80% to 10%. Figure 5 It can be seen that, It rose rapidly, and at the same time It decreases over time because the nodes for More data was transmitted, thus gaining more trust. Simultaneously, it is possible to... Figure 6 As seen in the image, the node... The energy game advantage decreases as forwarding activities increase.

[0167] Figures 7 to 10 This reflects the impact of node energy utilization on trust level. As the graphs show, this impact is only significant in the initial stages of node interaction and almost disappears as interaction activity increases. This is because in this evaluation model, direct interaction plays a decisive role in trust level, which can be reflected through the weighting parameters. These graphs also show that trust level increases slightly with increasing energy utilization, consistent with objective understanding.

[0168] In a MANET dynamic topology environment, a trust recalculation is triggered when the membership of a node's neighbor set changes. When a new node joins the neighbor set, the node needs to re-evaluate the trust of all its neighbors. As can be seen from the weighted fusion structure, under the condition that node attributes remain constant, the trust valuation of existing neighbors will exhibit a systematic downward bias. This phenomenon can be explained from the perspective of risk diversification: the expansion of the neighbor candidate set provides nodes with more forwarding path options, thus reducing the risk exposure caused by the failure of a single neighbor. Therefore, the node's dependence on a specific neighbor weakens, and the trust valuation is adjusted downward accordingly. Conversely, when a node is removed from the neighbor set, the trust valuation of the remaining neighbors shows a slight upward bias. This trend stems from the enhanced coupling relationship brought about by topology sparsity: the shrinking of the number of neighbors increases the topological dependence between the node and the remaining nodes, strengthening the structural position of the remaining neighbors in the communication path, thereby boosting their trust valuation.

[0169] In summary, this invention addresses the node trust assessment requirements in MANET and edge cloud convergence scenarios by constructing a dual-component fusion assessment framework, effectively solving the challenge of trust assessment for new nodes during cold starts. Existing technologies often use fixed, neutral initial trust values ​​for new nodes with no interaction history, making them vulnerable to malicious attacks. This invention, however, relies entirely on energy-based Bayesian game components to output initial trust values ​​during the cold start phase. Based on node-specific energy attributes and Bayesian Nash equilibrium inference, it obtains differentiated and objective assessment results without requiring manually preset fixed intermediate values, thus eliminating the security vulnerability of unfamiliar malicious nodes infiltrating the network using neutral initial values. As interaction data accumulates, the scheme gradually increases the weight of the observation components through a dynamic weighting mechanism, achieving a smooth transition from cold start to a stable interaction phase.

[0170] The proposed solution employs a dual-dimensional evaluation approach, considering both static resource attributes and dynamic real-time behavior, significantly improving evaluation accuracy. Compared to solutions relying solely on forwarding observations, this invention introduces the dimension of node energy's private attributes, enabling the prediction of potential self-interested tendencies in nodes. Compared to pure game theory models, this invention integrates real-world interaction observation data, allowing for timely capture of later behavioral anomalies in nodes. The solution distinguishes between self-retained energy and forwarding service energy to construct a two-dimensional Bayesian game utility function, accurately quantifying node forwarding costs and long-term collaborative benefits. Simultaneously, it utilizes a sliding window with a penalty coefficient to statistically analyze interaction behavior, mitigating the interference of long-standing invalid histories and capturing dynamic behavioral changes such as packet loss and passive forwarding in real time. After adaptive weighted fusion of these two dimensions, the trust value more closely reflects the true trustworthiness level of nodes, effectively improving the accuracy of distinguishing between normal, selfish, and malicious nodes.

[0171] The adaptive dynamic weighting mechanism can adapt to the entire process of network interaction evolution and has stronger dynamic adaptability. Unlike existing fusion solutions that use manually set fixed weights, this invention dynamically adjusts the weight ratio of the two types of components with the interaction frequency as the core variable. When there is a cold start with zero interaction, the weight of the game component is 1. As the number of interactions continues to increase, the weight of the observation component increases monotonically. In the single-neighbor topology scenario, it automatically and completely converges to the observation component. The evaluation logic throughout the process conforms to the law of node interaction evolution and can adapt to the MANET network environment with high mobility and frequent topology changes.

[0172] The multi-event-triggered trust update mechanism ensures the timeliness of trust values ​​and adapts to highly dynamic network topologies. The solution supports four update modes: topology change triggering, real-time packet interaction triggering, periodic batch updates, and emergency penalties for malicious or abnormal behavior. Simultaneously, it automatically reduces the computational space for multi-player game theory when neighbor nodes are added or removed, decoupling isolated bilateral games. This adapts to dynamic networking environments with high-speed node movement and frequent neighbor online / offline cycles, ensuring that trust values ​​are synchronized with the latest node status and network topology in real time, providing an accurate basis for routing and forwarding decisions.

[0173] The built-in dual defense mechanism of energy false reporting verification and collusion identification significantly improves network robustness and anti-attack capabilities. Energy cross-verification cross-checks the energy reported by nodes by cross-checking multi-neighbor interaction records, imposing a trust discount penalty on false reporting nodes, effectively suppressing false energy reporting behavior. Collusion identification, based on multi-neighbor joint behavior statistics, accurately identifies collusion attacks involving multi-node coordinated packet loss and forged forwarding records through multi-dimensional correlation calculations and hierarchical threshold determination. This dual defense mechanism effectively resists two typical network attacks: energy spoofing and node collusion, reducing the damage to data transmission caused by malicious behavior and improving the overall security of cloud services and the MANET converged network.

[0174] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0175] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A node security trust assessment method applicable to vehicle-mounted and sensor edge cloud MANETs, ​​characterized in that, Includes the following steps: Obtain the one-hop neighbor topology information of the evaluation node, the energy attribute information of each neighbor node, and historical interaction records; For each neighboring node, based on the energy attribute information, the node's energy is divided into two dimensions: its own transaction retention energy and forwarding service allocation energy. A Bayesian game utility function in a two-dimensional policy space is constructed, and the Bayesian Nash equilibrium is solved to obtain the node's maximum expected payoff. Based on the maximum expected payoff, the game trust component of the neighboring node is calculated. Based on the historical interaction records, the packet forwarding behavior of neighboring nodes is statistically analyzed, and the observed trust component of the neighboring node is calculated. The weighting factor is dynamically determined based on the interaction frequency between the evaluation node and its neighboring nodes. The weighting factor corresponds to the weight of the observation trust component, and the value of the weighting factor increases with the increase of the interaction frequency. The game-theoretic trust component and the observed trust component are weighted and fused using the weighting factor to obtain the comprehensive trust level of the neighbor node. The overall trust level is updated in response to changes in neighbor topology or the addition of new interaction records.

2. The node security trust assessment method applicable to vehicle-mounted and sensor edge cloud MANET according to claim 1, characterized in that: The game trust component of the neighbor node is calculated based on the maximum expected return as follows: Calculate the maximum expected payoff of the evaluation node when it participates in the game with all its neighboring nodes, and the maximum expected payoff after removing the target neighboring node. Use the difference between the two as the marginal contribution of the target neighboring node. When the marginal contribution is greater than 0, the game trust component is the ratio of the marginal contribution to the maximum expected return; When the marginal contribution is less than or equal to 0, the game trust component is set to 0.

3. The node security trust assessment method applicable to vehicle-mounted and sensor edge cloud MANET according to claim 1, characterized in that: The method also includes a game space trimming and adaptation step: When the neighbor set corresponding to the one-hop neighbor topology information does not meet the fully connected condition, the game computation space is trimmed and adapted. A fully connected subgraph in the neighbor set is selected as the game participant set to perform multi-player game solving; if no two nodes in the neighbor set have a direct link, the multi-player game is decoupled into multiple independent bilateral games, and the game trust component is calculated based on the maximum expected payoff of each bilateral game.

4. The node security trust assessment method applicable to vehicle-mounted and sensor edge cloud MANET according to claim 1, characterized in that: The process of statistically analyzing the packet forwarding behavior of neighboring nodes and calculating the observed trust component based on the historical interaction records specifically involves: The historical interaction records are statistically analyzed based on a sliding time window. The observed trust component is the ratio of the number of successfully forwarded data packets to the weighted total. The weighted total is the sum of the number of successfully forwarded data packets and the penalty term. The penalty term is the product of a preset penalty coefficient and the number of packet losses. The number of packet losses is the difference between the total number of forwarding requests and the number of successfully forwarded data packets.

5. The node security trust assessment method applicable to vehicle-mounted and sensor edge cloud MANET according to claim 1, characterized in that: The step of dynamically determining the weighting factor based on the interaction frequency between the evaluation node and its neighboring nodes is as follows: The weighting factor is determined by the proportion of the interaction frequency of the target neighbor node within the sliding time window to the total interaction frequency of all neighbor nodes of the evaluation node. When the evaluation node has no historical interaction records within the sliding time window, the weighting factor is 0, and the overall trust level is completely determined by the game trust component. When an evaluation node has only one neighbor node, the weighting factor is 1, and the overall trust level is determined entirely by the observed trust component.

6. The node security trust assessment method applicable to vehicle-mounted and sensor edge cloud MANET according to claim 1, characterized in that: The update of the overall trust level in response to changes in neighbor topology specifically includes: When a neighbor topology change is triggered, the game computation space is adjusted synchronously; when a new neighbor node is added, the set of game participants is expanded and the equilibrium is re-solved; when a neighbor node leaves, the set of game participants is reduced and the maximum expected payoff is recalculated. When a topology change leads to a change in the fully connected condition, the corresponding game space pruning or decoupling operation is performed.

7. The node security trust assessment method applicable to vehicle-mounted and sensor edge cloud MANET according to claim 1, characterized in that: The method also includes an energy false alarm verification step: The remaining energy value reported by the neighboring node is compared with the remaining energy value calculated locally. If the reported remaining energy value is greater than or equal to the recorded value, it is determined to be an energy false reporting behavior. After cross-verification with multiple neighbors, a trust discount penalty is applied to nodes that have engaged in energy misreporting, and the overall trust level of the corresponding node is updated by multiplying it by a preset discount factor.

8. The node security trust assessment method applicable to vehicle-mounted and sensor edge cloud MANET according to claim 1, characterized in that: The method also includes a conspiracy attack identification step: Within a preset sliding time window, the packet loss rate, energy reporting deviation, trust fluctuation amplitude, and interaction frequency density of each neighbor node are statistically analyzed. The average statistical indicators of third-party nodes are synchronized among one-hop neighbors, and the multi-dimensional behavior correlation coefficient of any two nodes toward the same target node is calculated. Collusive attack nodes are identified through a two-level threshold hierarchical determination, wherein the behavior correlation coefficient threshold in the two-level threshold is adaptively adjusted based on the ratio of the total number of local one-hop neighbors to the total number of nodes in the entire network. Apply trust penalties to nodes identified as complicit in the attack.

9. The node security trust assessment method applicable to vehicle-mounted and sensor edge cloud MANET according to claim 1, characterized in that: The solution to the Bayesian Nash equilibrium is obtained by maximizing the expected return of the nodes, and at least one lightweight processing method is used: Numerical iterative methods with maximum number of iterations and convergence accuracy thresholds; The lookup table method for pre-stored benchmark balancing strategies; A method for reducing the complexity of utility function calculations by compressing nonlinear computation terms is adopted.

10. A node security trust assessment device suitable for vehicle-mounted and sensor edge cloud MANET, characterized in that, include: The information acquisition module is used to acquire the one-hop neighbor topology information of the evaluation node, the energy attribute information of each neighbor node, and historical interaction records. The game calculation module is used to divide the energy of each neighbor node into two dimensions, namely, the energy reserved for its own transactions and the energy allocated for forwarding services, based on the energy attribute information. It constructs a Bayesian game utility function in a two-dimensional policy space, solves the Bayesian Nash equilibrium to obtain the maximum expected payoff of the node, and calculates the game trust component of the neighbor node based on the maximum expected payoff. The observation and calculation module is used to statistically analyze the packet forwarding behavior of neighboring nodes based on the historical interaction records, and calculate the observation trust component of the neighboring node. The weighted fusion module is used to dynamically determine a weighting factor based on the interaction frequency between the evaluation node and the neighbor node. The weighting factor corresponds to the weight of the observed trust component, and the value of the weighting factor increases with the increase of the interaction frequency. The game trust component and the observed trust component are weighted and fused through the weighting factor to obtain the comprehensive trust level of the neighbor node. The update execution module is used to update the overall trust level in response to changes in neighbor topology or the addition of new interaction records.

Images