A trust evaluation method and system for a mobile edge computing terminal node

Abstract

This invention discloses a trust assessment method and system for mobile edge computing terminal nodes, relating to the fields of edge computing, network security, and trust management. The assessment method includes the following steps: the terminal node generates local data and trains a local model according to model update instructions issued by the central server and edge nodes, and uploads the local update parameters to the edge nodes; the edge nodes collect behavioral information of the terminal nodes and interact with other edge nodes to share recommendation information for the terminal nodes; the current trust and historical trust of the computing node are calculated, and direct trust is obtained by integrating the current trust and historical trust; the recommended trust of the node is calculated based on behavioral factors such as transmission latency, interaction time difference, data integrity, node reliability, communication success rate, and number of interactions, and malicious recommendations are filtered based on indicators such as recommendation timeliness, recommendation distance, and external trust of recommendations; the comprehensive trust of the computing terminal is calculated by combining the final recommended trust and direct trust.

Description

Technical Field

[0001] This invention relates to the fields of edge computing, network security and trust management, and in particular to a trust assessment method and system for mobile edge computing terminal nodes. Background Technology

[0002] With the rapid development of the Internet of Things (IoT) and next-generation communication technologies, smartphones and other mobile terminals, along with other computationally intensive emerging devices, generate massive amounts of data daily, increasingly straining mobile and wireless networks. Mobile edge computing (MEC), with its advantages of low latency and rapid response, offers an effective solution to this high-load problem. MEC eliminates the need to offload large amounts of tasks to cloud centers, processing them directly at the network edge, closer to the user. While MEC alleviates the significant conflict between business demands and resource scarcity, its distributed deployment and complex service models present privacy challenges. Federated learning, by processing and analyzing data locally before uploading relevant parameters, effectively mitigates privacy concerns by eliminating the need to upload raw data.

[0003] In an MEC network environment, distributed mobile terminals collaborate with MEC edge nodes and a central server to perform a federated learning process. Terminal nodes train their local models using local datasets and parameters from the global model provided by the central server. Then, the edge nodes aggregate the collected local model parameters and communicate with the central server to perform global model aggregation. However, federated learning in MEC still presents several security challenges. The lack of trust between federated terminal nodes makes them vulnerable to external attacks or unreliable behavior due to limited resources. For example, attackers might intentionally upload malicious model parameters or tamper with data, leading to errors in the aggregated model. Therefore, it is necessary to assess the trustworthiness of the terminal nodes that generate the data.

[0004] Existing technical solution 1: J. Guo et al., "TFL-DT: A trust evaluation scheme for federated learning in digital twin for mobile networks," IEEE J. Sel. Area.Comm., vol. 41, no. 11, pp. 3548-3560, Aug. 2023, doi: 10.1109 / JSAC.2023.3310094. This solution designs a behavioral model based on node feature attributes and proposes a method called TFL-DT, which uses node behavioral data to calculate local trust and recommendation trust, and then obtains global trust. This method can accurately evaluate the trustworthiness of nodes under different behavioral patterns and performs better in resisting attacks from alternating honest and dishonest participants. Although the trust value evaluated by the TFL-DT scheme increases with the number of interaction rounds, it differs significantly from the actual value because it ignores the influence of malicious interactions and malicious recommendations.

[0005] Existing technical solution 2: Y. Cao, D. Liu, S. Zhang, T. Wu, F. Xue, and H. Tang, "T-FedHA: A trusted hierarchical asynchronous federated learning framework for the internet of things," Expert Syst. Appl., vol. 245, no. 123006, Jul. 2024, doi: 10.1016 / j.eswa.2023.123006, proposed a trusted hierarchical asynchronous federated learning framework, T-FedHA. This framework integrates distributed trust management, with edge servers calculating and maintaining the trust values ​​of devices. It employs two trust calculation methods based on device interaction, effectively eliminating malicious devices and improving the trustworthiness and security of the federated training process. Although T-FedHA uses historical trust data within a dynamic time window to calculate local trust, enabling timely detection of malicious node behavior, like TFL-DT, its performance evaluation is unstable and fluctuates significantly.

[0006] Represented by the two aforementioned technical solutions, most existing solutions focus on whether data has been tampered with, lacking detection of faulty terminals, and the collected trust factors are relatively limited, failing to provide a comprehensive assessment of nodes. Furthermore, relying on reputation or recommendation-based trust methods cannot guarantee the reliability of recommendations. Therefore, there is an urgent need for a trust assessment method and system for mobile edge computing terminal nodes that can address the aforementioned technical problems. Summary of the Invention

[0007] This solution addresses the problems and needs raised above by proposing a trust assessment method and system for mobile edge computing terminal nodes. It achieves the aforementioned technical objectives and brings about several other technical benefits due to the adoption of the following technical features.

[0008] One object of the present invention is to provide a trust evaluation method for mobile edge computing terminal nodes, comprising the following steps:

[0009] S10: The terminal node generates local data and trains the local model according to the model update instructions issued by the central server and edge nodes, and uploads the local update parameters to the edge nodes.

[0010] S20: Edge nodes collect behavioral information from terminal nodes and interact with other edge nodes to share recommendation information for terminal nodes;

[0011] S30: Calculate the current trust and historical trust of a node based on behavioral factors such as transmission delay, interaction time difference, data integrity, node reliability, communication success rate, and number of interactions, and obtain direct trust by integrating the current trust and historical trust;

[0012] S40: Calculate the node's recommendation trust based on behavioral factors such as transmission delay, interaction time difference, data integrity, node reliability, communication success rate, and number of interactions, and filter malicious recommendations based on indicators such as recommendation timeliness, recommendation distance, and external trust in recommendations.

[0013] S50: Integrated trust that combines final recommended trust and direct trust in computing terminals.

[0014] In addition, the trust assessment method for mobile edge computing terminal nodes according to the present invention may also have the following technical features:

[0015] In one example of the invention, in step S30, the current trust... The calculation method is as follows:

[0016]

[0017] In the formula, For the first One terminal node, For the first One edge node, For the first time Second interaction For terminal nodes No. Reliability measurement during the first interaction The success rate is expressed as:

[0018]

[0019] In the formula, For terminal nodes With edge nodes The number of successful interactions; For terminal nodes With edge nodes The total number of interactions.

[0020] In one example of the present invention, the terminal node No. Reliability measurement during the first interaction Its expression is as follows:

[0021]

[0022] In the formula, For the first Abnormal factors during the second interaction For the first The latency factor during the first interaction; For the first The interval between the time of the next interaction and the current time;

[0023] Among them, the Abnormal factors during the second interaction The expression for data integrity calculation is as follows:

[0024]

[0025] In the formula, It is the number of times the data is complete. For terminal nodes With edge nodes The number of successful interactions.

[0026] In one example of the present invention, in step S30, historical trust is defined as... Weighted average of behavioral reliability and communication success rate The calculation method is as follows:

[0027]

[0028] in, For terminal nodes No. Reliability measurement during the first interaction; Success rate; For terminal nodes With edge nodes The length of the context generated during the interaction; For the current moment, For the computing terminal in the first The moment of reliability of the next interaction.

[0029] In one example of the invention, in step S30, direct trust is established. The calculation expression is as follows:

[0030]

[0031] In the formula, For terminal nodes With edge nodes Current trust during interaction; Trust in history; Let this be the weight of the current trust level; Let be the weight of historical trust; where the expressions for current trust weight and historical trust weight are:

[0032]

[0033]

[0034] In the formula, For terminal nodes With edge nodes The length of the context generated during the interaction; For terminal nodes No. Reliability measurement during the first interaction; This represents the number of interactions.

[0035] In one example of the present invention, step S40, filtering malicious recommendations based on indicators of recommendation timeliness, recommendation distance, and external trust in the recommendations, specifically includes the following steps:

[0036] S41: After receiving the recommendations from the edge nodes, use the recommendation distance to check the similarity with the sender;

[0037] S42: Verify the timeliness of the recommendations to assess their temporal similarity;

[0038] S43: Based on the recommendation distance and recommendation timeliness, perform preliminary classification of the received recommendations to detect recommendation attacks; among them, those with a risk level of less than or equal to low are deleted;

[0039] S44: Calculate the external trust in the recommendations;

[0040] S45: Based on the preliminary classification results and external trust recommendations, detect and filter false recommendation attacks to obtain the final honest recommendations.

[0041] In one example of the present invention, step S50 specifically includes the following steps:

[0042] First, when the terminal interacts with the edge node, the edge node records and updates the terminal's behavior records in real time to form a context, and calculates the terminal's reliability;

[0043] Then, the edge nodes evaluate the direct trust of the terminal based on the latest reliability and the behavior at the current moment; at the same time, the current edge node continuously receives the recommended trust for the current terminal sent by other edge nodes, filters it through the constructed recommendation filtering model, and calculates the weighted average trust of all recommendations to obtain the final recommended trust value.

[0044] Finally, a comprehensive trust system is established by combining the final recommended trust and the direct trust of the computing terminal.

[0045] In one example of the present invention, in step S50, the expression for the comprehensive trust is:

[0046]

[0047] In the formula, and These are the weights for direct trust and indirect trust, respectively; where they are both in... Inside, satisfy ; For direct trust; To recommend and trust.

[0048] Another objective of this invention is to provide a trust evaluation system for mobile edge computing terminal nodes, comprising:

[0049] The parameter update module is configured to generate local data at the terminal node, train the local model according to the model update instructions issued by the central server and edge nodes, and upload the local update parameters to the edge node.

[0050] The information sharing module is configured to allow edge nodes to collect behavioral information from terminal nodes and interact with other edge nodes to share recommendation information for terminal nodes.

[0051] The direct trust calculation module is configured to calculate the current trust and historical trust of a node based on behavioral factors such as transmission latency, interaction time difference, data integrity, node reliability, communication success rate, and number of interactions, and to obtain direct trust by integrating the current trust and historical trust.

[0052] The recommendation trust calculation module is configured to calculate the recommendation trust of nodes based on behavioral factors such as transmission latency, interaction time difference, data integrity, node reliability, communication success rate and number of interactions, and to filter malicious recommendations based on indicators such as recommendation timeliness, recommendation distance and recommendation external trust.

[0053] The integrated trust calculation module is configured to combine the final recommended trust and the direct trust calculation terminal to calculate the integrated trust.

[0054] In one example of the present invention, the recommendation trust calculation module includes:

[0055] The similarity checking unit is configured to check the similarity of the sender using the recommendation distance after receiving the recommendation from the edge node.

[0056] A similarity assessment unit is configured to verify the timeliness of recommendations sent by nodes in order to assess their temporal similarity.

[0057] The recommendation classification unit is configured to perform preliminary classification of received recommendations based on recommendation distance and recommendation timeliness in order to detect recommendation attacks; wherein, recommendations with a risk level of less than or equal to the lowest level are deleted.

[0058] External trust unit, configured to compute external trust for recommendations;

[0059] The honest recommendation unit is configured to detect and filter false recommendation attacks based on the preliminary classification results and external trust in the recommendations, and obtain the final honest recommendations.

[0060] The present invention has the following advantages over the prior art:

[0061] This technical solution not only ensures the comprehensiveness and accuracy of trust assessment and increases assessment efficiency, but also solves the problem of unreliable trust evidence caused by malicious recommendations. It also avoids the inflexibility of using experience thresholds. It enables secure and reliable distributed collaborative training in mobile edge computing, alleviates the problem of privacy information leakage, can quickly identify malicious nodes, accurately assess node trust, and effectively filter malicious recommendations, thus improving detection and assessment performance and robustness.

[0062] The preferred embodiments of the invention will be described in more detail below with reference to the accompanying drawings, so as to facilitate an understanding of the features and advantages of the invention. Attached Figure Description

[0063] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. The drawings are merely illustrative of some embodiments of the present invention and are not intended to limit the scope of the present invention to all embodiments.

[0064] Figure 1 This is a flowchart illustrating the trust assessment method for mobile edge computing terminal nodes according to an embodiment of the present invention.

[0065] Figure 2 This is a schematic diagram of the recommended trust filtering model process according to an embodiment of the present invention;

[0066] Figure 3 This is the membership function of the fuzzy inference module 1 according to an embodiment of the present invention;

[0067] Figure 4 This is the fuzzy rule confusion matrix of the fuzzy inference module 1 according to an embodiment of the present invention;

[0068] Figure 5 This refers to the membership function of the fuzzy inference module 2 according to an embodiment of the present invention;

[0069] Figure 6 This is the fuzzy rule confusion matrix of the fuzzy inference module 2 according to an embodiment of the present invention;

[0070] Figure 7 The results of the proposed solution of the present invention before and after filtering malicious recommendations are shown in the embodiments of the present invention. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The same reference numerals in the drawings represent the same components. It should be noted that the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0072] Unless otherwise defined, the technical or scientific terms used herein shall 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 patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, “an” or “a” and similar terms do not necessarily indicate a quantity limitation. Terms such as “comprising” or “including” mean that the element or object preceding the word encompasses the element or object listed following the word and its 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; these relative positional relationships may change accordingly when the absolute position of the described object changes.

[0073] A trust assessment method for a mobile edge computing terminal node according to a first aspect of the present invention includes the following steps:

[0074] S10: The terminal node generates local data and trains the local model according to the model update instructions issued by the central server and edge nodes, and uploads the local update parameters to the edge nodes.

[0075] S20: Edge nodes collect behavioral information from terminal nodes and interact with other edge nodes to share recommendation information for terminal nodes;

[0076] S30: Calculate the current trust and historical trust of a node based on behavioral factors such as transmission delay, interaction time difference, data integrity, node reliability, communication success rate, and number of interactions, and obtain direct trust by integrating the current trust and historical trust;

[0077] Direct trust is calculated based on factors such as the node's current behavior, data integrity, historical records, and communication logs. This multi-factor approach increases the comprehensiveness of trust assessment. During the transmission of local updates between the terminal and edge nodes, the number of interactions, interaction time, and data integrity between the terminal and edge nodes are recorded in real time to calculate the terminal's current trust. However, the terminal's current trust only indicates its credibility at the current moment and does not reflect the trustworthiness of its behavior in previous moments or even earlier. Therefore, during the transmission model, the edge node updates the terminal's context based on communication conditions, forming a historical record to calculate the terminal's historical trust. Direct trust is obtained by integrating current trust and historical trust.

[0078] S40: Calculate the node's recommendation trust based on behavioral factors such as transmission delay, interaction time difference, data integrity, node reliability, communication success rate, and number of interactions, and filter malicious recommendations based on indicators such as recommendation timeliness, recommendation distance, and external trust in recommendations.

[0079] S50: Integrated trust that combines final recommended trust and direct trust in computing terminals.

[0080] This evaluation method not only ensures the comprehensiveness and accuracy of trust assessment and increases evaluation efficiency, but also solves the problem of unreliable trust evidence caused by malicious recommendations. It also avoids the inflexibility of using empirical thresholds. It enables secure and reliable distributed collaborative training in mobile edge computing, alleviates the problem of privacy information leakage, can quickly identify malicious nodes, accurately assess node trust, and effectively filter malicious recommendations. It has improved detection and evaluation performance as well as robustness.

[0081] In one example of the invention, in step S30, the current trust... The calculation method is as follows:

[0082]

[0083] In the formula, For the first One terminal node, For the first One edge node, For the first time Second interaction For terminal nodes No. Reliability measurement during the first interaction The success rate is expressed as:

[0084]

[0085] In the formula, For terminal nodes With edge nodes The number of successful interactions; For terminal nodes With edge nodes The total number of interactions.

[0086] In one example of the present invention, the terminal node No. Reliability measurement during the first interaction The expression for calculating the transmission delay and interaction time difference is as follows:

[0087]

[0088] In the formula, For the first Abnormal factors during the second interaction For the first The latency factor during the first interaction; For the first The interval between the time of the next interaction and the current time; and The earlier the interaction occurs, the lower the proportion of anomaly factors and time delay factors in the calculation. This is because since interaction can occur again or multiple times... Interaction, Explanation Increasingly recognized terminal . The introduction of this property can digitize it. This is obtained by statistically analyzing the time difference between uploading and receiving local updates.

[0089] Among them, the Abnormal factors during the second interaction The expression for data integrity calculation is as follows:

[0090]

[0091] In the formula, It is the number of times the data is complete. For terminal nodes With edge nodes The number of successful interactions.

[0092] In one example of the present invention, in step S30, both reliability and communication success rate are obtained based on the terminal's historical interaction records. Thus, historical trust is defined as... Weighted average of behavioral reliability and communication success rate The calculation method is as follows:

[0093]

[0094] in, For terminal nodes No. Reliability measurement during the first interaction; Success rate; For terminal nodes With edge nodes The length of the context generated during the interaction; For the current moment, For the computing terminal in the first The moment of reliability of the next interaction.

[0095] In one example of the invention, in step S30, direct trust is established. The calculation expression is as follows:

[0096]

[0097] In the formula, For terminal nodes With edge nodes Current trust during interaction; Trust in history; Let this be the weight of the current trust level; Let be the weight of historical trust; where the expressions for current trust weight and historical trust weight are:

[0098]

[0099]

[0100] In the formula, For terminal nodes With edge nodes The length of the context generated during the interaction; For terminal nodes No. Reliability measurement during the first interaction; This represents the number of interactions.

[0101] In one example of the present invention, step S40, filtering malicious recommendations based on indicators of recommendation timeliness, recommendation distance, and external trust in the recommendations, specifically includes the following steps:

[0102] S41: After receiving the recommendations from the edge nodes, use the recommendation distance to check the similarity with the sender;

[0103] S42: Verify the timeliness of the recommendations to assess their temporal similarity;

[0104] S43: Based on the recommendation distance and recommendation timeliness, perform preliminary classification of the received recommendations to detect recommendation attacks; among them, those with a risk level of less than or equal to low are deleted;

[0105] S44: Calculate the external trust in the recommendations;

[0106] S45: Based on the preliminary classification results and external trust recommendations, detect and filter false recommendation attacks to obtain the final honest recommendations.

[0107] When there is insufficient interaction between edge nodes and terminal nodes, it is necessary to communicate with other edge nodes that interact with the terminal nodes to obtain recommended trust. It is important to note that the recommended trust value is the combined trust generated by other edge nodes during their interactions with the terminal node. If the historical interaction records are zero, then it is considered direct trust.

[0108] Because of attacks, a node's identity, reputation, and other information can be altered. Therefore, recommendations generated by edge nodes are not always truthful and reliable. Although the forms of attacks vary, the characteristics that affect recommendation values ​​are consistent. To address this, we combined attack characteristics to statistically analyze recommendation distance and recommendation timeliness, and calculated external trust in the recommendation value to measure the honesty of the recommendations.

[0109] Since none of the obtained recommendation metrics can independently determine the honesty of the recommendations, the recommendation results are uncertain. Therefore, a recommendation trust filtering model is constructed by combining TS fuzzy logic, as shown in the appendix. Figure 2 As shown, the decision-making process includes two fuzzy inference modules, which classify different trust values ​​through predefined fuzzy rules, and ultimately detect and filter out malicious recommendations.

[0110] In one example of the present invention, step S50 specifically includes the following steps:

[0111] First, when the terminal interacts with the edge node, the edge node records and updates the terminal's behavior records in real time to form a context, and calculates the terminal's reliability;

[0112] Then, the edge nodes evaluate the direct trust of the terminal based on the latest reliability and the behavior at the current moment; at the same time, the current edge node continuously receives the recommended trust for the current terminal sent by other edge nodes, filters it through the constructed recommendation filtering model, and calculates the weighted average trust of all recommendations to obtain the final recommended trust value.

[0113] Finally, a comprehensive trust system is established by combining the final recommended trust and the direct trust of the computing terminal.

[0114] In one example of the present invention, in step S50, the expression for the comprehensive trust is:

[0115]

[0116] In the formula, and These are the weights for direct trust and indirect trust, respectively; where they are both in... Inside, satisfy ; For direct trust; To recommend and trust.

[0117] A trust assessment system for mobile edge computing terminal nodes according to a second aspect of the present invention includes:

[0118] The parameter update module is configured to generate local data at the terminal node, train the local model according to the model update instructions issued by the central server and edge nodes, and upload the local update parameters to the edge node.

[0119] The information sharing module is configured to allow edge nodes to collect behavioral information from terminal nodes and interact with other edge nodes to share recommendation information for terminal nodes.

[0120] The direct trust calculation module is configured to calculate the current trust and historical trust of a node based on behavioral factors such as transmission latency, interaction time difference, data integrity, node reliability, communication success rate, and number of interactions, and to obtain direct trust by integrating the current trust and historical trust.

[0121] The recommendation trust calculation module is configured to calculate the recommendation trust of nodes based on behavioral factors such as transmission latency, interaction time difference, data integrity, node reliability, communication success rate and number of interactions, and to filter malicious recommendations based on indicators such as recommendation timeliness, recommendation distance and recommendation external trust.

[0122] The integrated trust calculation module is configured to combine the final recommended trust and the direct trust calculation terminal to calculate the integrated trust.

[0123] This evaluation system not only ensures the comprehensiveness and accuracy of trust assessment and increases evaluation efficiency, but also solves the problem of unreliable trust evidence caused by malicious recommendations. It also avoids the inflexibility of using empirical thresholds. It enables secure and reliable distributed collaborative training in mobile edge computing, alleviates the problem of privacy information leakage, can quickly identify malicious nodes, accurately assess node trust, and effectively filter malicious recommendations. It has improved detection and evaluation performance as well as robustness.

[0124] In one example of the present invention, the recommendation trust calculation module includes:

[0125] The similarity checking unit is configured to check the similarity of the sender using the recommendation distance after receiving the recommendation from the edge node.

[0126] A similarity assessment unit is configured to verify the timeliness of recommendations sent by nodes in order to assess their temporal similarity.

[0127] The recommendation classification unit is configured to perform preliminary classification of received recommendations based on recommendation distance and recommendation timeliness in order to detect recommendation attacks; wherein, recommendations with a risk level of less than or equal to the lowest level are deleted.

[0128] External trust unit, configured to compute external trust for recommendations;

[0129] The honest recommendation unit is configured to detect and filter false recommendation attacks based on the preliminary classification results and external trust in the recommendations, and obtain the final honest recommendations.

[0130] This evaluation system enables secure and reliable distributed collaborative training in mobile edge computing, mitigating privacy information leakage issues. It can quickly identify malicious nodes, accurately assess node trust levels, and effectively filter malicious recommendations, thus improving detection and evaluation performance as well as robustness.

[0131] Specific examples:

[0132] like Figure 1 As shown, a trust assessment method for mobile edge computing terminal nodes includes:

[0133] Step 1: First, the terminal node generates local data and trains the local model according to the model update instructions issued by the central server and edge nodes. Then, the local update parameters are uploaded to the edge nodes.

[0134] Define the terminal node as Edge nodes are The number of terminal nodes and edge nodes is set to 100 each, and each terminal node is assigned an initial trust value of 0.5, ensuring that the initial state of the edge nodes is not zero trust. Edge nodes can communicate with each other, and the same terminal node can communicate with different edge nodes, thereby establishing recommended trust among the edge nodes.

[0135] Step 2: Edge nodes collect behavioral information from terminal nodes and interact with other edge nodes to share recommendation information for terminal nodes.

[0136] Terminal nodes are categorized into two types: benign nodes and malicious nodes. Benign nodes can upload the correct local model and transmit the correct data to edge nodes; malicious nodes can be faulty nodes or attacking nodes, whose purpose is to upload abnormal models and data or delay the upload of the current model, thereby affecting the aggregation of the global model.

[0137] Specifically, 100 terminal nodes were configured with behavioral information in six modes as shown in Table 1. In each round of simulation, the terminal nodes could behave completely benign or malicious, or alternate between benign and malicious behaviors. The purpose of setting up alternating verification in modes 3 and 4 is mainly to evaluate the detection performance of the proposed method as malicious behavior dynamically changes.

[0138] Table 1 Behavioral information of terminal nodes

[0139] model Behavioral information Mode 1 All edge nodes are benign. Mode 2 All edge nodes are malicious. Mode 3 Ten benign nodes and five malicious nodes alternate. Mode 4 Alternating between 10 benign nodes and 10 malicious nodes Mode 5 One benign node and one malicious node alternate Mode 6 Alternating between 5 benign nodes and 10 malicious nodes

[0140] Step 3: Calculate the current trust and historical trust of a node based on behavioral factors such as transmission latency, interaction time difference, data integrity, node reliability, communication success rate, and number of interactions.

[0141] The terminal's current trust is determined based on the interaction behavior at the current moment. The current trust is calculated as follows:

[0142]

[0143]

[0144]

[0145] in, for and The total number of interactions, For the first Abnormal factors during the second interaction For the first The latency factor during each interaction For the first The interval between the time of the next interaction and the current time. and The earlier the interaction occurs, the lower the proportion of anomaly factors and time delay factors in the calculation. This is because since interaction can occur again or multiple times... Interaction, Explanation Increasingly recognized terminal . The introduction of this property can digitize it. This is obtained by statistically analyzing the time difference between uploading and receiving local updates. Data integrity calculations are performed as follows:

[0146]

[0147] in, It is the number of times the data is complete. for and The number of successful interactions.

[0148] Reliability and communication success rate are both derived from the terminal's historical interaction records. Therefore, historical trust is defined as... The weighted average of behavioral reliability and communication success rate is calculated as follows:

[0149]

[0150] in, For the current moment, For the computing terminal in the first The moment of reliability of the next interaction.

[0151] The calculation method for direct trust is as follows:

[0152]

[0153]

[0154]

[0155] in, yes and The length of the context generated during the interaction.

[0156] Step 4: Calculate recommendation trust and filter malicious recommendations based on metrics such as recommendation timeliness, recommendation distance, and external trust in recommendations.

[0157] Recommended trust is represented as the relationship between the current edge node and other nodes. The weighted average of trust generated by each interacting edge node:

[0158]

[0159] in, This refers to the trust value generated by each edge node. To and The number of other edge nodes interacting.

[0160] Because of attacks, a node's identity, reputation, and other information can be altered. Therefore, recommendations generated by edge nodes are not always truthful and reliable. Although the forms of attacks vary, the characteristics that affect recommendation values ​​are consistent. To address this, we combined attack characteristics to statistically analyze recommendation distance and recommendation timeliness, and calculated external trust in the recommendation value to measure the honesty of the recommendations.

[0161] The recommended distance calculation method is as follows:

[0162]

[0163] in, Through All recommended centroid positions obtained, This is a randomly selected reference location. The range is .

[0164] The recommended method for calculating timeliness is as follows:

[0165]

[0166] in, It is the current time. To receive the transmission time included in the recommendation, a threshold is required. This is the time difference between when an edge node sends and receives recommendations. When the difference between the time contained in the received recommendation and the current time is greater than... At that time, This means that the current recommendations are outdated data and should be discarded.

[0167] The recommended method for calculating external trust is as follows:

[0168]

[0169] in, To and Interactive edge nodes The generated recommendation value, This represents the median of all recommendations, so that an initial trust can be attached to each recommendation. The range of values ​​is . The higher the value, the lower the external trust.

[0170] The membership function of fuzzy reasoning module 1 is as follows: Figure 3 As shown, the confusion matrix of the fuzzy rule is as follows: Figure 4 As shown, the risk level is divided into 6 levels based on the mapping results: extremely high (EH), very high (VH), high (H), low (L), very low (VL), and no risk (NR).

[0171] Because recommendations belonging to classes EH, VH, and H are very close in time and location, misclassification can occur, leading some reasonable recommendations to be mistakenly classified as attack-generated recommendations. Therefore, it is necessary to re-examine these recommendations to distinguish false recommendation attacks. A fuzzy inference module 2 was constructed to evaluate the external trust among recommendations belonging to classes EH, VH, and H. The membership function of fuzzy inference module 2 is as follows: Figure 5 As shown, the confusion matrix of the fuzzy rule is as follows: Figure 6 As shown.

[0172] Step 5: Combine the final recommended trust and the direct trust of the computing terminal to form a comprehensive trust.

[0173] The calculation method for overall trust is as follows:

[0174]

[0175] in, and These are the weights for direct trust and indirect trust, respectively. They are both in... Inside, satisfy .

[0176] To further illustrate the effectiveness of the proposed method, simulation experiments were conducted based on the terminal node and edge node information set in steps one and two. Under six behavioral modes, since existing TFL-DT and T-FedHA technologies do not include recommendation filtering, we only statistically analyzed the results of the proposed solution before and after filtering malicious recommendations, and compared them with the evaluation results without malicious recommendations.

[0177] When malicious recommendations are added to the system, some edge nodes provide false recommendation information to the current edge node. Figure 7 The results show that after adding malicious recommendations, the trust assessment of terminal nodes deviates significantly from the actual values. The main reason is that recommendation attacks cause the trust value of malicious nodes to increase or the trust value of benign nodes to decrease, thus affecting the overall trust assessment of terminal nodes and leading to errors.

[0178] By incorporating a recommendation filtering model into the solution, malicious recommendations are filtered out by analyzing factors such as the location and time of the recommendations, combined with the credibility of the recommendation values. After recommendation filtering, the trust assessment results for terminal nodes are ultimately closer to the actual trust values. Figure 7 The experimental results of (af) demonstrate the accuracy of the proposed scheme in node trust assessment, and also show that the scheme has the ability to identify malicious nodes and defend against and filter malicious recommendation attacks, which can ensure the reliable operation of edge nodes in the technology proposed in this invention.

[0179] Experimental results show that the proposed scheme can assess the trust level of terminal nodes quickly and accurately, ensuring the reliability of terminal nodes and data with its robust performance. Combined with a recommendation trust filtering model, it can restore the trust value of terminal nodes to the actual trust level, possessing the ability to identify malicious nodes and defend against and filter malicious recommendation attacks.

[0180] The foregoing description, with reference to preferred embodiments, details an exemplary implementation of the trust assessment method and system for mobile edge computing terminal nodes proposed in this invention. However, those skilled in the art will understand that various modifications and alterations can be made to the above specific embodiments without departing from the concept of this invention, and various combinations can be made to the various technical features and structures proposed in this invention without exceeding the protection scope of this invention, which is determined by the appended claims.

Claims

1. A trust assessment method for mobile edge computing terminal nodes, characterized in that, Includes the following steps: S10: The terminal node generates local data and trains the local model according to the model update instructions issued by the central server and edge nodes, and uploads the local update parameters to the edge nodes. S20: Edge nodes collect behavioral information from terminal nodes and interact with other edge nodes to share recommendation information for terminal nodes; S30: Calculate the current trust and historical trust of a node based on behavioral factors such as transmission delay, interaction time difference, data integrity, node reliability, communication success rate, and number of interactions, and obtain direct trust by integrating the current trust and historical trust; S40: Calculate the node's recommendation trust based on behavioral factors such as transmission delay, interaction time difference, data integrity, node reliability, communication success rate, and number of interactions, and filter malicious recommendations based on indicators such as recommendation timeliness, recommendation distance, and external trust in recommendations. S50: Integrated trust that combines final recommended trust and direct trust in computing terminals; In step S30, the current trust The calculation method is as follows: In the formula, For the first One terminal node, For the first One edge node, For the first time Second interaction For terminal nodes No. Reliability measurement during the first interaction The success rate is expressed as: In the formula, For terminal nodes With edge nodes The number of successful interactions; For terminal nodes With edge nodes The total number of interactions; In step S30, historical trust is defined as... Weighted average of behavioral reliability and communication success rate The calculation method is as follows: in, For terminal nodes No. Reliability measurement during the first interaction; Success rate; For terminal nodes With edge nodes The length of the context generated during the interaction; For the current moment, For the computing terminal in the first The moment when the reliability of the interaction is guaranteed; In step S30, direct trust is established. The calculation expression is as follows: In the formula, For terminal nodes With edge nodes Current trust during interaction; Trust in history; Let this be the weight of the current trust level; Let be the weight of historical trust; where the expressions for current trust weight and historical trust weight are: In the formula, For terminal nodes With edge nodes The length of the context generated during the interaction; For terminal nodes No. Reliability measurement during the first interaction; This represents the number of interactions.

2. The trust assessment method for mobile edge computing terminal nodes according to claim 1, characterized in that, terminal node No. Reliability measurement during the first interaction Its expression is as follows: In the formula, For the first Abnormal factors during the second interaction For the first The latency factor during the first interaction; For the first The interval between the time of the next interaction and the current time; Among them, the Abnormal factors during the second interaction The expression for data integrity calculation is as follows: In the formula, It is the number of times the data is complete. For terminal nodes With edge nodes The number of successful interactions.

3. The trust assessment method for mobile edge computing terminal nodes according to claim 1, characterized in that, In step S40, filtering malicious recommendations based on indicators such as recommendation timeliness, recommendation distance, and external trust in the recommendations specifically includes the following steps: S41: After receiving the recommendations from the edge nodes, use the recommendation distance to check the similarity with the sender; S42: Verify the timeliness of the recommendations to assess their temporal similarity; S43: Based on the recommendation distance and recommendation timeliness, perform preliminary classification of the received recommendations to detect recommendation attacks; among them, those with a risk level of less than or equal to low are deleted; S44: Calculate the external trust in the recommendations; S45: Based on the preliminary classification results and external trust recommendations, detect and filter false recommendation attacks to obtain the final honest recommendations.

4. The trust assessment method for mobile edge computing terminal nodes according to claim 1, characterized in that, Step S50 specifically includes the following steps: First, when the terminal interacts with the edge node, the edge node records and updates the terminal's behavior records in real time to form a context, and calculates the terminal's reliability; Then, the edge nodes assess the direct trust of the terminal based on the latest reliability and the behavior at the current moment; At the same time, the current edge node continuously receives recommendation trusts for the current terminal sent by other edge nodes. After filtering through the constructed recommendation filtering model, the weighted average trust of all recommendations is calculated to obtain the final recommendation trust value. Finally, a comprehensive trust system is established by combining the final recommended trust and the direct trust of the computing terminal.

5. The trust assessment method for mobile edge computing terminal nodes according to claim 1, characterized in that, In step S50, the expression for the comprehensive trust is: In the formula, and These are the weights for direct trust and indirect trust, respectively; where they are both in... Inside, satisfy ; For direct trust; To recommend and trust.

6. A trust assessment system for mobile edge computing terminal nodes, characterized in that, include: The parameter update module is configured to generate local data at the terminal node, train the local model according to the model update instructions issued by the central server and edge nodes, and upload the local update parameters to the edge node. The information sharing module is configured to allow edge nodes to collect behavioral information from terminal nodes and interact with other edge nodes to share recommendation information for terminal nodes. The direct trust calculation module is configured to calculate the current trust and historical trust of a node based on behavioral factors such as transmission latency, interaction time difference, data integrity, node reliability, communication success rate, and number of interactions, and to obtain direct trust by integrating the current trust and historical trust. The recommendation trust calculation module is configured to calculate the recommendation trust of nodes based on behavioral factors such as transmission latency, interaction time difference, data integrity, node reliability, communication success rate and number of interactions, and to filter malicious recommendations based on indicators such as recommendation timeliness, recommendation distance and recommendation external trust. The integrated trust calculation module is configured to combine the final recommended trust and the direct trust calculation terminal for integrated trust. Among them, current trust The calculation method is as follows: In the formula, For the first One terminal node, For the first One edge node, For the first time Second interaction For terminal nodes No. Reliability measurement during the first interaction The success rate is expressed as: In the formula, For terminal nodes With edge nodes The number of successful interactions; For terminal nodes With edge nodes The total number of interactions; Among them, historical trust is defined as Weighted average of behavioral reliability and communication success rate The calculation method is as follows: in, For terminal nodes No. Reliability measurement during the first interaction; Success rate; For terminal nodes With edge nodes The length of the context generated during the interaction; For the current moment, For the computing terminal in the first The moment when the reliability of the interaction is guaranteed; Among them, direct trust The calculation expression is as follows: In the formula, For terminal nodes With edge nodes Current trust during interaction; Trust in history; Let this be the weight of the current trust level; Let be the weight of historical trust; where the expressions for current trust weight and historical trust weight are: In the formula, For terminal nodes With edge nodes The length of the context generated during the interaction; For terminal nodes No. Reliability measurement during the first interaction; This represents the number of interactions.

7. The trust evaluation system for mobile edge computing terminal nodes according to claim 6, characterized in that, The recommendation trust calculation module includes: The similarity checking unit is configured to check the similarity of the sender using the recommendation distance after receiving the recommendation from the edge node. A similarity assessment unit is configured to verify the timeliness of recommendations sent by nodes in order to assess their temporal similarity. The recommendation classification unit is configured to perform preliminary classification of received recommendations based on recommendation distance and recommendation timeliness in order to detect recommendation attacks; wherein, recommendations with a risk level of less than or equal to the lowest level are deleted. External trust unit, configured to compute external trust for recommendations; The honest recommendation unit is configured to detect and filter false recommendation attacks based on the preliminary classification results and external trust in the recommendations, and obtain the final honest recommendations.

Images