A Method for Selecting Trusted Nodes for Unmanned Aerial Vehicles Based on Three-Dimensional Trust Variable Weight Fusion
By employing a three-dimensional trust-based variable weight fusion approach, which combines consensus behavior, communication quality, and energy consumption status, and utilizes variable weight fusion and time smoothing, the accuracy of node selection in drone swarms is solved, and the ability to identify malicious nodes is improved.
Patent Information
- Application Number
- CN202611577192.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-13
- Publication Date
- 2026-11-10
AI Technical Summary
In the absence of continuous central control, existing node selection methods for drone swarms are unable to accurately identify untrusted nodes. In particular, fixed weights in multi-dimensional assessments lead to abnormal dimensions being compensated by normal dimensions, and insufficient historical behavior or changes in energy consumption affect trust judgments, resulting in frequent misjudgments under collusion evaluation.
A three-dimensional trust assessment index is adopted, including consensus behavior, communication quality and energy consumption status. The trust value of drone nodes is calculated by variable weight fusion, and the direct and indirect trust values are combined for comprehensive evaluation. The accuracy is improved by time smoothing.
It improves the accuracy of drone trusted node selection, reduces the interference of 3D maneuvering and link fluctuations on node trust judgment, and enhances the ability to identify malicious nodes.
Smart Images

Figure CN123075385A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of network security and blockchain consensus technology for drone swarms, and specifically relates to a method for selecting trusted nodes for drones based on the fusion of three-dimensional trust conversion. Background Technology
[0002] Drone swarms typically consist of multiple drone nodes with sensing, communication, computing, and collaborative decision-making capabilities, and can be used in scenarios such as collaborative reconnaissance, disaster monitoring, edge computing, aerial relay, and on-chain mission data. Due to the high-speed movement of nodes, frequent changes in network topology, fluctuations in wireless links, and limited node resources, drone swarms are susceptible to abnormal behaviors such as silence, misdeployment, collusion, replay, and excessive energy consumption.
[0003] In order to identify untrusted nodes in the absence of continuous central control, existing solutions typically calculate trust values based on node history, packet forwarding behavior, or consensus results, and use these trust values for node selection, routing decisions, or blockchain consensus.
[0004] However, a single behavioral or communication metric is insufficient to reflect the overall trustworthiness of a drone node. For example, a node may have a good communication link and high remaining energy, but continuously generate incorrect digests during the consensus phase; it may also be misjudged as a silent or denial-of-service node due to temporary link interruptions caused by three-dimensional maneuvers.
[0005] To improve the completeness of the evaluation, nodes can be evaluated from multiple dimensions, including consensus behavior, communication quality, and energy consumption. However, when using fixed weights across multiple dimensions, low values in abnormal dimensions may be compensated for by high values in other normal dimensions. For example, a node whose consensus behavior trust value has significantly decreased may still receive a comprehensive score close to the trust threshold due to its high communication quality and remaining energy, thus allowing it to continue to participate in the consensus committee.
[0006] On the other hand, the behavioral dimension lacks sufficient historical consensus samples in the early stages of node startup; the energy consumption dimension is affected by normal propulsion energy consumption as flight time increases; and the communication quality dimension is affected by distance, elevation angle, shadow fading, and rapid fading. If cold start, propulsion energy consumption, and link reachability are not distinguished, it is easy to cause misjudgment of honest nodes and frequent switching of node sets.
[0007] Existing indirect trust methods may directly average neighbor ratings or assign weights based on the consistency between neighbor ratings and the target node's direct trust. In collusive evaluation scenarios, multiple malicious nodes can provide mutually consistent false ratings, thereby amplifying erroneous recommendations. Summary of the Invention
[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for selecting trusted nodes of drones based on three-dimensional trust variable weight fusion. It adopts three-dimensional evaluation indicators and uses a variable weight fusion method to evaluate the trust of drone nodes, which can improve the accuracy of drone trusted node selection.
[0009] The technical solution provided by this invention is as follows:
[0010] A method for selecting trusted nodes for unmanned aerial vehicles (UAVs) based on three-dimensional trust-based variable weight fusion includes:
[0011] Obtain consensus behavior metrics, communication quality metrics, and energy consumption status metrics for each drone node in the drone cluster.
[0012] The consensus behavior dimension indicators include: consensus success rate, consensus response rate, consensus message integrity rate, and voting consistency rate; the communication quality dimension indicators include: packet delivery rate, average latency, latency jitter, and link stability; and the energy consumption status dimension indicators include: remaining energy ratio, unit consensus energy consumption, and energy consumption rate.
[0013] Calculate the weights of each indicator within each dimension, and calculate the trust value of the drone node in each dimension based on the weights of the indicators within each dimension.
[0014] Determine the basic weights for each dimension, and based on the basic weights and the trust values of the drone nodes in each dimension, determine the actual weights of the trust values of the drone nodes in each dimension.
[0015] The direct trust value of the drone node is obtained by weighting and fusing the actual weights.
[0016] Determine the trusted recommended neighbors of the drone node, and calculate the trust value of the trusted recommended neighbors for the drone node, as the indirect trust value of the drone node;
[0017] The direct trust value and the indirect trust value are combined to obtain the comprehensive trust value of the drone node;
[0018] The drone node is selected as a trusted drone node based on the comprehensive trust value.
[0019] Preferably, the formula for calculating the weights within each dimension of the indicator is as follows:
[0020] ;
[0021] in, , , ;
[0022] In the formula, For the first Weights within each dimension of the indicator For the first The discriminatory power of each indicator For the first Information entropy of each indicator For drone nodes In the The proportion of each indicator This represents the number of drone nodes in the drone swarm. This is a preset minimum positive number to avoid the standardized denominator being zero. For drone nodes of The standardized values of each indicator; It represents the natural logarithm.
[0023] Preferably, the formula for calculating the trust value of the drone node in each dimension is as follows:
[0024] ;
[0025] in, For drone nodes In dimensions Trust value below, For dimension Inner The weight of each indicator, For drone nodes In dimension The next The standardized value of each indicator.
[0026] Preferably, the method for determining the actual weights of the trust values of drone nodes in each dimension is as follows:
[0027] Determine the upper bound of the variable weight for each dimension, and calculate the variable weight sharpness for each dimension based on the upper bound of the variable weight;
[0028] The variable weight state variables of the UAV node in each dimension are constructed using the variable weight sharpness and the trust value of the UAV node in each dimension, wherein:
[0029] ;
[0030] ;
[0031] ;
[0032] In the formula, Indicates drone node The variable power state quantity in the dimension of consensus behavior Indicates drone node Variable weight state quantities in the dimension of communication quality Indicates drone node Variable weight state quantities under the energy consumption state dimension; The basic weights representing the consensus behavior dimension This represents the basic weights of the communication quality dimension. The basic weights representing the energy consumption state dimension; Indicates drone node Trust value in the dimension of consensus behavior Indicates drone node Trust value in the dimension of communication quality Indicates drone node Trust value under the energy consumption state dimension; The sharpness of power shifts in the dimension of consensus behavior. Variable-weighted sharpness for the communication quality dimension. For the variable-weighted sharpness of the energy consumption state dimension, This is the energy consumption warning threshold;
[0033] Normalize the variable weight state variables of the drone node in each dimension to obtain the actual weight of the trust value of the drone node in each dimension.
[0034] Preferably, the formula for calculating the variable-weight sharpness of each dimension is as follows:
[0035] , ;
[0036] in, For dimension The sharpness of the change of power For dimension The upper limit of power transformation, For dimension The basic weights; Represents the consensus behavior dimension. Indicates the dimension of communication quality. This represents the energy consumption status dimension.
[0037] Preferably, the formula for calculating the direct trust value of a drone node is:
[0038] ;
[0039] in, Indicates drone node Direct trust value, Indicates drone node The actual weight of trust value under the consensus behavior dimension Indicates drone node The actual weight of trust value under the communication quality dimension Indicates drone node The actual weight of trust value under the energy consumption state dimension.
[0040] Preferably, the formula for calculating the indirect trust value of the drone node is:
[0041] ;
[0042] in, Indicates drone node Indirect trust value, Recommending neighbor nodes with trust For drone nodes Evaluation weights, Recommending neighbor nodes with trust For drone nodes The evaluation value, Indicates drone node A set of reliable recommended neighbors.
[0043] Preferably, the formula for calculating the comprehensive trust value is:
[0044] ;
[0045] In the formula, For drone nodes The overall trust value; The fusion coefficient represents direct trust.
[0046] Preferably, the UAV trusted node selection method based on three-dimensional trust-weighted fusion further includes:
[0047] The comprehensive trust value is smoothed over time to obtain the final trust value, and the drone node with the highest final trust value in the drone cluster is selected as the trusted drone node.
[0048] The formula for calculating the final trust value is as follows:
[0049] ;
[0050] In the formula, For drone nodes At the current assessment moment The final trust value, For drone nodes At the current assessment moment Overall trust value For drone nodes exist The final trust value at any given moment. This is the time smoothing coefficient.
[0051] The beneficial effects of this invention are:
[0052] The UAV trusted node selection method based on three-dimensional trust variable weight fusion provided by this invention adopts three-dimensional evaluation indicators and uses variable weight fusion to evaluate the trust of UAV nodes. This can weaken the compensation effect of high-value dimensions on abnormal weak dimensions under fixed weight, and reduce the interference of three-dimensional maneuvering, link fluctuation and natural power decay on node trust judgment, thereby improving the accuracy of UAV trusted node selection. Attached Figure Description
[0053] Figure 1 This is a schematic diagram of the drone trusted node selection method based on three-dimensional trust conversion fusion as described in this invention.
[0054] Figure 2 This is a schematic diagram of the overall architecture for drone node trust assessment as described in this invention.
[0055] Figure 3 This is a schematic diagram illustrating the alignment of the sliding window, evaluation period, and trust update time as described in this invention.
[0056] Figure 4 This is a schematic diagram of the three-dimensional trust index system for drone nodes described in this invention.
[0057] Figure 5 This is a schematic diagram comparing the penetration rate of malicious nodes under ablation conditions of different trust dimensions in the experimental examples of this invention.
[0058] Figure 6 This is a schematic diagram comparing the detection rates of malicious nodes under different weighted fusion mechanisms in the experimental examples of this invention. Detailed Implementation
[0059] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.
[0060] like Figure 1 As shown, the present invention provides a method for selecting trusted nodes of unmanned aerial vehicles based on three-dimensional trust-based variable weight fusion, and the specific implementation is as follows.
[0061] I. Overall Architecture and Data Acquisition
[0062] like Figure 2 As shown, the trust assessment architecture adopted in this invention includes a physical / network layer, a data acquisition and aggregation layer, a three-dimensional trust assessment layer, and a trust result application layer. Each layer can be implemented collaboratively by one or more of the following: the UAV node's local processor, the group master node, the ground control node, or the edge server.
[0063] The physical / network layer generates node motion state information, such as node location, velocity, link quality, packet transmission results, and energy state information. The data acquisition and aggregation layer maintains communication quality logs, neighbor set logs, consensus behavior logs, and energy consumption logs for each node, and uses a sliding window... Retain events from the most recent period.
[0064] The three-dimensional trust assessment layer in each assessment cycle Upon arrival, events within the sliding window are read. Within each dimension, the entropy weight method is used to calculate the dimensional trust value for the three dimensions. Based on the basic weights, the actual dimensional weights are calculated using the variable weight sensitive bottleneck formula. The weighted dimensional trust values yield the direct trust value, and the indirect trust value is obtained by integrating neighbor evaluations. The final trust value is obtained by weighting the direct and indirect trust values and performing time smoothing. Based on the trust result and ranking, the application layer performs trusted node selection, abnormal node isolation, group leader selection, or consensus node update.
[0065] like Figure 3 As shown, as a preferred option, if the sliding window... Seconds and evaluation cycle At the second level, the evaluation window for time 10 is [0 seconds, 10 seconds], and the evaluation window for time 15 is [5 seconds, 15 seconds]. Adjacent windows retain 5 seconds of overlapping data, which can avoid drastic jumps in the trust value when switching windows.
[0066] The node communication quality log records the event time, sender, receiver, message type, whether delivery was successful, one-way latency, and link quality; the neighbor set log records the neighbor set determined by the signal-to-noise ratio or packet delivery rate threshold at each sampling time; the consensus behavior log records whether a node has the conditions to participate, whether it responds, whether the vote is correct, whether the vote is consistent with the correct result, and the message integrity rate; the energy consumption log records the energy changes caused by different reasons such as propulsion, sending, receiving, and computation.
[0067] In a three-dimensional physical network implementation, drone nodes move within a region according to a three-dimensional Gauss-Markov process. (Drone node) The position is represented as drone nodes The position is represented as The three-dimensional distance and link elevation angle between the two are expressed as follows:
[0068] ;
[0069] ;
[0070] In the formula, and Each is a drone node and drone nodes 3D coordinates; For drone nodes and drone nodes The three-dimensional distance between them; This is the elevation angle of the link.
[0071] 3D distance Used for path loss and propagation delay calculations, link elevation angle The Rice factor is used to adjust the signal-to-noise ratio (SNR); path loss, Rice fading, and noise together determine the SNR, which in turn determines the transmission rate, bit error rate, and packet error rate. The resulting packet delivery rate, latency, jitter, and link stability of the UAV node can serve as data sources for communication quality dimensions.
[0072] The propulsion power of a drone is determined by both its horizontal and vertical velocity components. Propulsion energy consumption is used to assess the overall endurance of the drone, but is not directly used as a basis for judging communication / computation energy consumption attacks; the energy consumption generated by communication transmission, reception, and consensus computation is accumulated separately and used to calculate the unit consensus energy consumption and energy consumption rate.
[0073] II. Construction of Three-Dimensional Trust Indicators
[0074] like Figure 4 As shown, this invention constructs three trust dimensions: consensus behavior, communication quality, and energy consumption status. Each dimension contains multiple indicators with clearly defined positive and negative attributes, as detailed in Table 1.
[0075] Table 1. Indicators for Each Dimension
[0076]
[0077] For the consensus behavior dimension, a round is only counted in the behavior statistics if the node link is reachable, the relevant protocol message has been successfully delivered, and the node actually meets the conditions for participation. Honest nodes that cannot receive prepared messages due to link interruption are not counted as silent nodes.
[0078] For the energy consumption status dimension, the remaining energy ratio reflects the node's continuous service capacity, while unit consensus energy consumption and energy consumption rate are used to reflect abnormal communication or computing loads. Promoting separate energy consumption accounting is crucial to prevent normal energy consumption generated by three-dimensional ascents, descents, or high-speed maneuvers from masking energy consumption attack signals.
[0079] The specific calculation methods for each indicator are as follows. Let the current evaluation time be... The current sliding evaluation window is ,in, The length of the sliding evaluation window; the number of drone nodes to be evaluated is... drone nodes The effective consensus rounds are Each valid consensus round satisfies the requirements of the drone nodes. The link is reachable, the relevant protocol messages have been successfully delivered, and the drone node... Meeting the requirements for participation in the protocol; within the window, with drone nodes The relevant set of successful communication data packets is Within the sliding evaluation window, samples are obtained according to a preset sampling period. A set of neighbor samples, drone nodes In the The set of neighbors at each sampling time is denoted as .
[0080] (1) Consensus success rate The calculation formula is:
[0081] =
[0082] In the formula, For drone nodes The number of valid consensus rounds in which the consensus voting result or output result is consistent with the verified correct consensus result within the sliding evaluation window; This refers to the number of rounds in the set of valid consensus rounds.
[0083] (2) Consensus Response Rate The calculation formula is:
[0084] ;
[0085] In the formula, For drone nodes The number of rounds in a valid consensus round in which a response is completed before the timeout period specified in the protocol; The meaning is the same as above.
[0086] (3) Consensus message integrity rate The calculation formula is:
[0087] ;
[0088] In the formula, For drone nodes In the effective consensus round The actual number of protocol messages sent. This represents the number of protocol messages that should be sent in this round according to the consensus protocol; the summation range is... All valid consensus rounds in the process.
[0089] (4) Voting consensus rate The calculation formula is:
[0090] ;
[0091] In the formula, For drone nodes The number of valid votes that are consistent with objectively correct or verified results within the sliding evaluation window; For drone nodes The total number of valid votes actually generated within the window.
[0092] (5) Package delivery rate The calculation formula is:
[0093] ;
[0094] In the formula, For sliding evaluation window and drone node The number of relevant and successfully delivered data packets; This represents the total number of data packets that should be delivered under the same statistical criteria.
[0095] (6) Average delay The calculation formula is:
[0096] ;
[0097] In the formula, For successful communication of data packets End-to-end transmission delay; For sliding evaluation window and drone node The relevant set of successful communication data packets; .
[0098] (7) Delay jitter The calculation formula is:
[0099] ;
[0100] In the formula, , , and The meaning is the same as above; It is used to characterize the degree of dispersion of the end-to-end delay of successfully delivering data packets relative to the average delay.
[0101] (8) Link stability The calculation formula is:
[0102] ;
[0103] ;
[0104] In the formula, The number of times the neighbor set is sampled within the sliding evaluation window; and Representing drone nodes In the The set of neighbors at each sampling time and the previous time step; This represents the normalized change in the neighbor set between two adjacent sampling times; Represents the symmetric difference of sets. The union of sets, Indicates the number of elements in the set. When If the union is empty, then Set to the default neutral value or 1.
[0105] (9) Residual energy ratio The calculation formula is:
[0106] ;
[0107] In the formula, For drone nodes Remaining energy at the current assessment moment; For drone nodes The initial energy or rated maximum energy.
[0108] (10) Unit consensus energy consumption The calculation formula is:
[0109] ;
[0110] In the formula, For drone nodes The energy consumption increments generated by communication transmission, communication reception, and consensus computation within the sliding evaluation window do not include flight propulsion energy consumption. For drone nodes The number of effective consensus rounds within the sliding evaluation window.
[0111] (11) Energy consumption rate The calculation formula is:
[0112] ;
[0113] In the formula, The meaning is the same as above; This is the length of the sliding evaluation window. This metric only reflects the rate of change of communication and computing energy consumption per unit time.
[0114] When the denominator of any of the above ratios or average indicators is zero, the corresponding indicator can be set to a preset neutral value, or the corresponding indicator value from the previous evaluation period of the node can be used; consensus behavior indicators are based only on the set of effective consensus rounds. Statistics show that the unit consensus energy consumption and energy consumption rate only use communication energy consumption and computing energy consumption, thereby reducing misjudgments caused by link unreachability and normal flight propulsion energy consumption, respectively.
[0115] III. Intra-dimensional Entropy Weight Calculation
[0116] For each trust dimension, construct a system based on... One drone node and An indicator matrix composed of indicators To eliminate dimensional differences, the positive and negative indices in Table 1 were standardized respectively.
[0117] The formula for standardizing positive indicators is: ;
[0118] The formula for standardizing negative indicators is: ;
[0119] In the formula, For drone nodes In the The original values of each indicator for The standardized value, and The first The minimum and maximum values of each indicator in the current node set. A positive number is set to prevent the range from being zero or the logarithm from resulting in a zero value. When an indicator has no discriminative power in the current set of nodes, the standardized value of that indicator can be set to a neutral value. Specifically, It is a preset minimum positive number to avoid the standardized denominator being zero, and to smooth the proportion of indicators to avoid the undefined logarithm in the calculation of information entropy.
[0120] For the standardized indicators, calculate the indicator percentage, information entropy, discrimination, and in-dimension indicator weights in sequence:
[0121] ;
[0122] ;
[0123] ;
[0124] ;
[0125] In the formula, For drone nodes The proportion of the z-th indicator The number of drone nodes participating in the evaluation. For the first Information entropy of each indicator For the first The discriminatory power of each indicator For the first Weights within each dimension of the indicator.
[0126] The greater the information entropy, the smaller the difference of the indicator in the current node set, and the lower its weight within the dimension; the greater the discriminative power, the greater the contribution of the indicator to distinguishing the node state.
[0127] drone nodes In dimensions The trust value for the following dimension is represented as:
[0128] ;
[0129] In the formula, For drone nodes In dimensions Trust value below, For dimension Inner The weight of each indicator, These are the standardized values of the corresponding indicators; Represents the consensus behavior dimension. Indicates the dimension of communication quality. This represents the energy consumption status dimension.
[0130] Thus, the drone nodes were obtained respectively. Trust value in the consensus behavior dimension Trust value in the communication quality dimension Trust value for energy consumption status dimension The trust values for all three dimensions are limited to between 0 and 1.
[0131] Set the basic weights for each dimension of the pre-evaluation period. , and ,in, , and These are the basic weights for consensus behavior, communication quality, and energy consumption status of any drone node within the current evaluation period, respectively, and satisfy the following conditions: .
[0132] As a preferred option, to enable the variable weight strength to adapt to the basic weights, the upper bound of the variable weight for each dimension is determined based on the minimum and maximum values of the current three basic weights:
[0133] ;
[0134] In the formula, Indexed for the trust dimension; For dimension The upper limit of the change of power; For dimension The basic weights; and These are the minimum and maximum values of the basic weights in the three dimensions, respectively.
[0135] In other implementations, the upper bound of the variable weight can also be preset by configuration parameters. The variable weight sharpness is then determined by inverse solving based on the base weight and the upper bound of the variable weight.
[0136] ;
[0137] In the formula, For dimension The variable sharpness is used to control the rate at which the state quantity increases when the trust value of the corresponding dimension decreases. Represents the natural logarithm; and The meaning is the same as above, and satisfies .
[0138] For the consensus behavior dimension and the communication quality dimension, the variable weight state variables of each UAV node are constructed respectively:
[0139] ;
[0140] ;
[0141] In the formula, Indicates drone node The variable power state quantity in the dimension of consensus behavior Indicates drone node Variable weight state quantities in the dimension of communication quality; The basic weights representing the consensus behavior dimension The basic weights representing the communication quality dimension; Indicates drone node Trust value in the dimension of consensus behavior Indicates drone node Trust value in the dimension of communication quality; The sharpness of power shifts in the dimension of consensus behavior. Variable-weighted sharpness for the communication quality dimension; This represents an exponential function with the natural constant as its base.
[0142] The aforementioned state variables monotonically increase as the dimension trust value decreases. Therefore, low-trust dimensions receive higher weights in subsequent normalization, resulting in a bottleneck sensitivity effect.
[0143] Considering that the remaining energy of the drone node will naturally decrease as the normal mission time increases, this embodiment sets a warning line for the energy consumption status dimension. The energy consumption variable state variables of the drone node are represented as follows:
[0144] ;
[0145] In the formula, Indicates drone node Variable weight state quantities under the energy consumption state dimension The basic weights representing the energy consumption state dimension, Indicates drone node Trust value under the energy consumption state dimension For the variable-weighted sharpness of the energy consumption state dimension, This is the energy consumption warning line with a value between 0 and 1. hour, Keep as ; hour, Follow It decreases and then increases.
[0146] When the energy consumption state trust value is not lower than the warning line, the energy consumption dimension maintains a neutral basic weight; only when the energy consumption state deteriorates significantly will the penalty effect of the energy consumption dimension be increased.
[0147] Normalizing the variable-weight state variables of the three dimensions yields the actual weights of the trust values of the drone nodes in each dimension:
[0148] ;
[0149] in, ;
[0150] In the formula, Indicates drone node The actual weight of trust value under the consensus behavior dimension Indicates drone node The actual weight of trust value under the communication quality dimension Indicates drone node The actual weight of trust value under the energy consumption state dimension; For drone nodes The sum of the three dimensions of variable weight state quantities, and .
[0151] The final direct trust value for the drone node with variable weight is represented as follows:
[0152] ;
[0153] in, Indicates drone node Direct trust value, Indicates drone node The actual weight of trust value under the consensus behavior dimension Indicates drone node The actual weight of trust value under the communication quality dimension Indicates drone node The actual weight of trust value under the energy consumption state dimension. The value range is from 0 to 1.
[0154] When the trust values of the three dimensions are all high and the differences are small, the actual dimension weights are close to the base weights; when the trust value of a certain dimension decreases significantly, the actual weight of that dimension increases, so as to avoid its low value being masked by the high values of other dimensions.
[0155] Example of variable-weighted fusion computation: To illustrate the role of bottleneck-sensitive variable-weighting, let's assume the consensus behavior dimension trust value of a certain drone node. Trust value in the communication quality dimension Trust value for energy consumption status dimension The basic weight is Furthermore, the node has already passed the cold start period.
[0156] When using fixed weights, the direct trust value of this drone node is Abnormal behavior may be compensated by the other two high-value dimensions to be close to the isolation threshold.
[0157] According to the upper bound and sharpness calculation provided by this invention, the actual dimensional weights can be obtained as approximately The corresponding direct trust value for the change of power is approximately 0.383.
[0158] As this example shows, when the consensus behavior dimension becomes a low-value weakness, its actual weight automatically increases, and the overall trust value is further reduced, thereby improving the separability between abnormal nodes and honest nodes.
[0159] IV. Credible Indirect Trust and Time Smoothing
[0160] To enhance trust assessment by leveraging neighbor observations while reducing the impact of malicious recommendations, this embodiment employs a recommendation node admission screening based on a trust threshold.
[0161] Respondent credibility Find trusted recommended neighbor nodes The final trust value from the previous evaluation period is used to avoid using the current evaluation results to reverse-determine the evaluation weight in the same evaluation period.
[0162] Targeting drone nodes The set of trustworthy recommender neighbors is represented as:
[0163] ;
[0164] in, For nodes The current set of candidate recommended neighbors, As a trusted gating threshold, This is a set of anomalous nodes formed by historical verification evidence. Anomalous nodes, even if they maintain a link with the target node, are not eligible for recommendation. For drone nodes A set of reliable recommended neighbors; Recommend neighbor nodes as candidates; Retrieve candidate recommended neighbor nodes Final confidence level of the previous assessment period ; The trusted gating threshold is a value between 0 and 1; It is a set of anomalous nodes formed based on historical anomaly evidence.
[0165] According to drone nodes Counting positive consensus behavior within the sliding window and negative consensus behavior count Constructing behavioral testimonies with negative penalties:
[0166] ;
[0167] in, Not less than 1, used to amplify the impact of negative behavior on behavioral witnessing. In the formula, For drone nodes Behavioral witness value; For drone nodes Count of positive consensus actions within the current sliding evaluation window; Count negative consensus behaviors; The negative behavior penalty factor; the 1 in the numerator and the 2 in the denominator are smoothing constants, used to avoid undefined calculation results when there are few or no behavior samples.
[0168] Integrate behavioral witnessing and link witnessing into a per-neighbor evaluation:
[0169]
[0170] In the formula, Recommending neighbor nodes with trust Evaluation of drone nodes Neighbor rating score For drone nodes Behavioral witness value; Recommending neighbor nodes with trust For drone nodes The link witness value; and These are the fusion coefficients for behavioral witnessing and link witnessing, respectively. and Among them, trusted recommender neighbor nodes For drone nodes Link witness Package delivery rate observable by the neighbor Sure, .
[0171] The evaluator weighting considers both evaluator credibility and link observation reliability.
[0172] ;
[0173] In the formula, Recommend neighbors with credibility For drone nodes Evaluation weights; Recommending neighbor nodes with trust The final trust value of the previous assessment period; This is a credibility penalty index, used to amplify the weight differences between different credible evaluators; Neighboring nodes For drone nodes The link observation reliability ranges from 0 to 1, and can be taken as the link witness. Alternatively, it can be determined based on the number of valid link observation samples. The indirect trust value of node i is expressed as:
[0174] ;
[0175] In the formula, For drone nodes Indirect trust value; For nodes The set of trusted recommending neighbors; the numerator is the sum of the products of the neighbor-to-neighbor evaluation value and the corresponding evaluation weight of each trusted recommending neighbor; the denominator is the sum of the evaluation weights of all trusted recommending neighbors.
[0176] By combining direct and indirect trust values, a comprehensive trust value for the drone node is obtained:
[0177] ;
[0178] In the formula, For drone nodes The overall trust value for the current assessment period; and Unmanned nodes Direct and indirect trust values for the current assessment period; The fusion coefficient, representing direct trust, is primarily based on evidence of a node's directly observable behavior, communication, and energy consumption, with neighbor evaluations serving as supplementary evidence. A value of 0.7 is recommended. This is the fusion coefficient corresponding to the indirect trust value.
[0179] Afterwards, drone nodes with high overall trust values can be selected as trusted drone nodes.
[0180] As a preferred method, to reduce trust value flipping caused by single link fading or short-term behavioral anomalies, an exponentially weighted moving average is used for time smoothing:
[0181] ;
[0182] In the formula, This is the fusion coefficient between direct and indirect trust values. This is the time smoothing coefficient. This is the current assessment time. Among them, For nodes At the current assessment moment The final trust value; This represents the unsmoothed composite confidence value at the current evaluation moment. This is the final trust value for the previous evaluation period; The time smoothing coefficient requires a rapid response to anomalies. If a drone suddenly becomes silent, drops data incorrectly, or maliciously consumes energy, new observations must be able to quickly lower the trust value, so the current period cannot be too low. On the other hand, honest nodes cannot be immediately kicked out due to a momentary fading of the wireless link. Drone networks experience movement, shadow fading, fast fading, short-term packet loss, and latency fluctuations, so a certain amount of historical trust must be retained. A value of 0.6 is recommended.
[0183] The final trust value can be compared with one or more thresholds to classify nodes as trusted nodes, nodes to be observed, or abnormal nodes; or it can be directly sorted according to the final trust value for resource allocation and consensus node selection.
[0184] V. Trust Node Selection and Application Driven by Trust Result
[0185] First, based on the three-dimensional spatial location information of the drone nodes, the K-means clustering method is used to group the drone cluster. After grouping, trust calculation is performed on each drone node according to the aforementioned trust assessment method: the direct trust value of each drone node is calculated based on the trust values and corresponding actual weights of the consensus behavior dimension, communication quality dimension, and energy consumption state dimension; the trusted recommended neighbors of each drone node are determined, and the corresponding indirect trust values are calculated based on the behavioral witnesses and link witnesses provided by the trusted recommended neighbors; the direct trust values and indirect trust values are further fused, and the final trust value of each drone node is obtained through time smoothing.
[0186] Based on the final trust value, trusted nodes are selected within each node group. Specifically, the drone nodes within each node group are sorted from highest to lowest according to their final trust value, and trusted drone nodes are selected from the top-ranked drone nodes according to the preset number of trusted nodes for each node group. Thus, each node group can form a set of trusted nodes composed of drone nodes with high final trust values, providing candidate nodes for subsequent task allocation, group master node selection, or consensus node configuration.
[0187] The trusted drone node selection results can be used in scenarios such as drone mission execution node selection, group master node selection, and blockchain consensus node configuration. In group consensus applications, several nodes can be selected from the trusted drone nodes in each node group to form a consensus layer. The trusted drone node with the highest final trust value can serve as the group master node; nodes that do not enter the consensus layer but meet the corresponding protocol participation conditions can serve as verification nodes, thereby realizing group consensus node configuration driven by trust assessment results.
[0188] Test case
[0189] This experimental example uses a global logic clock, with a physical sampling step of 0.5 seconds, a trust evaluation period of 5 seconds, and a sliding window length of 10 seconds.
[0190] The basic weights are set to 0.50 for behavior dimension, 0.25 for communication quality dimension, and 0.25 for energy consumption dimension; the fusion coefficient of direct trust and indirect trust can be set to 0.70; the number of rounds for behavior cold start can be set to 50; and the energy consumption warning line can be set to 0.30.
[0191] The trusted indirect trust parameters can be set as follows: behavior witness weight 0.60, link witness weight 0.40, negative behavior penalty factor 2.0, evaluator credibility index 2.0, trust gating threshold 0.50, and minimum number of trusted neighbors 2. These parameters are for illustrative purposes only and can be adjusted according to the task risk level and network size.
[0192] The three-dimensional physical network can be set to a horizontal area of 1500m × 1500m and height layers of 100m to 400m; the average velocity is 15m / s, the Gauss-Markov memory factor is 0.75, and the upper limit of vertical velocity is 5m / s.
[0193] The wireless link can be configured with a 2.4 GHz carrier frequency, 10 MHz bandwidth, 20 dBm transmit power, -95 dBm noise power, 2.0 path loss index, 3 dB shadow standard deviation, 5 dB to 15 dB elevation-dependent Rice factor, and 1024 bit message length.
[0194] The attack injection can randomly select Byzantine nodes according to the proportion of malicious nodes during the initialization phase, and then configure the Byzantine nodes as silent, mis-voting, or colluding nodes according to a preset distribution. Silent nodes do not send the required protocol messages, mis-voting nodes send votes with different digests than the correct ones, and colluding nodes use a unified error digest.
[0195] Simulation tests may include: standard performance tests without malicious nodes; Byzantine fault tolerance tests with different proportions of malicious nodes; trust value tracking tests over a preset duration; and cross-tests with different numbers of nodes and different groups.
[0196] In an exemplary large-scale test, 160 drone nodes can be set, with three grouping conditions: 6 groups, 8 groups, and 10 groups. The proportion of multiple malicious nodes can range from 0 to 0.40, with 100 requests per test point and repeated 10 times.
[0197] Recordable output metrics include consensus success rate, average end-to-end latency, median latency, Latency, throughput, number of protocol messages, average trust value of honest nodes, average trust value of Byzantine nodes, trust difference, false positive rate of honest nodes, penetration rate of Byzantine nodes in the consensus layer, and number of node updates.
[0198] To verify the performance improvement of the multi-dimensional trust modeling and variable weight fusion mechanism in this invention on malicious node identification, ablation experiments were conducted under the same physical network conditions, attack node set, and random seed conditions. The experimental network consisted of 80 nodes divided into 8 node groups. Each group included 200 formal requests, with 50 rounds of warm-up before the formal experiments. The proportion of malicious nodes was set to 10%, 20%, 25%, 30%, 33%, and 40%, respectively, and the experiments were repeated using multiple fixed random seeds. The hybrid attack scenario included attack forms such as abnormal behavior, communication quality degradation, and abnormal energy consumption. The node isolation threshold was set to 0.5 to ensure that only the mechanism to be verified was changed in each group of experiments, while the other network environment and attack conditions remained consistent.
[0199] like Figure 5As shown, to verify the contribution of different trust dimensions to the anomaly detection capability of this invention, the behavior dimension, communication dimension, and energy consumption dimension were removed respectively, and compared with the complete three-dimensional trust model to form a dimension ablation experiment. Experimental results show that the complete three-dimensional model can consistently maintain the malicious node penetration rate at a low level under conditions where the proportion of malicious nodes is 10%–40%, with an average penetration rate of approximately 0.49%. After removing the behavior dimension, the average penetration rate increases to approximately 1.52%, and after removing the communication and energy consumption dimensions, it increases to approximately 9.57% and 12.08%, respectively. When the proportion of malicious nodes reaches 40%, the malicious node penetration rate of the complete three-dimensional model is only 1.54%, while after removing the behavior, communication, and energy consumption dimensions, it increases to 3.68%, 17.70%, and 20.10%, respectively. This indicates that the three trust dimensions of behavior, communication, and energy consumption can provide criteria for malicious behavior identification from different aspects, among which the communication state and energy consumption state have a more significant effect on inhibiting malicious nodes from entering the consensus layer.
[0200] To further verify the effectiveness of the weight-change mechanism of this invention, such as Figure 6 As shown, while maintaining the complete three-dimensional trust input, the variable-weight fusion method of this invention is compared with the equal-weight mechanism and the fixed-weight non-variable-weight fusion method to form a variable-weight mechanism ablation experiment. Experimental results show that as the proportion of malicious nodes increases from 10% to 40%, the malicious node detection rate of the variable-weight fusion method of this invention remains consistently between approximately 98.01% and 99.23%, with an average detection rate of approximately 98.78%; the detection rate of the fixed-weight non-variable-weight fusion method is approximately 27.50% to 33.12%, while the overall detection capability of the equal-weight mechanism is significantly lower. Meanwhile, the method of this invention can still maintain a consensus success rate of approximately 99.4% to 100% even with a high proportion of malicious nodes in the experiment.
[0201] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.
Claims
1. A method for selecting trusted nodes for unmanned aerial vehicles (UAVs) based on three-dimensional trust-based variable weight fusion, characterized in that, include: Obtain consensus behavior metrics, communication quality metrics, and energy consumption status metrics for each drone node in the drone cluster. The consensus behavior dimension indicators include: consensus success rate, consensus response rate, consensus message integrity rate, and voting consistency rate; the communication quality dimension indicators include: packet delivery rate, average latency, latency jitter, and link stability; and the energy consumption status dimension indicators include: remaining energy ratio, unit consensus energy consumption, and energy consumption rate. Calculate the weights of each indicator within each dimension, and calculate the trust value of the drone node in each dimension based on the weights of the indicators within each dimension. Determine the basic weights for each dimension, and based on the basic weights and the trust values of the drone nodes in each dimension, determine the actual weights of the trust values of the drone nodes in each dimension. The direct trust value of the drone node is obtained by weighting and fusing the actual weights. Determine the trusted recommended neighbors of the drone node, and calculate the trust value of the trusted recommended neighbors for the drone node, as the indirect trust value of the drone node; The direct trust value and the indirect trust value are combined to obtain the comprehensive trust value of the drone node; The drone node is selected as a trusted drone node based on the comprehensive trust value.
2. The UAV trusted node selection method based on three-dimensional trust-based variable weight fusion according to claim 1, characterized in that, The formula for calculating the weights within each dimension of each indicator is as follows: ; in, , , ; In the formula, For the first Weights within each dimension of the indicator For the first The discriminatory power of each indicator For the first Information entropy of each indicator For drone nodes In the The proportion of each indicator This represents the number of drone nodes in the drone swarm. This is a preset minimum positive number to avoid the standardized denominator being zero. For drone nodes of The standardized values of each indicator; It represents the natural logarithm.
3. The UAV trusted node selection method based on three-dimensional trust variable weight fusion according to claim 2, characterized in that, The formulas for calculating the trust value of drone nodes in each dimension are as follows: ; in, For drone nodes In dimensions Trust value below, For dimension Inner The weight of each indicator, For drone nodes In dimension The next The standardized value of each indicator.
4. The UAV trusted node selection method based on three-dimensional trust variable weight fusion according to any one of claims 1-3, characterized in that, The method for determining the actual weights of the trust values of drone nodes across various dimensions is as follows: Determine the upper bound of the variable weight for each dimension, and calculate the variable weight sharpness for each dimension based on the upper bound of the variable weight; The variable weight state variables of the UAV node in each dimension are constructed using the variable weight sharpness and the trust value of the UAV node in each dimension, wherein: ; ; ; In the formula, Indicates drone node The variable power state quantity in the dimension of consensus behavior Indicates drone node Variable weight state quantities in the dimension of communication quality Indicates drone node Variable weight state quantities under the energy consumption state dimension; The basic weights representing the consensus behavior dimension This represents the basic weights of the communication quality dimension. The basic weights representing the energy consumption state dimension; Indicates drone node Trust value in the dimension of consensus behavior Indicates drone node Trust value in the dimension of communication quality Indicates drone node Trust value under the energy consumption state dimension; The sharpness of power shifts in the dimension of consensus behavior. Variable-weighted sharpness for the communication quality dimension. For the variable-weighted sharpness of the energy consumption state dimension, This is the energy consumption warning threshold; Normalize the variable weight state variables of the drone node in each dimension to obtain the actual weight of the trust value of the drone node in each dimension.
5. The UAV trusted node selection method based on three-dimensional trust-based variable weight fusion according to claim 4, characterized in that, The formulas for calculating the variable-weight sharpness of each dimension are as follows: , ; in, For dimension The sharpness of the power shift, For dimension The upper limit of power transformation, For dimension The basic weights; Represents the consensus behavior dimension. Indicates the dimension of communication quality. This represents the energy consumption status dimension.
6. The UAV trusted node selection method based on three-dimensional trust-based variable weight fusion according to claim 5, characterized in that, The formula for calculating the direct trust value of a drone node is: ; in, Indicates drone node Direct trust value, Indicates drone node The actual weight of trust value under the consensus behavior dimension Indicates drone node The actual weight of trust value under the communication quality dimension Indicates drone node The actual weight of trust value under the energy consumption state dimension.
7. The UAV trusted node selection method based on three-dimensional trust-based variable weight fusion according to claim 6, characterized in that, The formula for calculating the indirect trust value of the drone node is as follows: ; in, Indicates drone node Indirect trust value, Recommending neighbor nodes with trust For drone nodes Evaluation weights, Recommending neighbor nodes with trust For drone nodes The evaluation value, Indicates drone node A set of reliable recommended neighbors.
8. The UAV trusted node selection method based on three-dimensional trust variable weight fusion according to claim 7, characterized in that, The formula for calculating the comprehensive trust value is as follows: ; In the formula, For drone nodes The overall trust value; The fusion coefficient represents direct trust.
9. The UAV trusted node selection method based on three-dimensional trust variable weight fusion according to claim 8, characterized in that, Also includes: The comprehensive trust value is smoothed over time to obtain the final trust value, and the drone node with the highest final trust value in the drone cluster is selected as the trusted drone node. The formula for calculating the final trust value is as follows: ; In the formula, For drone nodes At the current assessment moment The final trust value, For drone nodes At the current assessment moment Overall trust value For drone nodes exist The final trust value at any given moment. This is the time smoothing coefficient.
