A method for dynamically switching the topology of a UAV cluster for urban multi-obstacle environments

Abstract

The present application belongs to the technical field of micro-grid energy management and unmanned aerial vehicle intelligent inspection, and discloses a kind of unmanned aerial vehicle cluster topology dynamic switching control method for urban multi-obstacle environment. With link availability probability as the unified dimension, based on equivalent additional loss, fusion of shelter, multipath, LOS / NLOS and infrastructure availability, a discrete topology state space is constructed. A semi-Markov jump model is used to describe the residence time and transition law, to predict the jump probability and remaining residence time in a finite time domain. Under the guidance of prediction, candidate topologies, attitude fine-tuning / link activation strategies are generated, and minimum residence, hysteresis mechanism and switching penalty constraints are applied. The optimal topology is selected by distributed consistency optimization, and after execution, it is re-optimized by feedback rolling. This method reduces the risk of frequent topology switching and disconnection, and improves connectivity and end-to-end delay controllability.

Description

Technical Field

[0001] This invention belongs to the field of UAV swarm control and air-to-ground integrated wireless communication technology, focusing on methods for dynamic switching, rapid reconstruction, and emergency response of UAV swarm communication topology in complex environments with multiple obstacles, such as urban / suburban areas. The urban multi-obstacle environment includes at least: the street valley effect formed by high-density building clusters, frequent non-line-of-sight (NLOS) / line-of-sight (LOS) switching, strong multipath and occlusion caused by metal facades, transient link interruptions caused by dynamic obstacles such as pedestrians / vehicles, and fluctuations in communication infrastructure availability with changes in region / time period. Background Technology

[0002] Unmanned aerial vehicle (UAV) swarms have significant application value in urban low-altitude operations such as inspection, emergency response, and security. In urban / suburban multi-obstacle environments, swarm communication links often exhibit the following characteristics: (1) rapid LOS / NLOS switching caused by static obstacles such as buildings, bridges, and trees; (2) strong multipath and obstruction in street valley environments causing drastic fluctuations in link quality over short time scales; (3) sudden interruptions caused by dynamic obstacles such as vehicles, hoisting equipment, and temporary blockades; and (4) differences in coverage, penetration, capacity, and latency between different communication standards, leading to the need for on-demand switching of multi-mode links. These factors make "topology switching" a key capability for collaboration in urban UAV swarms: without modeling of topology dwell time and switching costs, problems such as topology multi-hop, local disconnection, mission interruption, and increased security risks are likely to occur.

[0003] Existing technologies for topology control of UAV swarms can be broadly categorized into three types. The first type consists of local or deterministic graph theory methods, such as degree centrality, k-connected graphs, and minimum spanning trees. While these methods are computationally fast (approaching linear time) and easy to implement in a distributed manner, they lack a global perspective and model stochastic processes, thus only being able to handle ideal open or weakly occluded environments. In urban environments with multiple obstacles, they are not robust enough to random occlusion, sudden interruptions, and non-stationary transitions, and the lag in topology switching can easily lead to temporary isolated nodes or swarm splits. The second category consists of stochastic transition models represented by continuous-time Markov chains (CTMC) or discrete Markov chains. Although these models can describe the topological state transition probabilities, they generally assume a constant transition rate and an exponential distribution of dwell time. This makes it difficult to characterize the non-exponential dwell characteristics that are common in urban / suburban environments. For example, complex shading (street valleys, overpasses / elevated roads, dense building clusters, etc.) leads to a heavy-tailed distribution of shading duration, extreme weather attenuation duration is affected by wind speed / rain intensity and follows a Weibull / gamma distribution, and multipath reflector clusters have short but highly random dwell times. As a result, the models have large prediction biases and control strategies that are too conservative or aggressive, making it difficult to achieve accurate forward-looking decisions. The third category is learning-based methods that have emerged in recent years, such as trajectory-communication joint optimization based on reinforcement learning or deep neural networks. Although they have made progress in some general scenarios, they have high computational complexity (requiring a lot of Monte Carlo simulation or online training), poor real-time performance, and rely heavily on simplified channel assumptions or expensive labels (such as calculating the shortest path in the entire network). They lack interpretability and stable generalization ability, and are difficult to directly adapt to unseen scenarios and multi-mode heterogeneous links.

[0004] In addition, existing methods are mostly designed for open or weakly obstructed environments, lacking a unified modeling and handling framework for the core mechanism of "frequent topology switching caused by multiple obstacles in the city". In particular, they lack event-triggered rapid rollback for sudden obstruction / infrastructure fluctuations, as well as stability-oriented switching suppression mechanisms (such as minimum dwell time, hysteresis threshold and switching penalty). As a result, topology oscillation and service quality degradation are prone to occur in multi-service concurrency and high-reliability tasks. Summary of the Invention

[0005] The core technical problem this invention aims to solve is that in complex urban environments with multiple obstacles, the combined effects of static occlusions such as buildings / vegetation and dynamic obstacles such as vehicles and crowds cause frequent line-of-sight / non-line-of-sight (LOS / NLOS) switching, short-term and drastic fluctuations in link quality due to multipath fading and occlusion, and significant changes in link availability over time and space due to differences in coverage and penetration capabilities of multiple communication standards. As a result, the communication topology of UAV swarms experiences frequent switching, partial disconnection, and degraded control performance.

[0006] To address the above problems, this invention proposes a dynamic topology switching control method for UAV swarms in urban multi-obstacle environments, comprising the following steps:

[0007] S1. Real-time monitoring of the multi-mode communication topology of drone swarms, extraction of link status between nodes and characteristics of urban multi-obstacle environment, and construction of discrete topology state space;

[0008] S2. Construct a semi-Markov transition model based on historical and real-time data to estimate the residence time distribution and state transition patterns of each topological state;

[0009] S3. Use a semi-Markov model to predict the probability of topological jumps and the expected remaining dwell time in the future finite time domain;

[0010] S4. Guided by the prediction results, generate a set of candidate connected graphs and construct a multi-mode switching / topology maintenance strategy;

[0011] S5. Select the optimal target topology through distributed consensus optimization, and output attitude adjustment and multi-mode link activation commands for each UAV;

[0012] S6. Perform topology reconstruction and feedback monitoring to achieve closed-loop dynamic control and online re-optimization.

[0013] Furthermore, step S1 includes the following sub-steps:

[0014] S1.1 Status Acquisition;

[0015] During each control cycle, each UAV collects and reports link measurements related to neighboring nodes, including received signal strength. Signal-to-noise ratio Packet loss rate LOS probability estimation and infrastructure availability factor Simultaneously, it collects urban multi-obstacle environment characteristics, including street valley indicators, occlusion intensity, and dynamic obstacle alarms;

[0016] S1.2 Link Comprehensive Quality Calculation and Availability Probability Assessment;

[0017] I. Equivalent Additional Losses in Link Construction And calculate the effective signal-to-noise ratio;

[0018] In the formula, The effective signal-to-noise ratio between nodes u and v at time t; This is the baseline path loss; Distance between nodes; Low noise level;

[0019] II. Define the link availability probability;

[0020]

[0021] In the formula, Let be the availability probability of the link between nodes u and v at time t, with a value range of [0,1]. The closer it is to 1, the more stable and available the link is. This represents the Sigmoid function. The difference between the effective signal-to-noise ratio and the threshold is mapped to the (0,1) interval, reflecting the probability that the signal-to-noise ratio meets the requirements; Indicates the signal-to-noise ratio threshold;

[0022] S1.3 Topology construction, hysteresis edge detection and discrete state space generation;

[0023] Use entry threshold With exit threshold Hysteresis mechanism for updating edge sets ,in, , For a moment The effective set of multimode links;

[0024] when Link establishment is permitted when Disconnect the link at one time, while keeping the rest unchanged, thus obtaining the topology map. The topology graph is clustered according to the number of connected components, network diameter, and algebraic connectivity to form a finite discrete set of topological states. Each state corresponds to a unique connectivity structure.

[0025] Furthermore, in step S1, the link equivalent additional loss The calculation expression is:

[0026]

[0027] In the formula, Indicates additional losses due to obstruction / diffraction; Indicates additional multipath loss; This indicates additional losses due to weather conditions; This represents the equivalent additional loss due to electromagnetic interference.

[0028] Furthermore, the semi-Markov transition model is constructed based on the discrete topological state space of step S1, and the residence time follows a Weibull or Gamma distribution.

[0029] The modeling process of the semi-Markov jump model is as follows:

[0030] The discrete topological state process is denoted as... When the system enters a state After that, its residence time random variable Follows distribution ,density Survival function ;

[0031] The corresponding transition probability density is defined as:

[0032]

[0033] In the formula, Let be the transition probability density, representing the system in state . Stay When time is specified, the state will jump to a specific state within a unit of time. The probability of; The transition probability is a residence time dependent probability, representing the system's state transition probability. Stayed After a certain time, jump to the status. The probability of; For state The probability density function of the dwell time;

[0034] The state transition kernel is defined in a residence-time-dependent form as follows:

[0035]

[0036] The probability density function of the Weibull distribution is:

[0037]

[0038] In the formula, For shape parameters, Controlling the distribution pattern: It degenerates into an exponential distribution, corresponding to the memorylessness of traditional Markov processes; The occurrence of heavy-tailed characteristics indicates a high probability of "long-term dwell" events, which is suitable for weakly connected / non-line-of-sight states. It exhibits a single-peak distribution and a typical peak in dwell time, making it suitable for scenarios where link status recovers rapidly. The scale parameter of the Weibull distribution is used to represent the scale parameter of the Weibull distribution.

[0039] The survival function is:

[0040]

[0041] The probability density function of the alternative gamma distribution is:

[0042]

[0043] In the formula, is the scale parameter of the gamma distribution; It is a gamma function;

[0044] For the case of no transition, the state preservation probability is expressed by the survival function. Description; Model parameters , / and transfer of nuclear Estimation is performed using the Expectation-Maximization (EM) algorithm.

[0045] Furthermore, step S3 includes the following sub-steps:

[0046] S3.1 Predict the probability of topological jumps within a finite time domain in the future;

[0047]

[0048] In the formula, Kronek The function takes the value 1 when i=j, i The value of j is 0; Let be a conditional survival function, representing the condition that a resident has already lived in the area. Given time, the system is in state The probability of remaining in the medium for more than time t; For conditional jump density, it indicates that the density has already been settled. Given time, the system is in state Staying in At time +t, the state jumps to the next state per unit time. The probability of;

[0049] S3.2 Predict the expected remaining dwell time within a finite time domain in the future;

[0050]

[0051] In the formula, This indicates that the current state is Already stationed Under the premise that the system can remain in that state for an average period of time before leaving it.

[0052] Furthermore, the specific content of step S4 is as follows:

[0053] S4.1 Generate a set of candidate connected graphs based on the prediction results and a finite set of actions. ;

[0054] Candidate actions in the action set include pose fine-tuning and multimodal link activation / suppression;

[0055] S4.2 Construct multi-mode handover / topology maintenance strategies;

[0056] In both candidate generation and screening, the following constraints are applied: (1) minimum residence time constraint, i.e., if the minimum residence time is not reached... Reverse switching is not allowed; (2) Hysteresis threshold constraint is implemented through step S3; (3) Switching penalty item is used to reduce the risk of frequent topology switching and service interruption.

[0057] Furthermore, the specific content of step S5 is as follows:

[0058] For each candidate topology Construct a comprehensive objective, and define the optimization objective as follows:

[0059]

[0060] In the formula, , , These are the weighting coefficients; For topology Laplacian matrix The second smallest eigenvalue; The expected algebraic connectivity is a weighted average of all possible future states; Indicates the cost of topology switching; This refers to the expected end-to-end latency metric.

[0061]

[0062] In the formula, , , These are the weighting coefficients for each cost; For drones The height change of the attitude adjustment amount; For drones The change in yaw angle of attitude adjustment; Indicates the yaw energy consumption coefficient; Multimode switching latency represents the time required to switch communication modes. , The link uv weights are the values ​​of the link uv in the old and new topologies.

[0063] The optimal target topology is selected as follows:

[0064]

[0065] In the formula, The optimal topology is found in the candidate set. The middle can enable the comprehensive objective function Topology that reaches its maximum value; Represents the set of candidate connected graphs; Represent the overall objective function;

[0066] A consensus-gradient distributed iterative solution is used: each node maintains local variables. It also exchanges local information with its one-hop neighbors and iteratively updates the information:

[0067]

[0068] In the formula, For drones The local variables at the k-th iteration include attitude adjustment and link decision; This is the iteration step size; For drones The set of one-hop neighbors, that is, with Direct communication drone nodes; For drones The gradient of the local objective function at the k-th iteration represents the direction of change of the local objective function;

[0069] This protocol does not require a global central processing unit; after convergence, all drones consistently obtain [the necessary information / data]. And generate specific instructions.

[0070] Furthermore, the specific content of step S6 is as follows:

[0071] After executing the command, continuous monitoring is performed. The system calculates the number of connected components and end-to-end time delay. When the deviation between the prediction and the actual result exceeds a threshold, it triggers an online update of the semi-Markov model parameters and threshold weights, and returns to steps S5–S6 for rolling re-optimization, thereby achieving adaptive closed-loop control of urban multi-obstacle environment disturbances.

[0072] Furthermore, the present invention also provides a dynamic control system for the topology of an unmanned aerial vehicle (UAV) swarm, including modules for performing the above-described methods: a topology monitoring and state construction module, a semi-Markov model construction and prediction module, a candidate topology generation and optimization module, a distributed decision-making and command output module, and an execution and closed-loop feedback module.

[0073] Beneficial effects:

[0074] This invention proposes a dynamic control method for multi-mode communication topology of UAV swarms based on a semi-Markov transition process. The method models and predicts the residence time distribution and transition law of the topology state through a semi-Markov transition process. Guided by the prediction results, it jointly considers connectivity, end-to-end latency and handover cost to generate topology handover decisions. Combined with distributed consistency optimization and closed-loop feedback of "state reporting - command issuance", it realizes online correction and re-optimization of the topology to improve the stability of topology handover and task continuity in urban multi-obstacle environments.

[0075] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0076] Figure 1 This is a flowchart of a dynamic switching control method for UAV swarm topology in urban multi-obstacle environments according to the present invention;

[0077] Figure 2 This is a schematic diagram of typical topology switching and emergency rollback in a multi-obstacle urban environment in this embodiment. Detailed Implementation

[0078] To make the technical solutions, advantages, and objectives of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the protection scope of this application.

[0079] To address the issues of frequent topology switching, partial disconnections, and latency degradation that often occur in UAV swarms operating in urban environments with multiple obstacles, especially under conditions of drastic fluctuations in multi-mode communication link quality, frequent LOS / NLOS transitions, and changes in infrastructure availability, this invention proposes a dynamic topology switching and online re-optimization method based on a semi-Markov transition process. This method uses link availability probability as a unified metric, abstracts topology evolution into finite discrete states, and employs a semi-Markov process to characterize dwell time and transition patterns. Candidate topologies are generated under prediction guidance, and the optimal topology is selected through distributed consensus optimization. Finally, rolling re-optimization is achieved through state feedback. The complete execution flow of the method is as follows: Figure 1 As shown, Figure 2 The topology switching and emergency fallback process is shown in the case of link drop triggered by street valley occlusion and dynamic obstacles.

[0080] This invention provides a method for dynamic switching control of UAV swarm topology in urban multi-obstacle environments, comprising the following steps:

[0081] S1. Real-time monitoring of the multi-mode communication topology of drone swarms, extraction of link status between nodes and characteristics of urban multi-obstacle environment, and construction of discrete topology state space;

[0082] S1.1 Status Acquisition;

[0083] Establishing a cluster communication graph during system initialization ,in For a collection of drone nodes, For a moment The effective multimode link set. During each control cycle, each UAV collects and reports link measurements related to neighboring nodes, including received signal strength. Signal-to-noise ratio Packet loss rate LOS probability estimation And infrastructure availability factors such as public networks / relays Simultaneously, urban multi-obstacle environment characteristics (such as street valley index, occlusion intensity, dynamic obstacle alarm, etc.) are collected as inputs for subsequent disturbance term parameterization.

[0084] S1.2 Link Comprehensive Quality Calculation and Availability Probability Assessment;

[0085] I. To uniformly quantify the impact of factors such as occlusion, multipath, and standard differences on the link, construct the equivalent additional loss of the link. And calculate the effective signal-to-noise ratio;

[0086]

[0087] In the formula, The effective signal-to-noise ratio between nodes u and v at time t; This is the baseline path loss; Distance between nodes; Low noise level;

[0088] II. Define the link availability probability;

[0089]

[0090] In the formula, Let be the availability probability of the link between nodes u and v at time t, with a value range of [0,1]. The closer it is to 1, the more stable and available the link is. This represents the Sigmoid function. The difference between the effective signal-to-noise ratio and the threshold is mapped to the (0,1) interval, reflecting the probability that the signal-to-noise ratio meets the requirements; Indicates the signal-to-noise ratio threshold;

[0091] S1.3 Topology construction, hysteresis edge detection and discrete state space generation;

[0092] Assuming an entry threshold With exit threshold satisfy (Hysteresis suppression), generate edge sets according to the following rules;

[0093] (1) If the current and Then ;

[0094] (2) If the current and Then from delete;

[0095] (3) Otherwise, remain unchanged.

[0096] Will be The induced topology graph is calculated based on the number of connected components. Average shortest path hop count Clustering or partitioning based on indicators such as key node coverage constraints yields a finite discrete set of topological states. Each state corresponds to a typical connectivity structure (such as fully connected, ring relay, chain relay, clustering / local splitting, etc.), thus forming a discrete topological state space.

[0097] Link equivalent additional loss It is parameterized by factors such as shading / street valley effect / multipath scattering / meteorology (see example below). To provide a feasible method for constructing the perturbation term, we can... Represented as the sum of several addendable terms:

[0098]

[0099] In the formula, Indicates additional losses due to obstruction / diffraction; Indicates additional multipath loss; This indicates additional losses due to weather conditions; This represents the equivalent additional loss due to electromagnetic interference.

[0100] The following are "exemplary" parameterizations for four extreme scenarios:

[0101] Example 1: Strong electromagnetic interference (corona pulse, etc.) near power transmission lines

[0102]

[0103] in The pulse interference coefficient is... It is the product of pulse arrival rate and intensity. For voltage level, Distance from tower / conductor This is the distance decay index.

[0104] Example 2: Multiple obstacles (building / vegetation penetration and diffraction, etc.) in suburban / residential / forested areas during power distribution line inspections can be characterized using simplified models of obstruction intensity or non-line-of-sight probability, for example:

[0105]

[0106] in The obstacle density coefficient (which can be determined by a combination of factors such as building density, street-to-valley width-to-height ratio, and vegetation density). Effective shading distance (which can be estimated from a 3D city model / terrain height map). This represents the flight altitude. This example can also be used for general urban street canyon occlusion scenarios. Parameterization.

[0107] Example 3: Extreme weather and infrastructure damage in power emergency communications

[0108]

[0109] in For meteorological attenuation parameters, Rainfall intensity, Wind speed; damaged infrastructure can be accessed via Directly depict, and in This enables rapid correction.

[0110] Example 4: Strong multipath / reflection clusters caused by dense metal equipment in substations

[0111]

[0112] in The reflection / cluster loss coefficient, For multipath cluster number, This is the attitude jitter / microDoppler correlation statistic (used to characterize short-time multipath instability).

[0113] S2. Construct a semi-Markov transition model based on historical and real-time data to estimate the residence time distribution and state transition patterns of each topological state;

[0114] Construction of a semi-Markov jump model and fitting of dwell time. The evolution of discrete topological states is modeled as a semi-Markov process: when the system is in state... At that time, their length of stay Follows distribution Survival function Dwell time can be fitted using a Weibull or Gamma distribution, and the transition patterns related to dwell time can be estimated through historical state sequences. To adapt to the non-stationarity of urban environments with multiple obstacles, a sliding window is used to update parameters online, making the model sensitive to changes in shading intensity, dynamic obstacle frequency, and infrastructure availability.

[0115] The discrete topological state process is denoted as... When the system enters a state After that, its residence time random variable Follows distribution ,density Survival function ;

[0116] The corresponding transition probability density is defined as:

[0117]

[0118] In the formula, Let be the transition probability density, representing the system in state . Stay When time is specified, the state will jump to a specific state within a unit of time. The probability of; The transition probability is a residence time dependent probability, representing the system's state transition probability. Stayed After a certain time, jump to the status. The probability of; For state The probability density function of the dwell time;

[0119] The state transition kernel is defined in a residence-time-dependent form as follows:

[0120]

[0121] The probability density function of the Weibull distribution is:

[0122]

[0123] In the formula, For shape parameters, Controlling the distribution pattern: It degenerates into an exponential distribution, corresponding to the memorylessness of traditional Markov processes; The heavy-tailed characteristic indicates a high probability of "long-term dwell" events, and is suitable for long-term weak connectivity or non-line-of-sight states caused by continuous occlusion or insufficient coverage in urban multi-obstacle environments. The occurrence of a pattern of first increasing and then decreasing indicates that there is a more typical peak scale in the dwell time, which is applicable to the "phased stability - phased switching" process caused by periodic or structural factors (such as the state recovery brought about by passing through the street valley section and turning out of the street corner). The scale parameter of the Weibull distribution; scale parameter (or The average dwell time can be controlled and adjusted online based on real-time environment and link quality statistics to reflect the differences in topology stability in different regions and time periods.

[0124] The survival function is:

[0125]

[0126] The probability density function of the alternative gamma distribution is:

[0127]

[0128] In the formula, is the scale parameter of the gamma distribution; It is a gamma function;

[0129] For the case of no transition, the state preservation probability is determined by the survival function. Description. Model parameters ( , / and transfer of nuclear Estimate using the Expectation-Maximization (EM) algorithm: Step S1 calculates the posterior of latent variables (based on the historical jump sequence and the link availability probability of step S1). (weighted) Maximize likelihood step:

[0130]

[0131] in, For parameter set, This serves as a historical jump index. The algorithm supports using the average link availability from step S1. Incorporating as a covariate (e.g., low) Time bias towards heavy tail To enhance the model's sensitivity to power disturbances, Bayesian updates or sliding window variants (with a typical window length of 10-30 jump events) are used in the online phase to achieve adaptive parameter correction, ensuring the model responds quickly to real-time scene changes (such as increased electromagnetic interference due to rising load rates).

[0132] S3. Use a semi-Markov model to predict the probability of topological jumps and the expected remaining dwell time in the future finite time domain;

[0133] S3.1 Predict the probability of topological jumps within a finite time domain in the future;

[0134] Given the current topology state and length of stay Define conditional survival function and conditional jump density Then in the future, within a finite time domain... The internal system is in a state The probability can be written as:

[0135]

[0136] To support multi-step transitions, the update equation can be written as follows:

[0137]

[0138] In the formula, Kronek The function takes the value 1 when i=j, i The value of j is 0; Let be a conditional survival function, representing the condition that a resident has already lived in the area. Given time, the system is in state The probability of remaining in the medium for more than time t; For conditional jump density, it indicates that the density has already been settled. Given time, the system is in state Staying in At time +t, the state jumps to the next state per unit time. The probability of;

[0139] S3.2 Predict the expected remaining dwell time within a finite time domain in the future;

[0140]

[0141] In the formula, This indicates that the current state is Already stationed Under the premise that the system can remain in that state for an average period of time before leaving it.

[0142] S4. Guided by the prediction results, generate a set of candidate connected graphs and construct a multi-mode switching / topology maintenance strategy;

[0143] S4.1 Generate a set of candidate connected graphs based on the prediction results and a finite set of actions. ;

[0144] Based on the predicted state arrival probability in the time domain, the topology corresponding to the high-probability states is selected as the candidate set. The candidate actions consist of two categories: one is a small number of drone attitude fine-tuning (altitude, yaw, etc.) to avoid occlusion, and the other is multi-mode link activation / suppression to improve the availability of critical links.

[0145] S4.2 Construct multi-mode handover / topology maintenance strategies;

[0146] To ensure switching stability, the following constraints are applied simultaneously during candidate generation and screening: (1) Minimum dwell time constraint, i.e., if the minimum dwell time is not reached... Reverse switching is not allowed; (2) Hysteresis threshold constraint is implemented through step S3; (3) Switching penalty item is used to reduce the risk of frequent topology switching and service interruption.

[0147] S5. Select the optimal target topology through distributed consensus optimization, and output attitude adjustment and multi-mode link activation commands for each UAV;

[0148] For each candidate topology Construct a comprehensive objective, and define the optimization objective as follows:

[0149]

[0150] In the formula, , , These are the weighting coefficients; For topology Laplacian matrix The second smallest eigenvalue; The expected algebraic connectivity is a weighted average of all possible future states; Indicates the cost of topology switching; This refers to the expected end-to-end latency metric.

[0151]

[0152] In the formula, , , These are the weighting coefficients for each cost; For drones The height change of the attitude adjustment amount; For drones The change in yaw angle of attitude adjustment; Indicates the yaw energy consumption coefficient; Multimode switching latency represents the time required to switch communication modes. , The link UV values ​​are the weights under the old and new topologies; the first term is the energy consumption for attitude adjustment. The first term is the yaw energy consumption coefficient; the second term is the multi-mode switching delay (higher for satellites, lower for self-organizing networks); and the third term is the link weight change. , and The coefficients are determined through offline grid search after quantization;

[0153] The optimal target topology is selected as follows:

[0154]

[0155] In the formula, The optimal topology is found in the candidate set. The middle can enable the comprehensive objective function Topology that reaches its maximum value; Represents the set of candidate connected graphs; Represent the overall objective function;

[0156] A consensus-gradient distributed iterative solution is used: each node maintains local variables. (Attitude / Link Decision) and exchange local information with one-hop neighbors, iteratively updating:

[0157]

[0158] In the formula, For drones The local variables at the k-th iteration include attitude adjustment and link decision; This is the iteration step size; For drones The set of one-hop neighbors, that is, with Direct communication drone nodes; For drones The gradient of the local objective function at the k-th iteration represents the direction of change of the local objective function;

[0159] This protocol does not require a global central processing unit; after convergence, all drones consistently obtain [the necessary information / data]. And generate specific instructions.

[0160] S6. Perform topology reconstruction and feedback monitoring to achieve closed-loop dynamic control and online re-optimization.

[0161] After executing the command, continuous monitoring is performed. Feedback quantities include the number of connected components and end-to-end time delay; when the deviation between the prediction and the actual value exceeds the threshold, the online update of the semi-Markov model parameters and threshold weights is triggered, and the process returns to steps S5–S6 for rolling re-optimization, thereby achieving adaptive closed-loop control of urban multi-obstacle environment disturbances.

[0162] exist Figure 2 In the building occlusion and dynamic obstacle triggering scenario shown, some links of the cluster experience issues after entering buildings or other areas. The continuous decline causes the topology to change from a "fully connected" structure to a "backbone relay" structure. When dynamic obstacles cause a sudden drop in the availability of critical edges, the system avoids repeated switching under switching constraints and quickly switches to a structure with higher algebraic connectivity through candidate topologies, maintaining the connectivity of critical control links and controllable end-to-end latency. Subsequently, during the environmental recovery phase, the system returns to a more efficient normal topology through rolling re-optimization, demonstrating the stable switching and online adaptive capabilities of this invention in urban multi-obstacle environments.

[0163] The present invention also provides a dynamic control system for the topology of an unmanned aerial vehicle (UAV) swarm, comprising modules for performing the above-described methods: a topology monitoring and state construction module, a semi-Markov model construction and prediction module, a candidate topology generation and optimization module, a distributed decision-making and command output module, and an execution and closed-loop feedback module.

[0164] It is hereby declared that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for dynamic topology switching control of UAV swarms in urban multi-obstacle environments, characterized in that, Includes the following steps: S1. Real-time monitoring of the multi-mode communication topology of drone swarms, extraction of link status between nodes and characteristics of urban multi-obstacle environment, and construction of discrete topology state space; S2. Construct a semi-Markov transition model based on historical and real-time data to estimate the residence time distribution and state transition patterns of each topological state; S3. Use a semi-Markov model to predict the probability of topological jumps and the expected remaining dwell time in the future finite time domain; S4. Guided by the prediction results, generate a set of candidate connected graphs and construct a multi-mode switching / topology maintenance strategy; S5. Select the optimal target topology through distributed consensus optimization, and output attitude adjustment and multi-mode link activation commands for each UAV; S6. Perform topology reconstruction and feedback monitoring to achieve closed-loop dynamic control and online re-optimization.

2. The method for dynamic switching control of UAV swarm topology in urban multi-obstacle environments according to claim 1, characterized in that, Step S1 includes the following sub-steps: S1.1 Status Acquisition; During each control cycle, each UAV collects and reports link measurements related to neighboring nodes, including received signal strength. Signal-to-noise ratio Packet loss rate LOS probability estimation and infrastructure availability factor Simultaneously, it collects urban multi-obstacle environment characteristics, including street valley indicators, occlusion intensity, and dynamic obstacle alarms; S1.2 Link Comprehensive Quality Calculation and Availability Probability Assessment; I. Equivalent Additional Losses in Link Construction And calculate the effective signal-to-noise ratio; In the formula, The effective signal-to-noise ratio between nodes u and v at time t; This is the baseline path loss; Distance between nodes; Low noise level; II. Define the link availability probability; In the formula, Let be the availability probability of the link between nodes u and v at time t, with a value range of [0,1]. The closer it is to 1, the more stable and available the link is. This represents the Sigmoid function. The difference between the effective signal-to-noise ratio and the threshold is mapped to the (0,1) interval, reflecting the probability that the signal-to-noise ratio meets the requirements; Indicates the signal-to-noise ratio threshold; S1.3 Topology construction, hysteresis edge detection and discrete state space generation; Use entry threshold With exit threshold Hysteresis mechanism for updating edge sets ,in, , For a moment The effective set of multimode links; when Link establishment is permitted when Disconnect the link at one time, while keeping the rest unchanged, thus obtaining the topology diagram. The topology graph is clustered according to the number of connected components, network diameter, and algebraic connectivity to form a finite discrete set of topological states. Each state corresponds to a unique connectivity structure.

3. The method for dynamic switching control of UAV swarm topology in urban multi-obstacle environments according to claim 2, characterized in that, In step S1, the link equivalent additional loss The calculation expression is: In the formula, Indicates additional losses due to obstruction / diffraction; Indicates additional multipath loss; This indicates additional losses due to weather conditions; This represents the equivalent additional loss due to electromagnetic interference.

4. The method for dynamic switching control of UAV swarm topology in urban multi-obstacle environments according to claim 3, characterized in that: The semi-Markov transition model is constructed based on the discrete topological state space of step S1, and the residence time follows a Weibull or gamma distribution. The modeling process of the semi-Markov jump model is as follows: The discrete topological state process is denoted as... When the system enters a state After that, its residence time random variable Follows distribution ,density Survival function ; The corresponding transition probability density is defined as: In the formula, Let be the transition probability density, representing the system in state . Stay When time is specified, the state will jump to a specific state within a unit of time. The probability of; The transition probability is a residence time dependent probability, representing the system's state transition probability. Stayed After a certain time, jump to the status. The probability of; For state The probability density function of the dwell time; The state transition kernel is defined in a residence-time-dependent form as follows: The probability density function of the Weibull distribution is: In the formula, For shape parameters, Controlling the distribution pattern: It degenerates into an exponential distribution, corresponding to the memorylessness of traditional Markov processes; The occurrence of heavy-tailed characteristics indicates a high probability of "long-term dwell" events, which is suitable for weakly connected / non-line-of-sight states; It exhibits a single-peak distribution and a typical peak in dwell time, making it suitable for scenarios where link status recovers rapidly. The scale parameter of the Weibull distribution is used to represent the scale parameter of the Weibull distribution. The survival function is: The probability density function of the alternative gamma distribution is: In the formula, is the scale parameter of the gamma distribution; It is a gamma function; For the case of no transition, the state preservation probability is expressed by the survival function. Description; Model parameters , / and transfer of nuclear Estimation is performed using the Expectation-Maximization (EM) algorithm.

5. The method for dynamic switching control of UAV swarm topology in urban multi-obstacle environments according to claim 4, characterized in that, Step S3 includes the following sub-steps: S3.1 Predict the probability of topological jumps within a finite time domain in the future; In the formula, Kronek The function takes the value 1 when i=j, i The value of j is 0; Let be a conditional survival function, representing the condition that a resident has already lived in the area. Given time, the system is in state The probability of remaining in the medium for more than time t; For conditional jump density, it indicates that the density has already been settled. Given time, the system is in state Staying in At time +t, the state jumps to the next state per unit time. The probability of; S3.2 Predict the expected remaining dwell time within a finite time domain in the future; In the formula, This indicates that the current state is Already stationed Under the premise that the system can remain in that state for an average period of time before leaving it.

6. The method for dynamic switching control of UAV swarm topology in urban multi-obstacle environments according to claim 5, characterized in that, The specific content of step S4 is as follows: S4.1 Generate a set of candidate connected graphs based on the prediction results and a finite set of actions. ; Candidate actions in the action set include pose fine-tuning and multimodal link activation / suppression; S4.2 Construct multi-mode handover / topology maintenance strategies; In both candidate generation and screening, the following constraints are applied: (1) minimum residence time constraint, i.e., if the minimum residence time is not reached... Reverse switching is not allowed; (2) Hysteresis threshold constraint is implemented through step S3; (3) Switching penalty item is used to reduce the risk of frequent topology switching and service interruption.

7. The method for dynamic switching control of UAV swarm topology in urban multi-obstacle environments according to claim 6, characterized in that, The specific content of step S5 is as follows: For each candidate topology Construct a comprehensive objective, and define the optimization objective as follows: In the formula, , , These are the weighting coefficients; For topology Laplacian matrix The second smallest eigenvalue; The expected algebraic connectivity is a weighted average of all possible future states; Indicates the cost of topology switching; This is the expected end-to-end latency metric. In the formula, , , These are the weighting coefficients for each cost; For drones The height change of the attitude adjustment amount; For drones The change in yaw angle of attitude adjustment; Indicates the yaw energy consumption coefficient; Multimode switching latency represents the time required to switch communication modes. , The link uv weights are the values ​​of the link uv in the old and new topologies. The optimal target topology is selected as follows: In the formula, The optimal topology is found in the candidate set. The middle can enable the comprehensive objective function Topology that reaches its maximum value; Represents the set of candidate connected graphs; Represent the overall objective function; A consensus-gradient distributed iterative solution is used: each node maintains local variables. It also exchanges local information with its one-hop neighbors and iteratively updates the information: In the formula, For drones The local variables at the k-th iteration include attitude adjustment and link decision; This is the iteration step size; For drones The set of one-hop neighbors, that is, with Direct communication drone nodes; For drones The gradient of the local objective function at the k-th iteration represents the direction of change of the local objective function; This protocol does not require a global central processing unit; after convergence, all drones consistently obtain [the necessary information / data]. And generate specific instructions.

8. The method for dynamic switching control of UAV swarm topology in urban multi-obstacle environments according to claim 7, characterized in that, The specific content of step S6 is as follows: After executing the command, continuous monitoring is performed. The system calculates the number of connected components and end-to-end time delay. When the deviation between the prediction and the actual result exceeds a threshold, it triggers an online update of the semi-Markov model parameters and threshold weights, and returns to steps S5–S6 for rolling re-optimization, thereby achieving adaptive closed-loop control of urban multi-obstacle environment disturbances.

9. A dynamic control system for unmanned aerial vehicle (UAV) swarm topology, characterized in that: It includes modules for performing the method according to any one of claims 1 to 8: a topology monitoring and state construction module, a semi-Markov model construction and prediction module, a candidate topology generation and optimization module, a distributed decision-making and instruction output module, and an execution and closed-loop feedback module.

Images