Network modeling method for reliable communication of unmanned aerial vehicle cluster

Abstract

The invention belongs to the technical field of unmanned aerial vehicle network communication and modeling, and particularly relates to an unmanned aerial vehicle cluster reliable communication-oriented network modeling method, which comprises the following steps of: establishing a unified state space containing dynamic characteristics and resource bearing capacity, and constructing a potential energy incidence matrix between neighbor nodes based on relative motion stability; in the unified state space, probability modeling is carried out on the motion behavior of the node in the prediction time window, and a position probability density function of the node at each moment in the future is obtained; calculating link connectivity probability based on the position probability density function, and performing weighted modeling on effective transmission delay, delay jitter and comprehensive packet loss probability of the link by taking the link connectivity probability as a reliability constraint factor; a dynamic attribute weight vector is given to each node, a probabilistic index evaluation vector is given to each edge according to the link connectivity probability, the effective transmission delay, the delay jitter and the comprehensive packet loss probability, and a graph model is obtained; according to the method, the problems of static model failure and risk perception deficiency are solved.

Description

Technical Field

[0001] This invention belongs to the field of UAV network communication and modeling technology, specifically relating to a network modeling method for reliable communication of UAV swarms. Background Technology

[0002] As drone technology continues to evolve towards swarming and collaboration, drone swarms are increasingly being used in industrial monitoring, emergency rescue, intelligent inspection, and battlefield reconnaissance. These applications typically involve the transmission of critical data such as industrial control commands, sensor data, and images and videos, placing stringent deterministic service requirements on communication networks. This means ensuring low latency, low jitter, and low packet loss rates simultaneously in complex and dynamic environments. If the network fails to consistently maintain these communication metrics within preset thresholds, it can easily lead to control failures, information loss, and even mission failure, severely impacting system security and mission reliability.

[0003] However, current UAV network modeling methods still have significant shortcomings in supporting highly dynamic, task-driven deterministic communication services, mainly in the following three aspects:

[0004] First, static modeling based on geometric distance or instantaneous topology is ill-suited to the dynamic nature of tasks. Existing methods often employ unit disk graphs or snapshot models based on instantaneous positions, constructing network topology solely based on current geometric distances or connectivity states. This type of modeling is essentially a static deterministic approach, failing to consider the continuous motion and state evolution of UAVs during mission execution. In scenarios involving high-speed UAV maneuvers or complex tactical maneuvers, the currently closest node may rapidly lose connection due to high-speed relative motion or sudden changes in heading. Topologies generated using such static models cannot predict future link connectivity risks, leading to frequent failures of upper-layer routing and scheduling strategies, ultimately causing service interruptions and performance fluctuations.

[0005] Second, trajectory modeling based on simplified motion assumptions lacks a quantitative description of uncertainty. While some studies have attempted to introduce node mobility prediction, most still rely on ideal motion models such as uniform linear motion or random waypoints, or assume that velocity follows a static Gaussian distribution. This makes it difficult to accurately characterize the frequent acceleration, sharp turns, obstacle avoidance, and other nonlinear maneuvers of UAVs in real-world missions. More importantly, existing methods mostly only output a single-point "expected trajectory" without modeling the probability distribution of prediction errors. This deterministic trajectory extrapolation cannot quantify the range of positional uncertainty, making it difficult for network models to assess the probabilistic risk of link interruption and failing to meet the stringent risk perception requirements of high-reliability communication.

[0006] Third, there is a lack of a unified quantitative modeling framework for multi-dimensional Quality of Service (QoS) constraints. Existing UAV network modeling research often focuses on connectivity analysis or only considers single performance indicators such as hop count and average latency, failing to systematically integrate multi-dimensional deterministic requirements such as end-to-end latency, latency jitter, packet loss rate, and link reliability. In UAV swarms with multiple services coexisting, different task flows have varying requirements for communication resources and reliability. Existing methods often introduce link quality assessment only during the routing or scheduling phase, failing to explicitly map the uncertainty of node movement into various QoS indicators in the future time domain during the basic network modeling phase. This results in the model being unable to accurately reflect the actual service carrying capacity of the network during dynamic operation.

[0007] In summary, existing UAV network modeling methods still face key challenges when dealing with task-driven dynamic scenarios, such as insufficient adaptability of static models, difficulty in quantifying motion uncertainties, and lack of unified representation of multi-dimensional service quality indicators. Summary of the Invention

[0008] To address the technical challenges faced by existing UAV network modeling methods in handling dynamic task scenarios, such as the failure of static snapshot models, the difficulty in accurately quantifying node motion uncertainties, and the lack of a unified representation framework for multi-dimensional reliability indicators, this invention proposes a network modeling method for reliable communication in UAV swarms.

[0009] The specific plan includes the following steps:

[0010] S1. Construct a unified state space and potential energy correlation matrix: Treat each UAV in the UAV cluster as an independent node, establish a unified state space that includes its dynamic characteristics and resource carrying capacity, and construct a potential energy correlation matrix between neighboring nodes based on relative motion stability.

[0011] S2. Constructing the trajectory probability distribution: In a unified state space, the motion behavior of nodes within the prediction time window is probabilistically modeled to predict the trajectory probability distribution of the acceleration vector of each node, thereby obtaining the position probability density function of each future time.

[0012] S3. Construct a multi-dimensional link quality quantification model with link connectivity probability as the core constraint: Calculate the link connectivity probability based on the location probability density function, and use the link connectivity probability as a reliability constraint factor to perform weighted modeling of the effective transmission delay, delay jitter and overall packet loss rate of the link.

[0013] S4. Construct a probabilistically weighted node-link directed graph model: Assign dynamic attribute weight vectors to each node based on a unified state space, and assign probabilistic index evaluation vectors to each edge based on link connectivity probability, effective transmission delay, delay jitter, and overall packet loss rate, to obtain a graph model that represents the real-time dynamic evolution trend of the UAV swarm.

[0014] The beneficial effects of this invention are:

[0015] This invention abandons the traditional static modeling method based on instantaneous geometric distance. By introducing a multidimensional state representation containing acceleration statistical features and a potential energy correlation matrix, it can effectively capture the dynamic evolution trend of UAV swarms during mission execution and is suitable for complex scenarios with high-speed maneuvering and frequent topological changes.

[0016] Unlike existing technologies that typically rely on simplified assumptions (such as uniform linear motion) to predict deterministic trajectories, this invention utilizes the probability distribution of neural network-predicted trajectories to quantify prediction errors and location uncertainties, thereby endowing the network model with the ability to perceive the risk of future link disruptions.

[0017] This invention proposes a modeling mechanism with link connectivity probability as the core constraint, which explicitly maps the connection risk at the topology level into a decrease in performance indicators such as latency, jitter, and packet loss rate. This can guide upper-layer protocols to actively avoid "pseudo-high-quality" links that are physically close but have poor stability, significantly improving the determinism and robustness of UAV swarm communication. Attached Figure Description

[0018] Figure 1 This is a flowchart of the UAV swarm network modeling based on location probability distribution prediction in this invention;

[0019] Figure 2 This is a diagram of the neural network architecture used in this invention to predict the probability distribution of UAV node positions. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figures 1-2This invention provides a network modeling method for reliable communication in UAV swarms. The core objective is to address the limitations of traditional deterministic snapshot modeling methods for UAVs operating in highly dynamic, mission-driven environments. By introducing a probabilistic modeling perspective, this invention predicts the probability distribution of future UAV movements, thereby quantifying their spatiotemporal uncertainty. Based on this, the invention proposes using link connectivity probability as the fundamental constraint for link reliability, explicitly mapping the dynamic evolution of the physical topology to key communication indicators such as latency, jitter, and packet loss rate, and then weighting and quantifying these indicators. This method constructs a network graph model with maneuver risk perception capabilities, thus providing effective support for upper-layer deterministic routing decisions.

[0022] In some embodiments, a network modeling method for reliable communication in unmanned aerial vehicle (UAV) swarms includes the following steps:

[0023] S1. Construct a unified state space and potential energy correlation matrix: Treat each UAV in the UAV cluster as an independent node, establish a unified state space that includes its dynamic characteristics and resource carrying capacity, and construct a potential energy correlation matrix between neighboring nodes based on relative motion stability.

[0024] This invention proposes a feature space construction method that can deeply characterize the movement trends and connectivity potential of unmanned aerial vehicle (UAV) swarms during mission execution, addressing the dynamic characteristics such as high-speed maneuverability and frequent topological changes exhibited by UAV swarms. The aim is to provide a foundational basis for describing dynamic characteristics in subsequent network modeling.

[0025] In some embodiments, step S1 specifically includes:

[0026] S11. When performing specific tasks (such as obstacle avoidance, formation changing, etc.), the dynamic characteristics of UAVs directly affect the stability of the communication link. Therefore, this invention considers the statistical characteristics of their maneuvering behavior and constructs a unified state space S for UAV swarm nodes, where the state space of node i represents S. i for:

[0027]

[0028] In the formula, pos i v represents the instantaneous position coordinates of node i in three-dimensional space. i This represents the velocity vector of node i in three-dimensional space, expressed through pos. i and v i Characterizes the current motion state of the drone; This represents the average acceleration of node i within a preset historical sliding window. C represents the variance of the acceleration of node i within a preset historical sliding window; i This represents the business processing capacity characteristics of node i, including the current queue backlog.

[0029] Specifically, in embodiments of the present invention, the acceleration variance Defined as the intensity of maneuver uncertainty, its value is directly related to the dynamic behavior of the UAV. A larger value indicates that the UAV is currently or is performing complex tactical maneuvers, or is subject to significant environmental interference. In this case, the probability prediction range of the UAV's future trajectory (i.e., the probability cloud) will expand accordingly, reflecting an increase in system state uncertainty.

[0030] S12. In the initial stage of modeling, to effectively select key neighbors and characterize the relative stability of relationships between nodes, a potential energy correlation matrix M for UAV swarm nodes is constructed, where m ij This represents the degree of association between node i and node j. The specific construction process includes:

[0031] S121. Calculate the distance potential energy term E between node i and node j. dist (i,j):

[0032]

[0033] In the formula, d ij σ represents the Euclidean distance between node i and node j; dist The distance sensitivity coefficient is used to adjust the rate of change of potential energy as distance increases; by introducing the distance potential energy term, the connectivity between nodes is quantified from a spatial geometric dimension.

[0034] S122. Calculate the motion stability factor E between node i and node j. move (i,j):

[0035]

[0036]

[0037] In the formula, Sim dir (i,j) represents the cosine similarity of the velocity directions between node i and node j, λ represents the relative velocity penalty coefficient, and v i v j Let ||·|| represent the velocity vectors of nodes i and j, and ||·|| represent the vector norm. This represents the relative velocity magnitude. When the velocity vectors of two nodes are at the same height, the relative velocity magnitude tends to 0, and the cosine similarity of the velocity directions tends to 1. At this point, the motion stability factor E... move When (i,j) approaches 1, it indicates that the motion stability is the best and the contribution to the correlation tightness is the highest.

[0038] S123. By fusing the distance potential energy term and the motion stability factor, the correlation density m between node i and node j is obtained. ij:

[0039]

[0040] In the formula, α represents the balance weight coefficient, which is used to balance the distance potential energy term and the motion stability factor.

[0041] Therefore, this invention distinguishes between "node pairs with similar motion trends" and "node pairs with diverging motion trends" from the modeling source, demonstrating its adaptability to dynamic task scenarios.

[0042] By modeling the unified state space and potential energy correlation matrix, a basic network state description that can reflect the dynamic maneuvering characteristics and potential communication relationships of UAV swarms is formed, providing input conditions for subsequent link reliability quantification modeling based on trajectory prediction uncertainty.

[0043] S2. Constructing the trajectory probability distribution: In a unified state space, the motion behavior of nodes within the prediction time window is probabilistically modeled to predict the trajectory probability distribution of the acceleration vector of each node, thereby obtaining the position probability density function of each future time.

[0044] This invention proposes a stochastic trajectory prediction method for unmanned aerial vehicles (UAVs) based on acceleration statistical modeling. Unlike the traditional approach of directly predicting deterministic trajectories, this study treats the acceleration that dominates trajectory changes during UAV movement as a random variable and models its statistical characteristics. For the frequent acceleration, deceleration, and turning maneuvers of UAVs during missions, analysis of experimental datasets such as Mid-Air reveals that UAV acceleration has more stable statistical characteristics than velocity and better conforms to a Gaussian distribution. Based on this, this invention assumes that the UAV's acceleration follows a Gaussian distribution to characterize its stochastic maneuvering behavior under various motion states. Furthermore, considering that the parameters of the acceleration distribution dynamically change with real-time velocity, this invention introduces a dynamic parameter modeling mechanism, enabling the acceleration distribution parameters to adaptively adjust according to the UAV's current motion state.

[0045] In some embodiments, it is assumed that the acceleration vectors of each node are... All follow a Gaussian distribution , where the mean vector With covariance matrix (Mainly reflected in the standard deviation vector) All of these are the current velocity vectors of the node. The function, i.e. and This allows the model to adaptively adjust the prediction boundary of the maneuver range based on different flight states of the UAV. Building on this, to capture the randomness of UAV maneuvering behavior, this embodiment employs a deep neural network model including Long Short-Term Memory (LSTM) and Hybrid Density Network (MDN) to predict the trajectory probability distribution of each node, such as... Figure 2 As shown, the specific process includes:

[0046] S21. Extract node i within the historical time window T obs The state-space representation within the network is used to construct the network input X. i,t ,in:

[0047]

[0048] In the formula, Let represent the instantaneous position coordinates of node i at time tk. This represents the velocity vector of node i at time tk. This represents the average acceleration of node i at time tk. This represents the variance of the acceleration of node i at time tk;

[0049] S22. Use a Long Short-Term Memory (LSTM) network to process the network input X. i,t By iteratively passing a forget gate, an input gate, and an output gate, the spatiotemporal correlation features of node motion are extracted to identify its current maneuvering mode (e.g., level flight, sharp turn, acceleration, etc.), thus obtaining a deep implicit state vector h that represents the motion trend. t ; can be represented as:

[0050]

[0051] S23. Transfer the deep hidden state vector h t Input the probability output layer of the hybrid density network to predict the acceleration distribution parameter set of node i at the next K time steps, where the acceleration distribution parameter of node i at the next time step τ is... , This represents the expected acceleration of node i at a future time τ. This represents the degree of uncertainty of node i at a future time τ. This represents the correlation coefficient between the components of node i at a future time τ;

[0052] S24. Based on the acceleration distribution parameter set and combined with the UAV kinematic equations, derive the trajectory probability distribution of node i in the next K time steps.

[0053] In some embodiments, based on the acceleration distribution parameter set output by the model and combined with the UAV's kinematic equations, the probability distribution of its position at future times can be derived. Since the Gaussian distribution is closed under linear transformation, the analytical form of the probability distribution of the node's position at future times can be obtained by performing a quadratic integration on the acceleration distribution parameter set.

[0054] Specifically, step S24 is based on the acceleration distribution parameters of node i at future time τ. Calculate the probability density function of the position of node i at a future time τ, including:

[0055] S241. Based on the expected acceleration Calculate the probabilistic cloud center trajectory of node i (i.e., the optimal estimated position).

[0056] Specifically, discrete-time kinematic equations are used, based on nodes. At the present moment initial state (position) With speed ), for the next K time points (step size is Perform recursive state updates:

[0057]

[0058]

[0059] In the formula, τ = 1, 2, ..., K represents the initial condition position. ,speed , This is the calculated probability cloud center position vector at the τ-th future time.

[0060] S242. Based on the degree of uncertainty Construct the covariance matrix that diffuses over time .

[0061] Specifically, firstly, the standard deviation vector output by the hybrid density network is used. Correlation coefficient between components Construct the acceleration covariance matrix Its elements are defined as:

[0062]

[0063] Subsequently, based on the error propagation law, the uncertainty of acceleration is mapped to the uncertainty of position. Considering that position is the second integral of acceleration with respect to time, and the uncertainty increases to the fourth power with the prediction step size, the position covariance matrix is ​​calculated as follows:

[0064]

[0065] In the formula, This represents the cumulative process noise covariance term.

[0066] S243. Based on the probability cloud center trajectory Covariance Matrix Calculate the probability density function f of the position of node i at a future time τ. i (x i ,τ), is represented as:

[0067]

[0068] In the formula, x i This represents the spatial coordinates of node i. Indicates transpose. This represents finding the inverse of a matrix. Represents a determinant.

[0069] This probability distribution-based prediction method can explicitly output the probability that a node will deviate from its predetermined route due to the accumulation of random maneuvers later in the prediction time window. This result provides accurate input for subsequent calculations of link connectivity probabilities, enabling subsequent communication quantification models to effectively perceive connectivity risks caused by physical movement.

[0070] S3. Construct a multi-dimensional link quality quantification model with link connectivity probability as the core constraint: Calculate the link connectivity probability based on the location probability density function and potential energy correlation matrix, and use the link connectivity probability as a reliability constraint factor to perform weighted modeling of the effective transmission delay, delay jitter and overall packet loss rate of the link.

[0071] This invention breaks through the limitations of traditional independent calculation of communication indicators by constructing a hierarchical modeling system with link connectivity probability as the core weight, and realizes the leap from mathematical expectation modeling to probability distribution quantitative modeling.

[0072] In some embodiments, link connectivity probability is a fundamental metric for evaluating the stability of the physical topology. Before calculating link connectivity probability, [the following is used:] As a priori screening threshold, only if the correlation between nodes i and j is strong. Only when the (preset threshold) is reached will the subsequent link connectivity probability P be calculated. link Calculate; otherwise, directly determine that the link is unavailable. This approach utilizes... This invention quickly eliminates false neighbors with diverging movement trends, reducing computational overhead. Furthermore, it breaks away from the traditional "0 / 1" binary decision model based on deterministic distance, instead employing the probability density function f of the future positions of node i and node j at time τ. i (x i,τ) and f j (x j Convolution and joint integral calculations are performed using τ.

[0073] The link connectivity probability is calculated based on the location probability density function and is expressed as follows:

[0074]

[0075] In the formula, f represents the spatial integration variable. i (x i Let I(·) denote the probability density function of the position of node i at future time τ, and let R be the indicator function. th P represents the effective communication radius. link (i,j,τ) represents the probability that the link between node i and node j will remain connected at the predicted time τ in the future.

[0076] The link connectivity probability metric is used to quantify the risk of communication link disruption caused by the uncertainty of UAV maneuvers. In actual calculations, the link connectivity probability value changes dynamically with the increase of the prediction step size, thus reflecting the risk evolution trend of the link within a future time window.

[0077] In some embodiments, the effective transmission delay, delay jitter, and overall packet loss rate of the link are weighted and modeled using the link connectivity probability as a reliability constraint factor, including:

[0078] S31. Define the effective transmission delay D of the link. eff :

[0079]

[0080] In the formula, E[D prop ] represents the probabilistic expectation of the dynamic propagation delay of the link, D trans D represents the transmission delay of the link. proc E[D] represents the processing latency of the link. queue ] represents the expected probability of queuing delay for a link. This formula shows that when the probability of link connectivity decreases, potential link outages may lead to data retransmission, route reconstruction, or buffering waits, thus inversely increasing the perceived effective delay.

[0081] Specifically, this embodiment of the invention addresses the dynamic nature and operational characteristics of drones by adjusting the link delay D between node i and node j. ij The latency is quantified as a superposition of propagation delay, transmission delay, processing delay, and queuing delay. Simultaneously, this invention uses LCP (Link Connectivity Probability) to weight and correct the above delays to obtain the effective transmission delay D of the link.eff The specific calculations for each term in the above formula are as follows:

[0082] Since the node locations are probabilistically distributed, the propagation delay is no longer a fixed value, E[D prop The formula for calculating ] is:

[0083]

[0084] In the formula, c represents the propagation speed of electromagnetic waves.

[0085] The transmission delay D of the link trans Related to links and link bandwidth:

[0086]

[0087] In the formula, B k R represents the length of the data frame. ij Indicates link e ij The sending rate.

[0088] Link processing delay D proc This represents the time it takes for a node to forward a data frame. Since the processor and workload of the drone node are known, the processing latency can be considered a constant value, denoted as D. proc =t proc , t proc It represents a constant value.

[0089] D queue This represents queuing latency, indicating the time a data frame spends waiting in the queue. In this scenario, it is assumed that nodes use a "first-come, first-served" scheduling rule, allocating dedicated priority queues for deterministic services to avoid resource contention with ordinary services. It is also assumed that service arrivals follow a Poisson distribution, the forwarding process of services by node i follows a binomial distribution, and the probability of service transmission at node i is p. ci The node's business processing speed is h di Each node has a fixed cache space threshold. Services exceeding the cache capacity will experience packet loss due to overflow, which is considered a deterministic scheduling failure.

[0090] Assume the maximum cache space capacity of the drone is Q. max The initial queue space occupied is Q. cur,i (Q cur,i ≤Q max In deterministic business f k (Data volume is B) k For example, the time from its generation to its forwarding to node i is Δt. i During this period, the arrival and processing of services at node i can be categorized into three scenarios, each corresponding to a different method for calculating queuing delay:

[0091] No need to queue: if △t i The number of service packets received by the internal node x and the initial queue occupancy Q cur,i The sum of these numbers is less than or equal to the number of business packages y (i.e., x+Q) processed in the same period. cur,i (-y≤0), the service does not need to queue, and the queuing delay D queue =0. The corresponding probability is:

[0092]

[0093] In the formula, λ arr,i Indicates the business flow f on node i k Average arrival rate, f X ( ) is the probability mass function of the Poisson distribution, f Y ( ) is the binomial distribution probability mass function.

[0094] Queuing is required and sufficient buffer is available: if 0 < x + Q cur,i -y≤Q max -B k That is, the remaining space in the queue can accommodate business f k Then business f k The queuing time depends on the amount of business to be processed and the node's processing speed, requiring the queue to wait for existing business to complete. The corresponding probability is:

[0095]

[0096] Buffer overflow: If x + Q cur,i -y> Q max -B k If an overflow or packet loss occurs, it is considered a scheduling failure, and the queuing delay D... queue =∞. The corresponding probability is P3 = 1 - P1 - P2.

[0097] In summary, by combining the queue information collected in real time, the current queuing delay can be estimated.

[0098] S32. Define the link delay jitter J eff :

[0099]

[0100] In the formula, Var(D eff ) represents the time delay variance, and η represents the weighting coefficient.

[0101] Specifically, this invention comprehensively considers link delay and link connectivity probability to model link jitter behavior. Delay jitter J effUsed to characterize the degree of latency fluctuation. In this embodiment, latency jitter originates not only from changes in queue length, but more importantly from the dynamic fluctuation of latency in the time domain. In the above formula, the first term corresponds to the latency variance, and the second term is a penalty term for the rate of change of the survival probability. If the prediction result indicates that the link has a significant connection / disconnection switching trend in the future (i.e., If the absolute value of the derivative is large, the jitter of the link will increase significantly. Among these, the delay variance Var(D) eff This mainly describes the fluctuation range of latency, calculated statistically through a sliding window:

[0102]

[0103] Where N is the number of samples within the sliding window, and D eff (t) represents the effective transmission delay of the t-th sample. This represents the average effective transmission delay within the window.

[0104] S33. Define the overall packet loss rate L total :

[0105]

[0106] In the formula, the link connectivity probability is used to represent the topological uncertainty packet loss term, quantifying the packet loss due to disconnection caused by movement exceeding the communication range; L phy This represents the physical layer bit error and packet loss items, quantifying packet loss caused by decoding errors due to low signal strength or other reasons; L queue This indicates a queued packet loss item, meaning the number of incoming requests exceeds the cache space Q. max The probability is derived from the cumulative probabilities of the Poisson and binomial distributions:

[0107]

[0108] In the formula, N arr N represents the volume of business received. proc This indicates the volume of business processed.

[0109] By incorporating the link connectivity probability as a product factor into the packet loss rate calculation, even with excellent channel quality (L... phy (Low) If trajectory prediction shows insufficient link stability, the system can still ensure that the total packet loss rate tends to 1, thereby achieving strict constraints on communication reliability in dynamic task scenarios.

[0110] S4. Construct a probabilistically weighted node-link directed graph model: Assign dynamic attribute weight vectors to each node based on a unified state space, and assign probabilistic index evaluation vectors to each edge based on link connectivity probability, effective transmission delay, delay jitter, and overall packet loss rate, to obtain a graph model that represents the real-time dynamic evolution trend of the UAV swarm.

[0111] In some embodiments, step S4 includes:

[0112] S41. Abstract each drone in the drone swarm as a vertex, and construct a graph model G=(V,E,W), where V represents the vertex set, E represents the edge set, and W represents the network attribute weight set. W node W edge These represent the dynamic attribute weight set and the probabilistic index evaluation vector set, respectively.

[0113] For vertex v i ∈V, assign dynamic attribute weight vector , Represents vertex v i The acceleration variance within a preset historical sliding window is used here to characterize the dynamic risk attribute; this attribute informs the upper-layer protocol whether the vertex is currently in a state of violent maneuvering, thereby helping to determine its stability as a forwarding hop point. i Represents vertex v i Business processing capability characteristics; E res Indicates the status of remaining resources.

[0114] Set the link reliability threshold P th When the calculation is obtained When the link is in the future, it is determined that the link has a valid link connectivity probability at time τ. Based on the potential energy correlation matrix, a directed edge E is established for vertex pairs that satisfy the potential energy correlation threshold and have a valid link connectivity probability. ij Assign probabilistic index evaluation vectors P link (i,j,τ) represents the probability that the link between node i and node j will remain connected at time τ in the future, D eff J represents the effective transmission delay. eff Indicates latency jitter, L total This represents the overall packet loss rate. Even if two links have the same physical delay, the more unstable link, as shown in the trajectory prediction, will be assigned a higher effective delay weight in the graph.

[0115] In summary, by integrating the multidimensional probabilistic attributes of nodes and links, a network view capable of dynamically evolving with the task progress is constructed. This graph model not only represents the network connection state at the current moment but also maintains topology snapshots for multiple future prediction steps in parallel through a time sliding window mechanism. In each update cycle, the system dynamically adjusts the evaluation vectors of each edge in the graph based on the probability cloud drift trend output by S2 and the link connectivity probability fluctuations calculated by S3, and uses a multi-step smoothing algorithm to eliminate pseudo-stable links caused by instantaneous maneuvers. The resulting probabilistic weighted graph model serves as the core topology foundation, providing risk-aware network state information for upper-layer protocols such as routing planning and resource scheduling. By using link connectivity probability as a priori constraint for routing criteria, upper-layer decision-making algorithms can find the globally optimal and long-term stable transmission path in the multidimensional Quality of Service (QoS) evaluation vector, based on the differentiated needs of services for determinism (such as real-time control flow prioritizing highly reliable paths), within a weight matrix that incorporates spatial uncertainty risk, thereby achieving deep coupling between physical motion laws and logical network topology.

[0116] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A network modeling method for reliable communication in unmanned aerial vehicle (UAV) swarms, characterized in that, Includes the following steps: S1. Construct a unified state space and potential energy correlation matrix: Treat each UAV in the UAV cluster as an independent node, establish a unified state space that includes its dynamic characteristics and resource carrying capacity, and construct a potential energy correlation matrix between neighboring nodes based on relative motion stability. S2. Constructing the trajectory probability distribution: In a unified state space, the motion behavior of nodes within the prediction time window is probabilistically modeled to predict the trajectory probability distribution of the acceleration vector of each node, thereby obtaining the position probability density function of each future time. S3. Construct a multi-dimensional link quality quantification model with link connectivity probability as the core constraint: Calculate the link connectivity probability based on the location probability density function and potential energy correlation matrix, and use the link connectivity probability as a reliability constraint factor to perform weighted modeling of the effective transmission delay, delay jitter and overall packet loss rate of the link. S4. Construct a probabilistically weighted node-link directed graph model: Assign dynamic attribute weight vectors to each node based on a unified state space, and assign probabilistic index evaluation vectors to each edge based on link connectivity probability, effective transmission delay, delay jitter, and overall packet loss rate, to obtain a graph model that represents the real-time dynamic evolution trend of the UAV swarm.

2. The network modeling method for reliable communication of UAV swarms according to claim 1, characterized in that, Step S1 specifically includes: S11. Construct a unified state space S for UAV swarm nodes, where the state space representation of node i is S. i for: In the formula, pos i v represents the instantaneous position coordinates of node i. i Represents the velocity vector of node i. This represents the average acceleration of node i within a preset historical sliding window. C represents the variance of the acceleration of node i within a preset historical sliding window. i This represents the business processing capability characteristics of node i; S12. Construct the potential energy correlation matrix M for UAV swarm nodes, where m ij This represents the degree of association between node i and node j. The specific construction process includes: S121. Calculate the distance potential energy term E between node i and node j. dist (i,j): In the formula, d ij σ represents the Euclidean distance between node i and node j; dist This represents the distance sensitivity coefficient, used to adjust the rate of change of potential energy as distance increases; S122. Calculate the motion stability factor E between node i and node j. move (i,j): In the formula, Sim dir (i,j) represents the cosine similarity of the velocity directions between node i and node j, v i v j Let represent the velocity vectors of nodes i and j, λ represent the relative velocity penalty coefficient, and ||·|| represent the vector norm; S123. By fusing the distance potential energy term and the motion stability factor, the correlation density m between node i and node j is obtained. ij : In the formula, α represents the balancing weight coefficient.

3. The network modeling method for reliable communication of UAV swarms according to claim 1, characterized in that, Assuming the acceleration vector of each node All follow a Gaussian distribution , where the mean vector With covariance matrix All of these are the current velocity vectors of this node. The function is used to predict the trajectory probability distribution of each node using a deep neural network model that includes long short-term memory networks and hybrid density networks. The specific process includes: S21. Extract node i within the historical time window T obs The state-space representation within the network is used to construct the network input X. i,t ,in: In the formula, Let represent the instantaneous position coordinates of node i at time tk. This represents the velocity vector of node i at time tk. This represents the average acceleration of node i at time tk. This represents the variance of the acceleration of node i at time tk; S22. Use a Long Short-Term Memory (LSTM) network to process the network input X. i,t Extract the deep hidden state vector h that represents the motion trend. t ; S23. Transfer the deep hidden state vector h t Given a hybrid density network, predict the acceleration distribution parameter set of node i at the next K time steps, where the acceleration distribution parameter of node i at future time τ is... , This represents the expected acceleration of node i at a future time τ. This represents the degree of uncertainty of node i at a future time τ. This represents the correlation coefficient between the components of node i at a future time τ; S24. Based on the acceleration distribution parameter set and combined with the UAV kinematic equations, derive the trajectory probability distribution of node i in the next K time steps.

4. The network modeling method for reliable communication of UAV swarms according to claim 3, characterized in that, Step S24: Based on the acceleration distribution parameters of node i at future time τ Calculate the probability density function of the position of node i at future time τ, specifically including: S241. Based on the expected acceleration Calculate the probabilistic cloud center trajectory of node i ; S242. Based on the degree of uncertainty Construct the covariance matrix that diffuses over time ; S243. Based on the probability cloud center trajectory Covariance Matrix Calculate the probability density function f of the position of node i at a future time τ. i (x i ,τ), is represented as: In the formula, x i This represents the spatial coordinates of node i. Indicates transpose. This represents finding the inverse of a matrix. Represents a determinant.

5. The network modeling method for reliable communication of UAV swarms according to claim 1, characterized in that, If the correlation between nodes i and j is strong If the link connectivity probability is calculated based on the location probability density function, then the link connectivity probability is directly determined to be unavailable; the link connectivity probability calculation formula is expressed as: In the formula, f represents the spatial integration variable. i (x i Let I(·) denote the probability density function of the position of node i at future time τ, and let R be the indicator function. th P represents the effective communication radius. link (i,j,τ) represents the probability that the link between node i and node j will remain connected at the predicted time τ in the future, where m th This indicates a preset threshold.

6. The network modeling method for reliable communication of UAV swarms according to claim 5, characterized in that, Using link connectivity probability as a reliability constraint factor, a weighted model is performed on the effective transmission delay, delay jitter, and overall packet loss rate of the link, including: S31. Define the effective transmission delay D of the link. eff : In the formula, E[D prop ] represents the probabilistic expectation of the dynamic propagation delay of the link, D trans D represents the transmission delay of the link. proc E[D] represents the processing latency of the link. queue ] represents the expected probability of queuing delay in the link; S32. Define the link delay jitter J eff : In the formula, Var(D eff ) represents the time delay variance, and η represents the weighting coefficient; S33. Define the overall packet loss rate L total : In the formula, L phy L represents the physical layer bit error and packet loss item. queue This indicates the item for lost packages in the queue.

7. The network modeling method for reliable communication of UAV swarms according to claim 1, characterized in that, Step S4 includes: S41. Abstract each drone in the drone swarm as a vertex, and construct a graph model G=(V,E,W), where V represents the set of drone vertices, E represents the set of communication link edges, and W represents the set of network attribute weights. W node W edge These represent the dynamic attribute weight set and the probabilistic index evaluation vector set, respectively. For vertex v i ∈V, assign a weight vector to the node attributes. , Represents vertex v i The acceleration variance within a preset historical sliding window is used here to characterize the dynamic risk attribute; C i Represents vertex v i The characteristics of its business processing capabilities; Indicates the status of remaining resources; Set the link reliability threshold P th When the calculation is obtained When the link is in the future, it is determined that the link has a valid link connectivity probability at time τ. Based on the potential energy correlation matrix, a directed edge E is established for vertex pairs that satisfy the potential energy correlation threshold and have a valid link connectivity probability. ij Assign probabilistic index evaluation vectors P link (i,j,τ) represents the probability that the link between node i and node j will remain connected at time τ in the future, D eff J represents the effective transmission delay. eff Indicates latency jitter, L total This represents the overall packet loss rate.

Images