A Method and System for Optimizing Unmanned Aerial Vehicle Communication Networks Based on Digital Twins

Abstract

A method and system for optimizing UAV communication networks based on digital twins are disclosed, relating to the field of UAV communication network optimization technology. The method includes acquiring measured communication state data of the UAV communication network in time slices; generating a priori twin state using initial simulation parameters in the initial time slice, and generating a priori twin state in subsequent time slices by reading the simulation parameters and policy boundaries fed back from the previous time slice; performing state assimilation correction based on the virtual-real synchronization deviation between the measured state and the priori twin state, and performing joint simulation, bottleneck link identification, and policy pre-simulation based on the corrected twin state; after the target policy is executed, generating the virtual-real deviation of policy execution based on the measured communication quality change and the pre-simulation saved results, and feeding back and updating the simulation parameters and policy boundaries for the next time slice. This invention improves the real-time performance, reliability, and closed-loop self-correction capability of communication optimization.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) communication network optimization technology, and in particular to a method and system for optimizing UAV communication networks based on digital twins. Background Technology

[0002] Unmanned aerial vehicle (UAV) communication networks typically consist of multiple mobile UAV nodes that transmit mission data, status data, and control information via wireless links. Because UAV nodes are in a dynamic, mobile state, factors such as distance between nodes, obstruction relationships, channel occupancy, external interference, routing load, and communication task priorities change rapidly over time, leading to significant uncertainties in link quality, transmission latency, routing stability, and communication reliability within the UAV communication network.

[0003] Existing UAV communication optimization methods typically adjust transmit power, switch channels, select routes, or choose relay nodes based on current link quality, remaining energy, or mission requirements. While these methods can improve local communication quality to some extent, their optimization process relies heavily on current measured data and lacks simulation prediction of subsequent communication quality changes. When the UAV network topology changes rapidly, directly experimenting with strategies in a real communication network can easily lead to frequent link switching, communication interruptions, increased energy consumption, or unstable data transmission for critical missions.

[0004] Digital twin technology can construct a twin model corresponding to a physical object in virtual space and drive the twin model update through measured data. When using digital twins for UAV communication network optimization, the execution effect of communication strategies can be simulated in advance in the digital twin space, thereby reducing the risk of trial and error in the real network. However, existing digital twin communication optimization schemes still have the following shortcomings:

[0005] First, some solutions only use digital twin models as communication status display or offline simulation tools, failing to form a time-slice closed loop of "previous time slice backfeeding parameters - current prior twin state - current state assimilation correction - current policy pre-play - post-execution backfeeding update", resulting in unclear sources of the current twin state and the effective time of backfeeding parameters.

[0006] Secondly, although some schemes can calculate the synchronization deviation between the measured state and the twin state, they do not distinguish between the virtual-real synchronization deviation used for current state assimilation and the virtual-real deviation of strategy execution used for model backfeeding after execution. This can easily lead to the execution feedback that has not yet been generated in the current time slice being mistakenly used to correct the current simulation state, resulting in unclear data baseline.

[0007] Third, while some solutions can predict link quality or allocate resources, they do not conduct targeted local strategy simulations around the bottleneck link set, resulting in a large amount of computational load for the whole network simulation and insufficient real-time performance of online optimization.

[0008] Fourth, some schemes only update the state after the strategy is executed, without converting the deviation between the measured changes in communication quality and the saved results of the strategy pre-simulation into parameters that can be fed back, such as propagation loss, interference coupling, link quality prediction confidence, and strategy boundary. This may cause the candidate communication optimization strategy in the next time slice to still deviate from the real communication environment.

[0009] Therefore, it is necessary to provide a method and system for optimizing UAV communication networks based on digital twins, so that the measured state, prior twin state, state assimilation, twin simulation, policy pre-playing, policy execution and post-execution feedback of the UAV communication network form a closed loop that is strictly advanced according to time slices, so as to improve the real-time performance, reliability and environmental adaptability of the communication optimization strategy. Summary of the Invention

[0010] The purpose of this invention is to provide a method and system for optimizing UAV communication networks based on digital twins, in order to solve the problems existing in the optimization process of UAV communication networks, such as high risk of trial and error in real networks, unclear digital twin state benchmarks, easy mixing of virtual and real synchronization deviations and post-execution feedback deviations, large amount of computational workload in repeated simulation of the entire network, and the inability to adjust the boundary of communication optimization strategies online according to the execution effect.

[0011] To achieve the above objectives, the present invention provides a method for optimizing unmanned aerial vehicle (UAV) communication networks based on digital twins, comprising: The measured communication status data of the UAV communication network is obtained according to time slices. The measured communication status data includes UAV node location, inter-node link quality, channel occupancy, interference intensity, and communication task priority. In the current time slice For the initial time slice, initial prior twin communication state data is generated using preset initial communication simulation parameters, initial model parameters obtained from offline training, and initial candidate communication optimization strategy boundaries; in the current time slice... When it is not the initial time slice, read the previous time slice. The communication simulation parameters and candidate communication optimization strategy boundaries have been completed after online feedback, and the current time slice is generated in the digital twin space. Prior twin communication state data; Based on timestamps, node identifiers, and link topology constraints, the measured communication state data is paired with the prior twin communication state data to calculate the virtual-real synchronization deviation. Based on the virtual-real synchronization deviation, state assimilation correction is performed only on the prior twin communication state data to obtain the corrected twin communication state data. The virtual-real synchronization deviation is not used to generate propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and policy boundary correction parameters. Based on the corrected twin communication state data, joint simulation of link propagation, interference coupling, and routing load is performed to generate communication quality prediction values ​​and a set of bottleneck links; Focusing solely on the aforementioned bottleneck link set, candidate communication optimization strategies, including at least two actions among transmit power adjustment, channel switching, relay node selection, and route reconstruction, are pre-simulated in the digital twin space to determine the target communication optimization strategy and issue it for execution. Collect the measured change in communication quality after the execution of the target communication optimization strategy, and compare the measured change in communication quality with the twin predicted change in communication quality saved before the target communication optimization strategy was issued and executed to generate the virtual-real deviation of strategy execution. Based on the aforementioned strategy, the propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and strategy boundary correction parameters are generated to synchronously update the next time slice due to the virtual-real deviation. Communication simulation parameters and candidate communication optimization strategy boundaries; When the absolute value of the virtual-to-real deviation of the strategy execution of at least one bottleneck link exceeds a preset deviation threshold, or when the bottleneck link set changes, local simulation and local pre-simulation are only re-executed for UAV nodes, adjacent links, and occupied channels associated with the changed bottleneck link set. The number of times the local simulation and local pre-simulation are executed in the same time slice does not exceed a preset upper limit.

[0012] The present invention also provides a UAV communication network optimization system based on digital twins, including a measured acquisition and prior generation module, a state assimilation and synchronization module, a communication quality simulation module, a strategy pre-execution module, an execution deviation feedback module, and a local re-simulation module.

[0013] The measured data acquisition and prior generation module is used to acquire measured communication status data of the UAV communication network in time slices, and in the current time slice For the initial time slice, initial prior twin communication state data is generated using preset initial communication simulation parameters, initial model parameters obtained from offline training, and initial candidate communication optimization strategy boundaries; in the current time slice... When it is not the initial time slice, read the previous time slice. The communication simulation parameters and candidate communication optimization strategy boundaries have been completed after online feedback, and the current time slice is generated in the digital twin space. Prior twin communication state data.

[0014] The state assimilation and synchronization module is used to pair the measured communication state data with the prior twin communication state data based on the timestamp, node identifier, and link topology constraints, calculate the virtual-real synchronization deviation, and perform state assimilation correction only on the prior twin communication state data based on the virtual-real synchronization deviation to obtain the corrected twin communication state data.

[0015] The communication quality simulation module is used to perform joint simulation of link propagation, interference coupling, and routing load based on the corrected twin communication state data, and generate communication quality prediction values, prediction confidence, and bottleneck link sets.

[0016] The strategy pre-executing module is used to pre-execute candidate communication optimization strategies only around the bottleneck link set, determine the target communication optimization strategy and issue it for execution, and save the twin prediction of communication quality changes before the target communication optimization strategy is issued and executed.

[0017] The execution deviation feedback module is used to collect the measured communication quality changes after the target communication optimization strategy is executed. It compares the measured communication quality changes with the twin-predicted communication quality changes saved before the target communication optimization strategy is issued and executed, generating a virtual-real deviation in strategy execution. Based on this virtual-real deviation, it generates propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and strategy boundary correction parameters to synchronously update the next time slice. The communication simulation parameters and candidate communication optimization strategy boundaries.

[0018] The local re-simulation module is used to re-perform local simulation and local pre-simulation only for UAV nodes, adjacent links and occupied channels associated with the changed bottleneck link set when the absolute value of the virtual-to-real deviation of the strategy execution of at least one bottleneck link exceeds a preset deviation threshold, or when the bottleneck link set changes. The number of times the local simulation and local pre-simulation are executed in the same time slice does not exceed a preset upper limit.

[0019] Compared with the prior art, the present invention has at least the following beneficial effects:

[0020] First, the present invention generates initial prior twin communication state data in the initial time slice using initial simulation parameters or offline training models. In non-initial time slices, it reads the communication simulation parameters and candidate communication optimization strategy boundaries that have been completed online in the previous time slice to generate the current prior twin communication state data. This makes the source of the twin state and the effective time of the infeasible parameters clear, and avoids the execution feedback parameters that have not yet been generated in the current time slice being mistakenly used for the current state generation.

[0021] Second, this invention will address the virtual-real synchronization deviation. It is limited to the state assimilation deviation between the measured communication state data of the current time slice and the prior twin communication state data, and it is explicitly stated that it will not be used to generate propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters and policy boundary correction parameters, thereby avoiding the confusion between state synchronization correction and post-execution feedback update.

[0022] Third, this invention will address the virtual-real deviation in strategy execution. The method is limited to the difference between the measured change in communication quality after the actual execution of the target communication optimization strategy and the twin predicted change in communication quality saved before the target communication optimization strategy was issued and executed. This ensures that the parameters fed back after execution have a clear data source and comparison benchmark.

[0023] Fourth, this invention performs joint simulation of link propagation, interference coupling, and routing load based on the corrected twin communication state data, generating communication quality prediction values, prediction confidence levels, and bottleneck link sets. This enables the communication optimization target to locate critical links where communication quality deteriorates, load is abnormal, or high-priority tasks are affected, thereby improving the targeting of UAV communication network optimization.

[0024] Fifth, this invention focuses only on the generation and pre-simulation of candidate communication optimization strategies for the bottleneck link set and its one-hop adjacent links, reducing the computational load of repeated simulations across the entire network; at the same time, by limiting the action intensity of candidate strategies for low-confidence links through prediction confidence, it helps to avoid generating aggressive strategies under low-confidence prediction conditions.

[0025] Sixth, this invention adjusts the propagation loss correction parameter, interference coupling correction parameter, link quality prediction confidence parameter, and policy boundary correction parameter based on the trend of the change in the sign and absolute value of the policy execution virtual-real deviation. When the policy execution virtual-real deviation maintains the same sign and the absolute value increases, the policy boundary is tightened; when the absolute value of the policy execution virtual-real deviation decreases or falls below the convergence threshold, the policy boundary is maintained or relaxed, thereby improving the stability of the closed-loop optimization process.

[0026] Seventh, this invention limits the number of times local simulation and local pre-simulation are executed within the same time slice when local re-simulation is triggered, and limits the generation of new policy execution virtual-real deviations when the amount of communication quality change after actual execution is not re-acquired, thereby avoiding the formation of a data cycle of "execution-deviation-feedback-re-execution-re-deviation" within the same time slice. Attached Figure Description

[0027] Figure 1 This is a flowchart of the UAV communication network optimization method based on digital twin provided by the present invention.

[0028] Figure 2 This is a block diagram of the composition of the UAV communication network optimization system based on digital twin provided by the present invention. Detailed Implementation

[0029] The following section describes the UAV communication network optimization method and system based on digital twins, using an application scenario of UAV swarms performing communication support tasks. The UAV communication network includes multiple UAV nodes, which transmit task data, status data, and control data via wireless links. Since the location, link distance, channel occupancy, external interference, routing load, and communication task priority of UAV nodes change dynamically over time, this embodiment uses time slices as the processing unit. The online backfeedback results from the previous time slice only take effect when prior twin communication status data is generated in the next time slice. Then, the current measured communication status data is used to perform state assimilation correction on the prior twin communication status data. Finally, based on the corrected twin communication status data, communication quality simulation, bottleneck link identification, policy pre-playing, policy execution, and post-execution backfeedback updates are performed.

[0030] In this embodiment, the virtual-real synchronization deviation It is used only to measure the synchronization difference between the measured communication state data of the current time slice and the prior twin communication state data, and only for state assimilation correction of the prior twin communication state data; policy execution virtual-real deviation. Generated only after the target communication optimization strategy has been actually executed, it is used to generate propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and strategy boundary correction parameters. Therefore, the state synchronization correction of the current time slice is distinguished from the online backfeedback update after execution, avoiding the use of currently executed feedback parameters that have not yet been generated to reverse-correct the current simulation state.

[0031] Example 1

[0032] like Figure 1 As shown, Embodiment 1 provides a method for optimizing UAV communication networks based on digital twins, including the following steps.

[0033] S1. Obtain measured communication status data of the UAV communication network according to time slices.

[0034] In the current time slice Within the time slice, measured communication status data of the UAV communication network is acquired. This measured communication status data includes UAV node location, inter-node link quality, channel occupancy, interference intensity, and communication task priority. UAV node location indicates the spatial status of each UAV node within the current time slice; inter-node link quality indicates the actual transmission quality of the wireless communication link between UAV nodes; channel occupancy indicates the degree to which each communication channel is occupied; interference intensity indicates the impact of co-channel interference, adjacent-channel interference, or external electromagnetic interference on the link; and communication task priority indicates the transmission guarantee level of different task data within the current time slice.

[0035] Specifically, for drone nodes Get its current time slice The position, velocity, heading, remaining energy, and communication task priority within the node are used to form a measured node state vector; for UAV nodes... With drone nodes Links between The system acquires link signal strength, packet loss rate, transmission delay, available bandwidth, and link connectivity status, forming a measured link state vector. For channels occupied by UAV nodes, it acquires channel occupancy rate, co-channel interference intensity, adjacent channel interference intensity, and channel switching history, forming a measured channel state vector. The measured node state vector, measured link state vector, and measured channel state vector together constitute the current time slice. The measured communication status data.

[0036] In a preferred embodiment, a task priority protection window is also generated based on the communication task priority. The task priority protection window refers to the time range within a preset continuous time slice during which stability protection is performed on high-priority task links. For links carrying high-priority tasks, within the task priority protection window, subsequent candidate communication optimization strategies must not frequently change their main route, relay node, or target channel, unless the predicted communication quality value of the link is lower than the forced handover threshold.

[0037] The task priority protection window can be represented as: , in, For the task In the current time slice The generated task priority protection window, Number the communication task. For the current time slice, For the task The corresponding protection duration, For the current time slice Start Time Slice The continuous protection interval until the end.

[0038] In this step, the current time slice The data acquisition methods described above serve as a common time reference for subsequent a priori twin state generation, state assimilation correction, communication quality simulation, policy pre-playing, and post-execution feedback; the UAV node identifier serves as a data binding reference between the measured node and the twin node; the link topology serves as a data coupling reference between node state, link state, and channel state; and the task priority protection window serves as a constraint input for subsequent policy selection and policy boundary updates. Through these data acquisition methods, the simulation objects in the digital twin space can have a clear data source and time reference.

[0039] Through the above processing, the digital twin space is provided with the real communication state input of the current time slice, so that subsequent state assimilation, communication quality simulation and policy pre-play can simultaneously consider node movement, link fluctuation, channel congestion, interference changes and communication task priority.

[0040] S2. Generate prior twin communication state data for the current time slice.

[0041] Complete the current time slice After collecting the measured communication status data, determine the current time slice. Is this the initial time slice?

[0042] In the current time slice For the initial time slice, since there is no online feedback result from the previous time slice, initial prior twin communication state data is generated in the digital twin space using preset initial communication simulation parameters, initial model parameters obtained from offline training, and initial candidate communication optimization strategy boundaries. The preset initial communication simulation parameters may include initial parameters for the link propagation model, initial parameters for the interference coupling model, initial parameters for the routing load model, and initial parameters for link quality prediction confidence. The initial candidate communication optimization strategy boundaries may include initial transmit power adjustment boundaries, initial channel switching boundaries, initial relay node selection boundaries, and initial route reconstruction boundaries.

[0043] In the current time slice When it is not the initial time slice, read the previous time slice. The communication simulation parameters and candidate communication optimization strategy boundaries have been completed after online feedback. The communication simulation parameters include link propagation model parameters, interference coupling model parameters, routing load model parameters, and link quality prediction confidence parameters; the candidate communication optimization strategy boundaries include transmit power adjustment boundaries, channel switching boundaries, relay node selection boundaries, and route reconstruction boundaries. The previous time slice... The updated communication simulation parameters are applied to the digital twin model to generate the current time slice. Prior twin communication state data.

[0044] The prior twin communication state data includes prior twin node state vectors, prior twin link quality, and prior twin channel state. The prior twin communication state data represents the predictive twin state of the current communication network state formed by the digital twin space based on initial model parameters or the feedback results from the previous time slice, before assimilation using the measured communication state data of the current time slice.

[0045] In this step, the previous time slice The policy execution virtual-to-real deviation has been transformed into propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and policy boundary correction parameters at the end of the previous processing round, and the communication simulation parameters and candidate communication optimization policy boundaries have been updated. Therefore, in the current time slice In this context, the prior twin communication state data is not generated from the post-execution feedback that has not yet occurred in the current time slice, but rather from the initial parameters or the model state after the previous time slice has been fed back.

[0046] To ensure clear data versioning, this embodiment sets version identifiers for the measured communication state data, prior twin communication state data, corrected twin communication state data, and policy pre-simulation results. The version identifier for the measured communication state data records the actual data source for the current time slice; the version identifier for the prior twin communication state data records that it was generated from the initial model parameters or the model refeedback from the previous time slice; the version identifier for the corrected twin communication state data records that it has undergone state assimilation in the current time slice; and the version identifier for the policy pre-simulation results ensures that the virtual-to-real deviation calculation used in subsequent policy execution calls the twin-predicted communication quality change amount saved before the target communication optimization policy was issued and executed.

[0047] Through the above processing, a twin prior benchmark for the current time slice state assimilation is formed, the source of the twin state of the current time slice is clarified, the backfeed parameters that have not yet been generated after the current execution are incorrectly used for the current state generation, and the initial time slice can be started normally.

[0048] S3. Based on timestamps, node identifiers, and link topology constraints, calculate the virtual-real synchronization deviation and perform state assimilation correction on the prior twin communication state data.

[0049] After obtaining the measured communication state data and the prior twin communication state data, they are paired according to timestamps, node identifiers, and link topology constraints. Specifically, the measured node state vector is paired with the prior twin node state vector according to the node identifier; the measured link quality is paired with the prior twin link quality according to the link topology; and the measured channel state is paired with the prior twin channel state according to the channel identifier. After pairing, the current time slice is calculated. virtual-real synchronization deviation .

[0050] The virtual-real synchronization deviation satisfies: , in, For the current time slice The virtual-real synchronization deviation For a collection of drone nodes, and Number the drone nodes. For drone nodes The node state weights, For drone nodes In the current time slice The measured node state vector, For drone nodes In the current time slice The prior twin node state vector Denotes the vector norm, superscript Indicates the measured domain, superscript Indicates the prior twin field, For the current time slice The set of inter-node links, For drone nodes With drone nodes The link between them For link Link quality weights, For link In the current time slice The measured link quality, For link In the current time slice The quality of the prior twin link.

[0051] When the virtual-real synchronization deviation When the value exceeds the preset synchronization threshold, the node state weight is used as the basis for further action. Link quality weight Identify the main sources of deviation and prioritize performing state assimilation correction on twin UAV nodes and twin links that contribute significantly to the deviation, to obtain the corrected twin node state vectors and corrected twin link quality: , , in, For drone nodes In the current time slice The corrected twin node state vector For link In the current time slice The corrected twin link quality, superscript This represents the corrected twin domain. For drone nodes In the current time slice The node state assimilation coefficient, For link In the current time slice Link quality assimilation coefficient, and The values ​​are all located in Within the range.

[0052] It should be noted that the virtual-real synchronization deviation in this step It is only used for state assimilation correction in the current time slice, and not for generating propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and policy boundary correction parameters. Propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and policy boundary correction parameters are generated only from the virtual-to-real deviation of policy execution after the target communication optimization policy is executed.

[0053] In this step, the prior twin communication state data undergoes state assimilation correction to form corrected twin communication state data. This corrected twin communication state data serves as the input for the subsequent joint simulation of link propagation, interference coupling, and routing load in S4. This processing method ensures that the initial simulation state of the current time slice closely approximates the actual communication state, while maintaining the time order of subsequent feedback updates.

[0054] Through the above processing, the prior twin state of the current time slice is synchronously corrected, reducing the distortion of the current simulation initial state caused by initial model error or residual error of the model prediction in the previous time slice, and ensuring that the functional boundaries of state assimilation and post-execution backfeedback are clear.

[0055] S4. Based on the corrected twin communication state data, perform joint simulation of link propagation, interference coupling, and routing load to generate communication quality prediction values ​​and prediction confidence levels.

[0056] In the digital twin space, based on the corrected twin communication state data obtained from S3, a link propagation model, an interference coupling model, and a routing load model are constructed, and the current time slice of each twin link is analyzed. Joint simulation of communication quality within the system.

[0057] The link propagation model is used to predict the twin link propagation strength based on the corrected spatial distance between twin UAV nodes, relative altitude, node movement speed, heading angle, and environmental attenuation parameters; the interference coupling model is used to predict the twin interference coupling strength based on the corrected channel occupancy status, co-channel interference, adjacent channel interference, adjacent link transmit power, and channel switching history; and the routing load model is used to predict the twin routing load delay based on communication task priority, routing path, link-carrying traffic, queue length, and task priority protection window.

[0058] In one specific implementation, the predicted communication quality value for each twin link satisfies: , in, For link In the current time slice The predicted quality of twin communication, For link In the current time slice The propagation strength of twin links, For link In the current time slice The twin interference coupling strength, For link In the current time slice Twin routing load latency, For link In the current time slice Twin channel occupancy strength, , , and These are the quality weights for link propagation strength, interference coupling strength, routing load delay, and channel occupancy strength, respectively.

[0059] In this step, the link propagation strength Predicted values ​​for communication quality Produces a positive contribution, twin interference coupling strength Twin routing load latency and twin channel occupancy strength Predicted values ​​for communication quality It has a negative impact.

[0060] In a preferred embodiment, a prediction confidence score is further generated based on the historical strategy execution virtual-real deviation, the current virtual-real synchronization deviation, and the number of effective feedback samples. The prediction confidence score is used to constrain the action strength of subsequent candidate communication optimization strategies. The prediction confidence score can be expressed as: , in, For link In the current time slice The prediction confidence level This is a range cutoff function used to restrict the value within parentheses to a certain range. Within the interval, For link In the previous time slot The strategy execution deviates from the real one. For the current time slice The virtual-real synchronization deviation For link In the current time slice The number of valid feedback samples accumulated previously, , and These are the confidence weights for the historical policy execution virtual-to-real bias, virtual-to-real synchronization bias, and the reciprocal of the number of valid feedback samples, respectively. In the initial time slice, if there is no policy execution virtual-to-real bias from the previous time slice, then... Take 0, if the link If no valid feedback samples have been accumulated, then Set it to 0 so that the prediction confidence level is determined by the current virtual-real synchronization deviation and the preset confidence weight.

[0061] when When the link is below the confidence threshold, The link is marked as low confidence. Low confidence links still participate in communication quality prediction and bottleneck link identification, but their corresponding transmit power adjustment range, route reconstruction range, and relay node replacement range are limited.

[0062] In this step, the corrected twin communication state data is used as input for the co-simulation, and the predicted communication quality value is... As the basis for identifying bottleneck links in S5, and also as the input for calculating the pre-strategy revenue value in S7; prediction confidence. Used to constrain the generation and simulation strength of subsequent candidate communication optimization strategies.

[0063] Through the above processing, the communication quality of each twin link in the current time slice is predicted. The movement of UAV nodes, wireless propagation, interference superposition, routing load, channel congestion and prediction confidence are coupled into a computable communication quality prediction result, and the prediction confidence is avoided from exceeding a reasonable range by interval truncation.

[0064] S5. Generate a set of bottleneck links based on the predicted communication quality values.

[0065] Based on the communication quality prediction values ​​of each twin link obtained in S4, a bottleneck link set is identified. This bottleneck link set consists of twin links whose communication quality prediction values ​​are below a quality threshold, twin links whose communication quality prediction values ​​decrease by more than a threshold between adjacent time slots, and twin links carrying high-priority communication tasks with routing loads exceeding a load threshold. Specifically, in the initial time slot where there is no communication quality prediction value from the previous time slot, the judgment of the decrease in communication quality prediction values ​​between adjacent time slots is not enabled; the bottleneck link set is determined solely based on the current communication quality prediction value, communication task priority, and routing load. From the time slot where there is a communication quality prediction value from the previous time slot, the judgment of the decrease in communication quality prediction values ​​between adjacent time slots is enabled.

[0066] Specifically, for the link If its twin communication quality prediction value If the quality value is below the quality threshold, the link is added to the bottleneck link set. If the predicted communication quality value of the link decreases more than the decrease threshold between adjacent time slices, the link is added to the bottleneck link set. If the link carries a high-priority communication task and its routing load or queue length is higher than the load threshold, the link is added to the bottleneck link set even if its current predicted communication quality value has not yet fallen below the quality threshold.

[0067] After generating the bottleneck link set, the one-hop adjacent links of the bottleneck link set are further determined. A one-hop adjacent link is a link that shares at least one UAV node with the bottleneck link. For bottleneck links... With drone nodes Other directly connected links and drone nodes Any other directly connected links can serve as one-hop adjacent links to the bottleneck link. Subsequent candidate communication optimization strategies are generated only for the bottleneck link set and its one-hop adjacent links.

[0068] In a preferred embodiment, a bottleneck region topology fingerprint is also generated based on the bottleneck link set. The bottleneck region topology fingerprint is used to record the local topology, channel occupancy pattern, and task carrying status associated with the bottleneck links. The bottleneck region topology fingerprint can be represented as: , in, Bottleneck area In the current time slice Topological fingerprint, Number the bottleneck areas. Bottleneck area In the current time slice The set of bottleneck link identifiers included. Bottleneck area In the current time slice The set of one-hop adjacent node identifiers Bottleneck area In the current time slice The set of occupied channel identifiers Bottleneck area In the current time slice The priority level of the tasks they carry. Bottleneck area In the current time slice Link quality level, Bottleneck area In the current time slice Interference level.

[0069] In this step, the predicted communication quality, the link quality decline trend, the communication task priority, the routing load, and the prediction confidence jointly determine the bottleneck link set; the bottleneck link set further limits the generation range of candidate communication optimization strategies in S6; the topology fingerprint of the bottleneck region is used for historical experience matching in subsequent strategy pre-playing.

[0070] Through the above processing, the key links affecting the reliability of the UAV communication network are located in the digital twin space, and the optimization target is narrowed from the entire network link to the local communication area that has a greater impact on communication quality, thereby reducing the amount of online computing.

[0071] S6. Generate candidate communication optimization strategies only around the bottleneck link set.

[0072] After the bottleneck link set is determined, candidate communication optimization strategies are generated only for the bottleneck link set and its one-hop adjacent links. Candidate communication optimization strategies include at least two actions from the following: transmit power adjustment, channel switching, relay node selection, and route reconstruction. A relay node is a UAV node that forwards communication data between the source UAV node and the target UAV node.

[0073] The generation boundary of candidate communication optimization strategies is determined by the strategy boundary correction results after online feedback in the previous time slice. The strategy boundaries include transmit power adjustment boundaries, channel switching boundaries, relay node selection boundaries, and route reconstruction boundaries. The transmit power adjustment boundary limits the range of transmit power that UAV nodes are allowed to increase or decrease within the current time slice; the channel switching boundary limits the set of target channels allowed for switching and the number of switching attempts; the relay node selection boundary limits the range of UAV nodes that can participate in relay forwarding and their load limits; and the route reconstruction boundary limits the number of route hops and the range of route paths that can be changed.

[0074] Specifically, when the link propagation strength of the bottleneck link is lower than the propagation strength threshold and the interference coupling strength is lower than the interference strength threshold, a transmit power adjustment action is generated; when the channel occupancy strength or interference coupling strength of the bottleneck link is higher than the corresponding threshold, a channel switching action is generated; when the direct communication quality between the nodes at both ends of the bottleneck link is lower than the quality threshold and the surrounding adjacent nodes meet the relay quality conditions, a relay node selection action is generated; when the bottleneck link is located in the congested area of ​​the current routing path or the routing load delay is higher than the delay threshold, a route reconstruction action is generated. These actions are not generated in isolation, but rather are generated based on a combination of the bottleneck link's link propagation strength, interference coupling strength, channel occupancy strength, routing load delay, communication task priority, and prediction confidence.

[0075] In a preferred embodiment, if the bottleneck link is within the task priority protection window, a protection action screening rule is introduced when generating candidate communication optimization strategies. The protection action screening rule includes: prohibiting repeated changes to the main route of the same high-priority task within at least two consecutive time slices; prohibiting switching high-priority tasks to relay nodes with a predicted confidence level below a confidence threshold; when channel switching is necessary, prioritizing channels with higher historical switching success rates for the same task; and when route reconstruction is necessary, retaining at least one critical relay node from the previous stable route.

[0076] In this step, the bottleneck link set and one-hop adjacent links output by S5 determine the target of the candidate communication optimization strategy; the boundary of the candidate communication optimization strategy updated in the previous time slice determines the search range of the candidate communication optimization strategy; and the task priority protection window and prediction confidence determine the action strength of the candidate communication optimization strategy.

[0077] Through the above processing, a combination of communication optimization actions can be generated in advance in the digital twin space, avoiding blindly adjusting the transmission power, channel, relay node or routing path directly in the real UAV communication network.

[0078] S7. Perform a pre-play of candidate communication optimization strategies in the digital twin space and calculate the strategy pre-play payoff value.

[0079] For each candidate communication optimization strategy generated by S6, the strategy is pre-executed in the digital twin space, and the predicted communication quality, predicted energy consumption, link switching cost, and routing disturbance cost of the bottleneck link set and its one-hop adjacent links are recalculated. The strategy pre-playing does not involve immediately issuing control commands to the real UAV communication network, but rather simulating the changes in communication state after the strategy is executed in the digital twin space.

[0080] In one specific implementation, the strategy pre-simulation payoff value satisfies: , in, Optimize candidate communication strategies In the current time slice The strategy preview return value, Number the candidate communication optimization strategy. Optimize candidate communication strategies In the current time slice Twin prediction of communication quality gain, Optimize candidate communication strategies In the current time slice Twin prediction of energy consumption cost Optimize candidate communication strategies In the current time slice The cost of twin link switching, Optimize candidate communication strategies In the current time slice The cost of twin routing perturbation, , , and These are the weights for the gains in predicting communication quality, predicting energy consumption costs, link switching costs, and routing disturbance costs, respectively.

[0081] During the strategy simulation, if the candidate communication optimization strategy includes a transmit power adjustment action, the transmit power of the relevant UAV nodes is updated in the digital twin space, and the twin interference coupling strength of the link and its adjacent links is recalculated; if the candidate communication optimization strategy includes a channel switching action, the channel state of the target link is updated, and the channel occupancy strength and interference coupling strength are recalculated; if the candidate communication optimization strategy includes a relay node selection action, the target relay node is added to the routing path, and the routing load delay and energy consumption cost are recalculated; if the candidate communication optimization strategy includes a route reconstruction action, the set of links traversed by the data flow is updated, and the total path communication quality and routing disturbance cost are recalculated.

[0082] After each candidate communication optimization strategy completes its pre-performance, its twin-predicted communication quality change, predicted energy consumption cost, link switching cost, routing disturbance cost, and strategy pre-performance benefit value are written into the strategy pre-performance record, and a pre-performance version identifier is set. For subsequent candidate communication optimization strategies selected as the target communication optimization strategy, the twin-predicted communication quality change saved during its pre-performance is used as the comparison benchmark for S9 to calculate the virtual-to-real deviation of strategy execution.

[0083] In this step, candidate communication optimization strategies are used as inputs, and predicted communication quality, energy consumption cost, link switching cost, and routing disturbance cost are used as intermediate results. The strategy pre-simulation benefit value is also considered. The target communication optimization strategy is selected by S8 as the basis; the twin predicted communication quality change amount saved during the target communication optimization strategy pre-run serves as the fixed benchmark for deviation calculation after S9 execution.

[0084] Through the above processing, the execution effect of candidate communication optimization strategies can be evaluated without affecting the operation of the real UAV communication network, reducing ineffective adjustments and frequent switching in the real network, and ensuring that the virtual-real deviation of subsequent strategy execution has a clear twin pre-simulation comparison object.

[0085] S8. Determine the target communication optimization strategy based on the strategy pre-simulation results and issue it for execution.

[0086] After obtaining the policy pre-simulation benefit values ​​of each candidate communication optimization strategy, the candidate communication optimization strategies are constrained and screened. Specifically, candidate communication optimization strategies that cause the UAV node's transmit power to exceed the transmit power adjustment boundary, the number of channel handovers to exceed the channel handover boundary, the relay node load to exceed the relay node selection boundary, or the route hop count to exceed the route reconstruction boundary are eliminated.

[0087] Among the remaining candidate communication optimization strategies, the candidate communication optimization strategy with the highest strategy pre-simulation benefit value is selected first; when the strategy pre-simulation benefit values ​​of two candidate communication optimization strategies are the same or the difference is lower than the preset difference threshold, the candidate communication optimization strategy that covers more bottleneck links is selected as the target communication optimization strategy; if the number of bottleneck links covered is the same, the candidate communication optimization strategy with lower link switching cost is selected first; if the link switching cost is the same, the candidate communication optimization strategy with lower routing disturbance cost is selected first.

[0088] In a preferred embodiment, if the target communication optimization strategy changes the main route, relay node, or target channel carrying high-priority communication tasks, it is determined whether the forced handover conditions are met before execution. If the forced handover conditions are not met, the target communication optimization strategy is downgraded to a candidate strategy, and a second-best candidate communication optimization strategy is selected as the new target communication optimization strategy; if the forced handover conditions are met, the target communication optimization strategy is allowed to execute, and a protection fallback flag is added to the strategy execution packet.

[0089] Once the target communication optimization strategy is determined, a strategy execution package is generated and distributed to the actual UAV communication network. The strategy execution package includes UAV node identifiers, target transmit power, target channel, target relay node, target route, execution effective time, and rollback conditions. The UAV node identifier is used to identify the target of the strategy; the target transmit power is used to control the transmit parameters of the relevant UAV nodes; the target channel is used to control link switching; the target relay node is used to determine the forwarding node; the target route is used to determine the task data transmission path; the execution effective time is used to ensure that multiple UAV nodes execute the strategy synchronously; and the rollback conditions are used to restore the communication to the previous stable communication state if the communication quality does not meet the preset improvement requirements after execution.

[0090] In this step, the policy pre-simulation benefit value output by S7 and the number of bottleneck links covered jointly determine the target communication optimization policy; the task priority protection window determines whether the target communication optimization policy needs to be subject to task protection verification; the actual execution results generated after the target communication optimization policy is issued serve as the input for S9 to calculate the measured communication quality change and the virtual-real deviation of policy execution.

[0091] Through the above processing, the strategies with better verification results in the digital twin space are applied to the real UAV communication network, which reduces the risks of power overrun, frequent switching, relay overload, excessive routing disturbance and short-term interruption of high-priority tasks while ensuring the improvement of communication.

[0092] S9. Collect the measured changes in communication quality after the execution of the target communication optimization strategy, generate the virtual-real deviation of strategy execution, and generate online backfeeding parameters.

[0093] After the target communication optimization strategy is executed, the post-execution link quality of the bottleneck link set and its one-hop adjacent links is collected and compared with the link quality before execution to obtain the measured change in communication quality. Simultaneously, the twin-predicted change in communication quality saved during the S7 pre-exercise of the target communication optimization strategy is read and compared to obtain the virtual-actual deviation of strategy execution.

[0094] The virtual-to-real deviation of the strategy execution is calculated according to the bottleneck link, satisfying the following: , in, For link In the current time slice The strategy execution deviates from the real one. After the target communication optimization strategy is executed, the link will be optimized. In the current time slice The measured change in communication quality, superscript This represents the actual measured results after execution. The predicted changes in communication quality, simulated and saved in the digital twin space before the target communication optimization strategy is issued and executed, are indicated by the superscript. This indicates the results of the twin simulation saved before execution.

[0095] when When the measured improvement after implementing the target communication optimization strategy is lower than the improvement predicted by the twin, it indicates that the digital twin space may underestimate propagation loss or interference coupling strength, or overestimate the link quality prediction confidence. In this case, improving the link... The corresponding propagation loss correction parameters and interference coupling correction parameters are applied, and the link quality prediction confidence parameter is reduced. When When the measured improvement after implementing the target communication optimization strategy is higher than the improvement predicted by the twin, it indicates that the digital twin space may have overly conservatively estimated the propagation conditions or interference impact on this link; at this point, reducing the link... The corresponding propagation loss correction parameters or improved link quality prediction confidence parameters.

[0096] It should be noted that the propagation loss correction parameter, interference coupling correction parameter, link quality prediction confidence parameter, and policy boundary correction parameter are determined solely by the virtual-to-real deviation of the policy execution. Generation, not caused by virtual-real synchronization deviation Generation. Virtual-real synchronization deviation. Used only for state assimilation correction in S3, policy execution virtual-real bias Only then is it used for online re-upload and update after execution.

[0097] In a preferred embodiment, the online backfeed parameters are generated using a dual-speed online backfeed method. Dual-speed online backfeed includes fast backfeed and slow backfeed. Fast backfeed is used to quickly correct the propagation loss correction parameters, interference coupling correction parameters, and policy boundary correction parameters of the current bottleneck link based on a large policy deviation within a single time slice. Slow backfeed is used to correct the link quality prediction confidence parameters and model base parameters based on the policy deviation trend over at least two consecutive time slices.

[0098] Dual-speed online recharge can be represented as: , , in, For link In the current time slice Rapid recharge parameters, For link In the previous time slot Rapid recharge parameters, To achieve rapid recharge step size, For link In the current time slice The strategy execution deviates from the real one. For link In the current time slice Slow reinjection parameters, For link In the previous time slot Slow reinjection parameters, For slow refill step size, For link Up to the current time Multiple time-slice strategies are used to smooth the virtual-to-real deviation value.

[0099] In this step, the policy execution virtual-real deviation serves as the sole source of deviation for generating online feedback parameters; propagation loss correction parameters and interference coupling correction parameters are used to update the communication simulation parameters for the next time slice; link quality prediction confidence parameters are used to adjust the confidence weight of the communication quality prediction value for the next time slice in the policy pre-play; and policy boundary correction parameters are used to update the candidate communication optimization policy boundary for the next time slice.

[0100] Through the above processing, the actual execution effect of the target communication optimization strategy is fed back to the digital twin space, enabling the digital twin space to continuously correct itself based on the response results of the real UAV communication network, and ensuring that the online backfeed parameters have a clear post-execution feedback source.

[0101] S10. Utilize online backfeed parameters to synchronously update the communication simulation parameters and candidate communication optimization strategy boundaries for the next time slice, and perform local re-simulation judgment.

[0102] Based on the propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and policy boundary correction parameters generated by S9, the next time slice is updated synchronously. The communication simulation parameters and candidate communication optimization strategy boundaries.

[0103] Specifically, the propagation loss correction parameter applies to the link propagation model, correcting the propagation strength of twin links under the same link or similar spatial conditions in the next time slice; the interference coupling correction parameter applies to the interference coupling model, correcting the interference superposition relationship in the same channel or adjacent channels in the next time slice; and the link quality prediction confidence parameter is used to update the prediction confidence of the corresponding link in the next time slice. And it applies to the credibility weight of the communication quality prediction value in the calculation of the policy pre-simulation benefit; when the link When the link quality prediction confidence parameter decreases, the prediction confidence for the next time slice... The corresponding reduction reduces the impact of the predicted benefits of the candidate communication optimization strategies for that link, or increases the priority of resimulating that link; when the link When the link quality prediction confidence parameter is improved, the prediction confidence of the next time slice... The corresponding increase, but still passed Limited to Within the interval; the policy boundary correction parameter acts on the candidate communication optimization policy generation process, so that the range of transmit power adjustment, channel switching, relay node selection and route reconstruction in the next time slice are adjusted according to the actual execution effect.

[0104] When the virtual-to-real deviation of the strategy execution remains the same sign and its absolute value increases for at least two consecutive time slices, it indicates that the prediction error of the digital twin space for the relevant bottleneck link continues to expand. At this time, the transmit power adjustment boundary and route reconstruction boundary of the corresponding bottleneck link should be narrowed, and the quality threshold for candidate communication optimization strategies to enter the pre-exercise should be increased. When the absolute value of the virtual-to-real deviation of the strategy execution decreases or falls below the convergence threshold for at least two consecutive time slices, it indicates that the prediction result of the digital twin space for the relevant bottleneck link gradually stabilizes. At this time, the boundary of the candidate communication optimization strategy for the corresponding bottleneck link should be maintained or relaxed.

[0105] In a preferred embodiment, a policy boundary recovery criterion is further provided. This criterion is used to determine whether the narrowed candidate communication optimization policy boundary can be restored to a normal range. The policy boundary recovery criterion can be expressed as: , in, For link In the current time slice Boundary restoration markers, Indicates permission to access the link The boundaries of candidate communication optimization strategies are recovered. This indicates that access to the link is not allowed. The boundaries of candidate communication optimization strategies are recovered. To recover the time slice index within the criterion observation interval, To restore the observation length of the criterion, For time slices Up to the current time slice The observation interval For link In time slice The strategy execution deviates from the real one. To restore the threshold, For link In time slice The prediction confidence level To restore the confidence threshold; when the number of available historical time slices is less than At that time, the observation interval is formed by existing historical time slices. If there are no usable historical time slices or the above recovery conditions are not met, then... .

[0106] When the absolute value of the virtual-to-real deviation of the strategy execution on at least one bottleneck link exceeds a preset deviation threshold, or when the bottleneck link set changes, link propagation, interference coupling, routing load joint simulation, and candidate communication optimization strategy pre-playing are only re-executed for the UAV nodes, adjacent links, and occupied channels associated with the changed bottleneck link set. Associated UAV nodes include UAV nodes at both ends of the bottleneck link and UAV nodes with a one-hop adjacency relationship with the bottleneck link; adjacent links include links that share endpoints with the bottleneck link; occupied channels include the channels currently used by the bottleneck link, target candidate channels, and channels that cause co-channel or adjacent-channel interference to the bottleneck link.

[0107] Within the same time slice, the number of executions of local simulation and local pre-performance does not exceed a preset limit. The results of local simulation and local pre-performance are used to update the pre-performance results of candidate communication optimization strategies for the associated region or the local strategy candidate pool for the next time slice; unless new changes in communication quality after actual execution are re-acquired, new policy execution virtual-real deviations will not be generated again. This limitation avoids data loops of repeated execution, repeated backfeeding, and repeated re-pre-performance within the same time slice.

[0108] In a preferred embodiment, local resimulation further includes local sample isolation. When the absolute value of the virtual-to-real deviation of the strategy execution of at least one bottleneck link within a bottleneck region exceeds a preset deviation threshold, the feedback samples generated in the current time slice of that bottleneck region are temporarily isolated and not immediately used for global model training, but only for local model updates of that bottleneck region and its similar topological fingerprint regions. If the isolated samples are verified to be stable and effective in subsequent time slices, they are then added to the global training samples.

[0109] In this step, the online backfeeding parameters generated in S9 are used as input. The updated communication simulation parameters affect the generation of the prior twin communication state in the next time slice S2 and the joint simulation in S4. The updated candidate communication optimization strategy boundary affects the generation of candidate strategies in the next time slice S6. The local re-simulation judgment affects whether to quickly recalculate the associated region.

[0110] Through the above processing, the virtual-real deviation after execution is transformed into simulation parameters and policy boundary updates for the next time slice, enabling the UAV communication network optimization process to have self-correction capabilities. It also reduces the computational load of repeated simulations of the entire network when the bottleneck area changes or the execution deviation is too large, while avoiding the formation of infinite data loops within the same time slice.

[0111] Example 2

[0112] like Figure 2 As shown in Embodiment 2, a UAV communication network optimization system based on digital twins is provided for executing the method described in Embodiment 1. The system includes a measured data acquisition and prior knowledge generation module, a state assimilation and synchronization module, a communication quality simulation module, a strategy pre-simulation and execution module, an execution deviation feedback module, and a local re-simulation module.

[0113] The measured data acquisition and prior generation module is used to acquire measured communication status data of the UAV communication network in time slices, and in the current time slice For the initial time slice, initial prior twin communication state data is generated using preset initial communication simulation parameters, initial model parameters obtained from offline training, and initial candidate communication optimization strategy boundaries; in the current time slice... When it is not the initial time slice, read the previous time slice. The communication simulation parameters and candidate communication optimization strategy boundaries have been completed after online feedback, and the current time slice is generated in the digital twin space. Prior twin communication state data.

[0114] The measured acquisition and prior generation module obtains the UAV node location, inter-node link quality, channel occupancy, interference intensity, and communication task priority, and forms measured node state vector, measured link state vector, and measured channel state vector; at the same time, it generates prior twin node state vector, prior twin link quality, and prior twin channel state based on the initial model parameters or the communication simulation parameters that have been fed back in the previous time slice.

[0115] The state assimilation and synchronization module is used to pair the measured communication state data with the prior twin communication state data based on timestamps, node identifiers, and link topology constraints, calculate the virtual-real synchronization deviation, and perform state assimilation correction only on the prior twin communication state data based on the virtual-real synchronization deviation to obtain the corrected twin communication state data. The state assimilation and synchronization module does not generate propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and policy boundary correction parameters based on the virtual-real synchronization deviation; its output corrected twin communication state data is transmitted to the communication quality simulation module.

[0116] The communication quality simulation module is used to perform joint simulations of link propagation, interference coupling, and routing load based on the corrected twin communication state data, generating predicted communication quality values, prediction confidence scores, and a bottleneck link set. The communication quality simulation module includes a link propagation unit, an interference coupling unit, a routing load unit, a prediction confidence unit, and a bottleneck link identification unit. The link propagation unit calculates the twin link propagation strength based on the spatial state between the corrected twin UAV nodes; the interference coupling unit calculates the twin interference coupling strength based on channel occupancy and interference intensity; the routing load unit calculates the twin routing load delay based on communication task priority, task priority protection window, and path-carrying traffic; the prediction confidence unit generates a prediction confidence score based on the historical policy execution virtual-real deviation, virtual-real synchronization deviation, and the number of effective feedback samples, and limits the prediction confidence score to a certain range using an interval truncation function. Within the interval; the bottleneck link identification unit is used to determine the bottleneck link set based on the predicted communication quality value, the magnitude of communication quality degradation, the load of high-priority communication tasks, and the prediction confidence.

[0117] The link quality prediction confidence parameter is used to update the prediction confidence of the link corresponding to the next time slice. And it applies to the credibility weight of the communication quality prediction value in the calculation of the policy pre-simulation benefit; when the link When the link quality prediction confidence parameter decreases, the prediction confidence for the next time slice... The corresponding reduction reduces the impact of the predicted benefits of the candidate communication optimization strategies for that link, or increases the priority of resimulating that link; when the link When the link quality prediction confidence parameter is improved, the prediction confidence of the next time slice... The corresponding increase, but still passed Limited to Within the interval; the policy boundary correction parameter acts on the candidate communication optimization policy generation process, so that the range of transmit power adjustment, channel switching, relay node selection and route reconstruction in the next time slice are adjusted according to the actual execution effect.

[0118] The strategy pre-executing module is used to pre-execute candidate communication optimization strategies, including at least two actions among transmit power adjustment, channel switching, relay node selection and route reconstruction, focusing only on the bottleneck link set, determine the target communication optimization strategy and issue it for execution, and save the twin prediction of communication quality changes before the target communication optimization strategy is issued and executed.

[0119] The strategy pre-simulation execution module includes a candidate strategy generation unit, a strategy pre-simulation unit, a strategy filtering unit, a task protection verification unit, and a strategy distribution unit. The candidate strategy generation unit generates candidate communication optimization strategies based on the bottleneck link set and its one-hop adjacent links; the strategy pre-simulation unit simulates the execution effect of each candidate communication optimization strategy in a digital twin space and calculates the strategy pre-simulation benefit value; the strategy filtering unit eliminates candidate communication optimization strategies that exceed the transmit power adjustment boundary, channel switching boundary, relay node selection boundary, or route reconstruction boundary; the task protection verification unit determines whether the target communication optimization strategy disrupts high-priority communication tasks within the task priority protection window; and the strategy distribution unit generates strategy execution packets and distributes them to the real UAV communication network.

[0120] The execution deviation feedback module is used to collect the measured communication quality changes after the target communication optimization strategy is executed. It compares the measured communication quality changes with the twin-predicted communication quality changes saved before the target communication optimization strategy is issued and executed, generating a virtual-real deviation in strategy execution. Based on this virtual-real deviation, it generates propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and strategy boundary correction parameters to synchronously update the next time slice. The communication simulation parameters and candidate communication optimization strategy boundaries.

[0121] The execution deviation feedback module includes an execution quality acquisition unit, an execution deviation calculation unit, a feedback parameter generation unit, a model parameter update unit, and a policy boundary update unit. The execution quality acquisition unit collects the measured changes in communication quality after the execution of the target communication optimization strategy. The execution deviation calculation unit calls the twin predictions of communication quality changes saved before the target communication optimization strategy was issued and executed to calculate the virtual-to-real deviation of the strategy execution, avoiding using state assimilation deviation as the basis for feedback after execution. The feedback parameter generation unit generates propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and policy boundary correction parameters based on the virtual-to-real deviation of the strategy execution. The model parameter update unit updates the link propagation model, interference coupling model, and routing load model. The policy boundary update unit updates the transmit power adjustment boundary, channel switching boundary, relay node selection boundary, and route reconstruction boundary.

[0122] The local re-simulation module is used to re-perform local simulation and local pre-simulation only for UAV nodes, adjacent links, and occupied channels associated with the changed bottleneck link set when the absolute value of the virtual-to-real deviation of the strategy execution on at least one bottleneck link exceeds a preset deviation threshold, or when the bottleneck link set changes. The number of executions of the local simulation and local pre-simulation within the same time slice does not exceed a preset upper limit. The local re-simulation module receives the virtual-to-real deviation of strategy execution from the execution deviation feedback module, receives bottleneck link set change information from the communication quality simulation module, and determines the associated UAV nodes, adjacent links, and occupied channels. Subsequently, it calls the communication quality simulation module to recalculate the predicted communication quality value for the associated area and calls the strategy pre-simulation execution module to re-execute the candidate communication optimization strategy pre-simulation.

[0123] After the local re-simulation module is triggered, the results of local simulation and local pre-simulation are used to update the pre-simulation results of candidate communication optimization strategies for the associated region or the local strategy candidate pool for the next time slice. Unless new changes in communication quality after actual execution are re-acquired, new policy execution virtual-real deviations are not generated again. This process avoids the formation of infinite loops within the same time slice due to local re-simulation.

[0124] In the above system, the measured acquisition and prior generation module, the state assimilation and synchronization module, the communication quality simulation module, the policy pre-simulation execution module, the execution deviation feedback module, and the local re-simulation module are connected in the order of data flow. The measured acquisition and prior generation module outputs measured communication state data and prior twin communication state data; the state assimilation and synchronization module outputs corrected twin communication state data; the communication quality simulation module outputs communication quality prediction values, prediction confidence, and bottleneck link sets; the policy pre-simulation execution module outputs the target communication optimization policy, policy pre-simulation results, and policy execution package; the execution deviation feedback module outputs the policy execution virtual-real deviation, online feedback parameters, updated communication simulation parameters, and updated policy boundaries; and the local re-simulation module re-feeds the associated region to the communication quality simulation module and the policy pre-simulation execution module when the trigger conditions are met.

[0125] The communication simulation parameters and candidate communication optimization strategy boundaries fed back from the execution deviation feedback module to the measured acquisition and prior generation module only take effect when prior twin communication state data is generated in the next time slice, and do not reverse the completed strategy pre-play results and strategy execution virtual-real deviation calculation results in the current time slice.

[0126] Through the above specific implementation methods, this invention forms a closed-loop processing link: "Initial parameters or previous time-slice refeeding parameters generate the current prior twin state—current measured state and prior twin state are assimilated—communication quality simulation is performed based on the corrected twin state—policy pre-playing is performed around the bottleneck link—target policy is actually executed—post-execution deviation generates online refeeding parameters—updates simulation parameters and policy boundaries for the next time slice—local re-simulation is performed when necessary." This processing method reduces virtual-real synchronization deviation. Deviation between actual and actual strategy execution They are respectively responsible for the different functions of state assimilation and post-execution backfeeding, avoiding the confusion of twin state benchmarks caused by mixing the two, and avoiding data looping within the same time slice by limiting the number of local resimulations, thereby improving the logical consistency, real-time performance and reliability of the UAV communication network optimization process.

[0127] The embodiments described above are only for illustrating the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, without departing from the technical concept of the present invention, the time slice length, threshold parameters, model training methods, number of policy pre-drills, online backfeeding frequency, upper limit of local re-simulation times, and module deployment methods in the above embodiments can be adaptively adjusted according to the scale of the UAV communication network, communication task type, channel environment, node movement mode, and computing power conditions. As long as the above adjustments are still based on the state assimilation relationship between the measured communication state data and the prior twin communication state data, using the corrected twin communication state data to identify bottleneck links, and optimizing the UAV communication network through policy pre-drilling, real execution feedback, and online parameter backfeeding, they should all fall within the scope of protection of the present invention.

Claims

1. A method for optimizing unmanned aerial vehicle (UAV) communication networks based on digital twins, characterized in that, include: The measured communication status data of the UAV communication network is obtained according to time slices. The measured communication status data includes UAV node location, inter-node link quality, channel occupancy, interference intensity, and communication task priority. In the current time slice For the initial time slice, initial prior twin communication state data is generated using preset initial communication simulation parameters, initial model parameters obtained from offline training, and initial candidate communication optimization strategy boundaries; in the current time slice... When it is not the initial time slice, read the previous time slice. The communication simulation parameters and candidate communication optimization strategy boundaries have been completed after online feedback, and the current time slice is generated in the digital twin space. Prior twin communication state data; Based on timestamps, node identifiers, and link topology constraints, the measured communication state data is paired with the prior twin communication state data to calculate the virtual-real synchronization deviation. Based on the virtual-real synchronization deviation, state assimilation correction is performed only on the prior twin communication state data to obtain the corrected twin communication state data. The virtual-real synchronization deviation is not used to generate propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and policy boundary correction parameters. Based on the corrected twin communication state data, joint simulations of link propagation, interference coupling, and routing load are performed to generate predicted communication quality values ​​and a set of bottleneck links. Focusing solely on the aforementioned bottleneck link set, candidate communication optimization strategies, including at least two actions from transmit power adjustment, channel switching, relay node selection, and route reconstruction, are pre-simulated in the digital twin space to determine the target communication optimization strategy and execute it. Collect the measured change in communication quality after the execution of the target communication optimization strategy, and compare the measured change in communication quality with the twin predicted change in communication quality saved before the target communication optimization strategy was issued and executed to generate the virtual-real deviation of strategy execution. Based on the aforementioned strategy, propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and strategy boundary correction parameters are generated to synchronously update the next time slice due to virtual-real deviation. Communication simulation parameters and candidate communication optimization strategy boundaries; When the absolute value of the virtual-to-real deviation of the strategy execution of at least one bottleneck link exceeds a preset deviation threshold, or when the bottleneck link set changes, local simulation and local pre-simulation are only re-executed for UAV nodes, adjacent links, and occupied channels associated with the changed bottleneck link set. The number of times the local simulation and local pre-simulation are executed in the same time slice does not exceed a preset upper limit.

2. The UAV communication network optimization method based on digital twin as described in claim 1, characterized in that, Pairing the measured communication state data with the prior twin communication state data includes: forming a measured node state vector based on the UAV node position, speed, heading, remaining energy, and communication task priority; forming a measured link state vector based on the inter-node link signal strength, packet loss rate, transmission delay, available bandwidth, and link connectivity; and forming a measured channel state vector based on the channel occupancy rate, co-channel interference intensity, adjacent channel interference intensity, and channel switching history. The previous time slice The communication simulation parameters, after online backfeedback, are applied to the digital twin model to generate the current time slice. The prior twin node state vector, prior twin link quality, and prior twin channel state; The measured node state vector is paired with the prior twin node state vector according to the node identifier, and the measured link state vector is paired with the prior twin link quality according to the link topology. Version identifiers are set for the measured communication state data, the prior twin communication state data, the corrected twin communication state data, and the strategy pre-simulation results, so that the twin predicted communication quality change amount saved before the target communication optimization strategy is issued and executed can be called when calculating the virtual-real deviation of the strategy execution.

3. The UAV communication network optimization method based on digital twin as described in claim 1, characterized in that, The virtual-real synchronization deviation satisfies: ， in, For the current time slice The virtual-real synchronization deviation, For a collection of drone nodes, and Number the drone nodes. For drone nodes The node state weights, For drone nodes In the current time slice The measured node state vector, For drone nodes In the current time slice The prior twin node state vector Denotes the vector norm, superscript Indicates the measured domain, superscript Indicates the prior twin field, For the current time slice The set of inter-node links, For drone nodes With drone nodes The link between them For link Link quality weights, For link In the current time slice The measured link quality, For link In the current time slice Prior twin link quality; When the virtual-real synchronization deviation exceeds a preset synchronization threshold, state assimilation correction is performed on the twin drone nodes and twin links that contribute significantly to the deviation, resulting in the corrected twin node state vector and the corrected twin link quality: ， ， in, For drone nodes In the current time slice The corrected twin node state vector, For link In the current time slice The corrected twin link quality, superscript This represents the corrected twin domain. For drone nodes In the current time slice The node state assimilation coefficient, For link In the current time slice The link quality assimilation coefficient, and and The values ​​are all located in Within the range.

4. The method for optimizing UAV communication networks based on digital twins according to claim 1, characterized in that, Generating the predicted communication quality values ​​includes: based on the corrected twin communication state data, calculating the twin link propagation strength, twin interference coupling strength, twin route load delay, and twin channel occupancy strength, and calculating the predicted communication quality value for each twin link according to the following formula: ， in, For link In the current time slice The predicted quality of twin communication, For link In the current time slice The propagation strength of twin links, For link In the current time slice The twin interference coupling strength, For link In the current time slice Twin routing load latency, For link In the current time slice Twin channel occupancy strength, , , and These are the quality weights for link propagation strength, interference coupling strength, routing load delay, and channel occupancy strength, respectively. It generates prediction confidence based on historical strategy execution virtual-real bias, current virtual-real synchronization bias, and the number of effective feedback samples: ， in, For link In the current time slice The prediction confidence level This is a range cutoff function used to restrict the value within parentheses to a certain range. Within the interval, For link In the previous time slot The strategy execution deviates from the real one. For the current time slice The virtual-real synchronization deviation, For link In the current time slice The number of valid feedback samples accumulated previously, , and These are the confidence weights for the historical policy execution virtual-to-real bias, virtual-to-real synchronization bias, and the reciprocal of the number of valid feedback samples, respectively. In the initial time slice, if there is no policy execution virtual-to-real bias from the previous time slice, then... Take 0, if the link If no valid feedback samples have been accumulated, then Set it to 0 so that the prediction confidence level is determined by the current virtual-real synchronization deviation and the preset confidence weight.

5. The UAV communication network optimization method based on digital twin according to claim 4, characterized in that, The bottleneck link set consists of twin links whose predicted communication quality is lower than the quality threshold, twin links whose predicted communication quality decreases more than the decrease threshold between adjacent time slots, and twin links carrying high-priority communication tasks with routing loads higher than the load threshold. Specifically, in the initial time slot where there is no predicted communication quality value for the previous time slot, the judgment of the decrease in predicted communication quality between adjacent time slots is not enabled; the bottleneck link set is determined solely based on the current predicted communication quality value, communication task priority, and routing load. From the time slot where there is a predicted communication quality value for the previous time slot, the judgment of the decrease in predicted communication quality between adjacent time slots is enabled. After the bottleneck link set is determined, the one-hop adjacent links of the UAV nodes that share the bottleneck link set are further determined. The candidate communication optimization strategy is generated only for the bottleneck link set and its one-hop adjacent links. When link Prediction confidence When the confidence threshold is lower than that, the link is marked as a low-confidence link, and the transmit power adjustment range, route reconstruction range, and relay node replacement range in the candidate communication optimization strategies corresponding to the link are restricted.

6. The method for optimizing UAV communication networks based on digital twins according to claim 1, characterized in that, The generation boundary of the candidate communication optimization strategy is determined by the strategy boundary correction result after the online backfeeding was completed in the previous time slice. The candidate communication optimization strategy boundary includes transmit power adjustment boundary, channel switching boundary, relay node selection boundary and route reconstruction boundary. When the link propagation strength of the bottleneck link is lower than the propagation strength threshold and the interference coupling strength is lower than the interference strength threshold, a transmit power adjustment action is generated; when the channel occupancy strength or interference coupling strength of the bottleneck link is higher than the corresponding threshold, a channel switching action is generated; when the direct communication quality between the nodes at both ends of the bottleneck link is lower than the quality threshold and the surrounding adjacent nodes meet the relay quality conditions, a relay node selection action is generated; when the bottleneck link is located in the congested area of ​​the current routing path or the routing load delay is higher than the delay threshold, a route reconstruction action is generated.

7. The UAV communication network optimization method based on digital twin according to claim 1, characterized in that, The pre-simulation of the candidate communication optimization strategies includes: executing each candidate communication optimization strategy in the digital twin space to obtain the corresponding twin-predicted communication quality gain, twin-predicted energy consumption cost, twin-link switching cost, and twin-routing disturbance cost, and calculating the strategy pre-simulation benefit value. ， in, Optimize candidate communication strategies In the current time slice The strategy preview return value, Number the candidate communication optimization strategy. Optimize candidate communication strategies In the current time slice Twin prediction of communication quality gain, Optimize candidate communication strategies In the current time slice Twin prediction of energy consumption cost Optimize candidate communication strategies In the current time slice The cost of twin link switching, Optimize candidate communication strategies In the current time slice The cost of twin routing perturbation, , , and These are the revenue weights for predicting communication quality gain, predicting energy consumption cost, link switching cost, and routing disturbance cost, respectively. After each candidate communication optimization strategy completes its pre-performance, its twin predicted communication quality change, predicted energy consumption cost, link switching cost, routing disturbance cost, and strategy pre-performance benefit value are written into the strategy pre-performance record, and a pre-performance version identifier is set.

8. The UAV communication network optimization method based on digital twin according to claim 7, characterized in that, The target communication optimization strategy is determined by eliminating candidate communication optimization strategies that cause the UAV node's transmit power to exceed the transmit power adjustment boundary, the number of channel switching to exceed the channel switching boundary, the relay node's load to exceed the relay node selection boundary, or the number of route hops to exceed the route reconstruction boundary. Among the remaining candidate communication optimization strategies, the candidate communication optimization strategy with the highest strategy pre-show benefit value is selected first; when the strategy pre-show benefit values ​​of two candidate communication optimization strategies are the same or the difference is lower than the preset difference threshold, the candidate communication optimization strategy that covers more bottleneck links is selected as the target communication optimization strategy. If the target communication optimization strategy changes the main route, relay node, or target channel carrying high-priority communication tasks, it is determined whether the forced switching conditions are met before execution; if the forced switching conditions are not met, the target communication optimization strategy is downgraded to a backup strategy.

9. The method for optimizing UAV communication networks based on digital twins according to claim 1, characterized in that, The virtual-to-real deviation of the strategy execution is calculated according to the bottleneck link, satisfying the following: ， in, For link In the current time slice The strategy execution deviates from the real one. After the target communication optimization strategy is executed, the link will be optimized. In the current time slice The measured change in communication quality, superscript This represents the actual measured results after execution. The predicted changes in communication quality, simulated and saved in the digital twin space before the target communication optimization strategy is issued and executed, are indicated by the superscript. This represents the results of the twin simulation saved before execution; when This indicates that the measured improvement after implementing the target communication optimization strategy is lower than the improvement predicted by the twins, thus improving the link... The corresponding propagation loss correction parameters and interference coupling correction parameters are used to reduce the link quality prediction confidence parameters; when This indicates that the measured improvement after implementing the target communication optimization strategy is higher than the improvement predicted by the twins, thus reducing link... The corresponding propagation loss correction parameters or improved link quality prediction confidence parameters; The propagation loss correction parameter, interference coupling correction parameter, link quality prediction confidence parameter, and policy boundary correction parameter are generated only by the policy execution virtual-real deviation and not by the virtual-real synchronization deviation. When the virtual-to-real deviation of the strategy execution remains the same sign and the absolute value increases for at least two consecutive time slices, the transmit power adjustment boundary and route reconstruction boundary of the corresponding bottleneck link are narrowed, and the quality threshold for candidate communication optimization strategies to enter the pre-play is increased; when the absolute value of the virtual-to-real deviation of the strategy execution decreases or falls below the convergence threshold for at least two consecutive time slices, the boundary of the candidate communication optimization strategy for the corresponding bottleneck link is maintained or relaxed. Once local simulation and local pre-play are triggered within the same time slice, no new policy execution virtual-real deviation will be generated unless new changes in communication quality after actual execution are re-acquired.

10. A UAV communication network optimization system based on digital twins, characterized in that, include: The measured data acquisition and prior knowledge generation module is used to acquire measured communication status data of the UAV communication network in time slices and generate data in the current time slice. For the initial time slice, initial prior twin communication state data is generated using preset initial communication simulation parameters, initial model parameters obtained from offline training, and initial candidate communication optimization strategy boundaries; in the current time slice... When it is not the initial time slice, read the previous time slice. The communication simulation parameters and candidate communication optimization strategy boundaries have been completed after online feedback, and the current time slice is generated in the digital twin space. Prior twin communication state data; The state assimilation and synchronization module is used to pair the measured communication state data with the prior twin communication state data based on timestamps, node identifiers, and link topology constraints, calculate the virtual-real synchronization deviation, and perform state assimilation correction only on the prior twin communication state data based on the virtual-real synchronization deviation to obtain the corrected twin communication state data. The virtual-real synchronization deviation is not used to generate propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and policy boundary correction parameters. The communication quality simulation module is used to perform joint simulation of link propagation, interference coupling, and routing load based on the corrected twin communication state data, and generate communication quality prediction values, prediction confidence, and bottleneck link sets. The strategy pre-executing module is used to pre-execute candidate communication optimization strategies, including at least two actions among transmit power adjustment, channel switching, relay node selection and route reconstruction, focusing only on the bottleneck link set, determine the target communication optimization strategy and issue it for execution, and save the twin prediction of communication quality changes before the target communication optimization strategy is issued and executed. The deviation feedback module is used to collect the measured changes in communication quality after the target communication optimization strategy is executed. It compares these measured changes with the twin-predicted changes in communication quality saved before the target communication optimization strategy was issued and executed, generating a virtual-to-real deviation in strategy execution. Based on this virtual-to-real deviation, it generates propagation loss correction parameters, interference coupling correction parameters, link quality prediction confidence parameters, and strategy boundary correction parameters to synchronously update the next time slice. Communication simulation parameters and candidate communication optimization strategy boundaries; The local re-simulation module is used to re-perform local simulation and local pre-simulation only for UAV nodes, adjacent links, and occupied channels associated with the changed bottleneck link set when the absolute value of the virtual-to-real deviation of the strategy execution of at least one bottleneck link exceeds a preset deviation threshold, or when the bottleneck link set changes. The number of times the local simulation and local pre-simulation are executed in the same time slice does not exceed a preset upper limit. The communication simulation parameters and candidate communication optimization strategy boundaries fed back to the measured acquisition and prior generation module by the execution deviation feedback module only take effect in the next time slice.

Images