Distributed multi-uav consensus formation control method for communication delay

Abstract

The present application relates to the technical field of multi-rotor unmanned aerial vehicle cluster cooperative control, and particularly relates to a distributed multi-unmanned aerial vehicle consistency formation control method for coping with communication delay, each unmanned aerial vehicle performing the following steps: time synchronization with neighboring machines and periodically obtaining communication quality measurement values to generate a link quality index; classifying communication links based on the link quality index, constructing an information pair composed of an expected trajectory sequence and a deviation feedback sequence in the same predicted time domain, and broadcasting; receiving neighboring machine information pairs, and when the link is degraded or interrupted, superimposing the deviation feedback sequence on the expected trajectory sequence corresponding to the absolute time point according to the time stamp to generate a closed-loop correction estimate; adjusting consistency control law parameters according to real-time link classification results, enabling or disabling compensation control based on the closed-loop correction estimate to generate flight control instructions; when all neighboring machine links are interrupted, generating a dynamic safety window based on the inertial measurement unit drift rate and the flight state, and performing cooperative flight within the window based on the last closed-loop correction estimate.

Description

Technical Field

[0001] This invention relates to the field of multi-rotor unmanned aerial vehicle (UAV) swarm collaborative control technology, and in particular to a distributed multi-UAV consensus formation control method for addressing communication delays. Background Technology

[0002] Distributed multi-UAV consensus formation control is a key technology for UAV swarms to perform tasks such as collaborative inspection, mapping, and emergency communication networking. Under this control architecture, each UAV autonomously adjusts its flight state by interacting only with its communication neighbors, enabling the entire swarm to form and maintain the desired formation configuration without a central node. However, in real-world operating environments, wireless communication links between UAVs often experience random delays and intermittent interruptions due to physical obstructions such as buildings, trees, and terrain undulations, as well as electromagnetic clutter interference. Communication delays typically fluctuate between 20ms and 150ms, and in severe cases, the links become completely blocked. This degradation in communication quality undermines the stability of distributed consensus formation control, leading to formation configuration oscillations, deviations, or even divergence and disintegration.

[0003] Existing technical solutions for addressing communication latency mainly fall into two categories: one is to pre-consider a fixed upper bound on latency during the consensus protocol design phase and design a fixed-parameter controller using robust control theory. However, the performance of this method degrades significantly when latency changes dynamically. The other category uses state predictors to estimate the current state of neighboring machines to compensate for information transmission delays. However, these solutions are mostly open-loop predictions, failing to consider prediction biases caused by external disturbances such as gusts and inter-machine airflow coupling. Errors accumulate over time, resulting in limited compensation accuracy. Furthermore, when communication links are interrupted for extended periods, existing solutions generally lack proactive and reliable topology recovery and security degradation mechanisms, making it easy for individual machines to break away from the formation and lose their collaborative capabilities. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a distributed multi-UAV consensus formation control method to address communication delays. It aims to improve the problem that existing distributed UAV consensus formation control methods suffer from slow or even unstable formation convergence due to the accumulation of open-loop prediction errors under conditions of random jumps in communication delays and external disturbances.

[0005] This invention provides the following technical solution: a distributed multi-UAV consensus formation control method to address communication latency, wherein each UAV performs the following steps: S1. Synchronize with the execution time of neighboring machines in the communication network, and periodically acquire communication quality measurement values ​​with each neighboring machine, and generate a link quality index according to preset calculation rules; S2. Based on the link quality index, classify each communication link as good, downgraded, or interrupted, construct an information pair consisting of the expected trajectory sequence and the deviation feedback sequence generated for the same prediction time domain, and broadcast it at a fixed frequency; S3. Receive the information pairs broadcast by each neighboring machine. When it is determined that the communication link with a target neighboring machine is in a degraded or interrupted state, the received deviation feedback sequence is superimposed on the corresponding absolute time point of the expected trajectory sequence according to the timestamp to generate a closed-loop correction estimate of the current state of the target neighboring machine. S4. Based on the real-time hierarchical results of the communication links between the aircraft and its neighboring aircraft, adjust the control parameters of the formation consistency control law, and enable or disable compensation control based on the closed-loop correction estimate to generate the aircraft's flight control commands. S5. When it is determined that the communication link with all neighboring aircraft is interrupted, a dynamic safety window is generated based on the drift rate output in real time by the local inertial measurement unit and the current flight status, and cooperative flight is carried out within the dynamic safety window based on the last received closed-loop correction estimate. S6. When the duration of the communication link interruption with all neighboring machines exceeds the dynamic security window, execute the preset step-by-step security degradation strategy.

[0006] The present invention has the following beneficial effects: 1. In this invention, by constructing an information pair consisting of an expected trajectory sequence and a deviation feedback sequence generated for the same prediction time domain, and superimposing the deviation feedback sequence onto the corresponding absolute time point of the expected trajectory sequence at the receiving end based on the timestamp to generate a closed-loop correction estimate, the error accumulation of the existing open-loop prediction scheme under external disturbances is effectively suppressed, and the accuracy and robustness of formation consistency convergence under the condition of random jump in communication delay are significantly improved.

[0007] 2. In this invention, a normalized link quality index is generated through an environment adaptive mapping model, and the classification threshold is dynamically determined based on the statistical distribution characteristics of the link quality index. This enables the link status classification to automatically adapt to different communication environments such as open areas, forest areas, and urban canyons, avoiding the problem of inaccurate classification under fixed thresholds in different scenarios and improving the scenario generalization ability of the solution.

[0008] 3. In this invention, by determining the coupling coefficient and speed damping coefficient of the consistency control law on the continuous mapping function based on the specific value of the link quality index, a smooth transition of control parameters as communication quality degrades is achieved, eliminating the sudden changes in control quantity and transient oscillations in formation caused by traditional segmented parameter switching at the hierarchical boundary, and improving the stability of formation flight. Attached Figure Description

[0009] Figure 1This is a flowchart illustrating the distributed multi-UAV consensus formation control method for addressing communication delays proposed in this invention. Figure 2 This is a schematic diagram of the link quality perception and classification process of the distributed multi-UAV consensus formation control method for dealing with communication delays proposed in this invention. Figure 3 This is a schematic diagram of the closed-loop correction estimation and compensation control process of the distributed multi-UAV consensus formation control method for dealing with communication delay proposed in this invention. Figure 4 This diagram illustrates the dynamic safety window for chain breakage and the step-by-step degradation process of the distributed multi-UAV consensus formation control method for addressing communication delays proposed in this invention. Detailed Implementation

[0010] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] Example 1: In a first embodiment of the present invention, the present invention provides a distributed multi-UAV consensus formation control method to address communication latency, such as... Figures 1-4 As shown, it includes: S1. Synchronize with the execution time of neighboring machines in the communication network, periodically acquire communication quality measurement values ​​between each neighboring machine, and generate a link quality index according to preset calculation rules.

[0012] Specifically, after each drone powers on, it first synchronizes its time with neighboring drones in the communication network. Time synchronization employs a two-stage mechanism: when GNSS signals are valid, each drone uses the second pulse signal output by its GNSS receiver as a global clock reference, running a lightweight, precise time synchronization protocol to control the network-wide clock synchronization error to the microsecond level. When a drone enters a GNSS signal obstruction area, causing the second pulse signal to be lost, the drone automatically switches to a consensus-based distributed time synchronization algorithm. It broadcasts its clock quality assessment value to neighboring drones within its communication range, selects the neighbor with the best clock quality in the current network as a reference node, and estimates and corrects the clock offset through bidirectional message exchange to maintain network-wide clock synchronization. The clock quality assessment value is calculated as follows: the average frequency drift of the drone's clock over the most recent 10 synchronization cycles is weighted by the current GNSS signal lock state. The clock quality assessment value is set to the highest value when the GNSS signal is locked, and gradually decreases as the lock duration increases after the GNSS signal is lost. The overall weighting uses a linear weighting method, and the weighting formula is as follows: the clock quality assessment value equals the product of the GNSS lock-in coefficient and the lock-in weight, plus the product of the normalized frequency drift and the frequency drift weight. Specifically, the GNSS lock-in coefficient is 1 when the GNSS signal is locked and 0 when it is unlocked; the lock-in weight is 0.6; the normalized frequency drift is 1 minus the ratio of the average frequency drift to the preset maximum allowable drift, with the upper limit of this ratio truncated to 1; and the frequency drift weight is 0.4. The preset maximum allowable drift is 10 microseconds per second.

[0013] After time synchronization, each UAV periodically measures its bidirectional communication latency, packet loss rate, and received signal strength with neighboring UAVs every 100 milliseconds as communication quality measurements. The bidirectional communication latency is measured as follows: the UAV sends a probe packet carrying a transmission timestamp to the target neighbor; upon receiving the packet, the target neighbor immediately sends back a response packet carrying both a reception timestamp and a transmission timestamp. The UAV calculates the round-trip latency based on the timestamp in the response packet and its local reception time, and takes half of this as the estimated bidirectional communication latency for that period. The packet loss rate is measured by calculating the ratio of the number of data packets actually received by the UAV from the target neighbor to the number of data packets that should have been received within a preset sliding window. The packet loss rate is obtained by subtracting this ratio from 1. The received signal strength is measured by directly reading the received signal strength indicator value from the data radio's receive link indicator register.

[0014] The bidirectional communication delay, packet loss rate, and received signal strength obtained from the above measurements are input into an environment adaptive mapping model to generate a normalized link quality index. The construction and calculation method of this environment adaptive mapping model is as follows. First, to eliminate the difference in dimensions of each measurement value, each measurement value is standardized. Let the bidirectional communication delay obtained in the current period be... The packet loss rate is The received signal strength is Calculate the standardized values ​​for each of the three measurements: ; ; ; in, , , These are the average values ​​of bidirectional communication latency, packet loss rate, and received signal strength within a preset historical sliding window, respectively. , , These represent the corresponding standard deviations. The preset duration of the historical sliding window is 30 seconds, and the number of data points within the window is 300.

[0015] Then, the three standardized values ​​are input into a weighted fusion function to calculate the initial link quality score: ; in, The Sigmoid function is defined as follows: ; in, , , These are weighting coefficients, with values ​​of 0.4, 0.35, and 0.25 respectively. Since higher received signal strength indicates better link quality, and its direction is opposite to delay and packet loss rate, therefore... The sigmoid function value is negativeed before being weighted.

[0016] The initial link quality score is mapped to the 0-1 range using another Sigmoid function to generate the normalized link quality index (LQI): ; Where k is the mapping steepness parameter, with a value of 10. When the initial link quality score... When the LQI is 0.5, the LQI is 0.5; when When the LQI is above 0.5, it approaches 1; when When it is below 0.5, the LQI approaches 0.

[0017] In the aforementioned environment adaptive mapping model, the mean and standard deviation of each measurement value are continuously updated online through a sliding window to adapt to changes in the statistical characteristics of different communication environments. When entering a new communication environment, the statistical data within the sliding window are gradually updated, and the model output is adaptively adjusted accordingly, thereby achieving a normalized representation of link quality under different scenarios.

[0018] At this point, each drone obtains a Link Quality Index (LQI) with each neighboring drone. This LQI will be used for link classification, control parameter adjustment, and compensation control decisions in subsequent steps.

[0019] S2. Based on the link quality index, each communication link is classified as good, degraded, or interrupted. An information pair consisting of the expected trajectory sequence and the deviation feedback sequence generated for the same prediction time domain is constructed and broadcast at a fixed frequency.

[0020] Specifically, after each UAV obtains the Link Quality Index (LQI) with its neighboring UAVs in S1, the following hierarchical process is performed on each communication link: The statistical distribution characteristics of the link quality index between the UAV and its neighboring UAVs over a historical period are obtained. In this embodiment, the historical period is defined as the LQI sampling values ​​for the most recent 300 periods, corresponding to a 30-second sliding window. The statistical distribution characteristics include the mean LQI within this window. and standard deviation The first threshold is dynamically determined based on the above statistical distribution characteristics. Second threshold ,and The threshold is calculated as follows: ; ; in, and This is a preset offset coefficient. In this embodiment... The value is 0.5. The value is 0.5.

[0021] The link quality index generated in the current period With the first threshold Second threshold Comparisons are made to classify the corresponding communication links as good, degraded, or interrupted: when When, the corresponding communication link is classified as good; when When, the corresponding communication link is classified as downgraded; when When this happens, the corresponding communication link will be classified as an interruption.

[0022] The aforementioned dynamic threshold mechanism enables the classification boundary to adaptively adjust to the overall LQI distribution of the current communication environment. In open environments, the mean LQI is higher and its fluctuations are smaller, resulting in a correspondingly higher threshold and stricter classification standards. In heavily obstructed environments such as forest areas or urban canyons, the mean LQI decreases and its fluctuations increase, leading to a corresponding reduction in the threshold to avoid frequent misjudgments of interruptions. The sliding window is continuously updated, and the mean and standard deviation are recalculated periodically to ensure that the threshold responds promptly to changes in the communication environment.

[0023] Each drone constructs and broadcasts information pairs at a fixed frequency of 10 Hz, meaning a broadcast period of 100 milliseconds, consistent with the communication quality measurement period in S1. The specific process for constructing information pairs within each period is as follows: The first step is to generate the expected trajectory sequence for this cycle. A lightweight distributed model predictive controller is run on the local machine. Using its current state as initial conditions, and satisfying dynamic constraints and formation configuration constraints, the optimal control sequence and the corresponding predicted state sequence within the future time window are solved. The length of the future time window is... Where N is the number of prediction steps, and its value is 10. This is the single-step prediction step size, set to 100 milliseconds. The generated expected trajectory sequence contains data from... arrive There are N discrete time points for the predicted state, and each predicted state includes three-dimensional position and three-dimensional velocity. Let the current time be t, and the expected trajectory sequence generated in this cycle is represented as: ; in, Let be the predicted state vector for the i-th future step.

[0024] The second step is to generate the current cycle deviation feedback sequence. The deviation feedback sequence is the error sequence between the expected trajectory sequence generated in the previous cycle (time t-T) and the actual flight trajectory within the corresponding time period, where T is the broadcast period, which is 100 milliseconds. Let the expected trajectory sequence generated in the previous cycle at time t-T be... Its coverage time window is The aircraft records the actual flight status at each moment within this time window through its onboard integrated navigation system, forming an actual trajectory sequence. The deviation feedback sequence is obtained by subtracting the actual trajectory sequence from the expected trajectory sequence of the previous period at each corresponding absolute time point: ; in, , which is the state error vector at the corresponding absolute time point, including position error and velocity error.

[0025] It should be noted that the expected trajectory sequence generated in this period Although the time window covered by the deviation feedback sequence EE is not exactly the same, each error term in the deviation feedback sequence The timestamps strictly correspond to the prediction points at the same absolute time in the previous cycle's prediction sequence. When using the data, the receiver overlays the corresponding items in the deviation sequence onto the corresponding prediction points in the expected trajectory sequence based on the absolute timestamps. Here, "correspondence" refers to the consistency of absolute time coordinates, thus forming a closed-loop correction with strict time alignment.

[0026] The third step is to generate the expected trajectory sequence for this period. The bias feedback sequence E is combined with the bias feedback sequence E to form an information pair, which is then broadcast at a fixed frequency. The broadcast data packet format is as follows: frame header, local identifier, timestamp, expected trajectory sequence, bias feedback sequence, and frame tail checksum. Each data packet carries an absolute timestamp for time alignment by the receiver.

[0027] The construction and broadcasting of information pairs are performed continuously at a fixed frequency, regardless of the communication link quality. The local machine generates and broadcasts information pairs periodically. When some neighboring machines are unable to receive the information pairs broadcast by the local machine due to link interruption, the local machine does not change its broadcasting behavior. Once the link is restored, the neighboring machines can directly obtain the latest information pairs. For the local machine, even if the link with a neighboring machine is interrupted, it still receives information pairs broadcast by other neighboring machines and maintains cooperative flight using the last received and closed-loop corrected state information.

[0028] During each broadcast cycle, each UAV appends its currently maintained one-hop neighbor list to the broadcast information pair. This list contains the identifiers of all its directly communicating neighbors and their corresponding link quality indices. Upon receiving a neighbor's broadcast one-hop neighbor list, the UAV merges it with its own one-hop neighbor list to construct a two-hop neighbor topology map centered on itself, and labels the link quality indices on each path. The two-hop neighbor topology map is fully updated once per broadcast cycle. Based on this topology map, the UAV selects all indirect neighbors reachable by a two-hop path. For each candidate backup neighbor, a handover cost is calculated. The handover cost is equal to the sum of the link quality index attenuation values ​​of each link segment on the path from the UAV to the candidate backup neighbor, where the link quality index attenuation value is defined as 1 minus the link quality index of that link segment. The candidate neighbor with the lowest handover cost is selected as the optimal backup neighbor, and its identifier is stored in the UAV's backup neighbor list for rapid topology recovery after all communication links are interrupted. When the number of the UAV's one-hop neighbors exceeds a preset threshold, the update cycle of the two-hop neighbor topology map is adaptively extended by a factor equal to the number of one-hop neighbors divided by the preset threshold and rounded up. The preset threshold value is 8. During broadcast cycles that have not yet reached the update cycle, the local machine only broadcasts the changed parts of the one-hop neighbor list, i.e., incremental updates, to reduce communication overhead.

[0029] S3. Receive the information pairs broadcast by each neighboring machine. When it is determined that the communication link with a target neighboring machine is in a degraded or interrupted state, the received deviation feedback sequence is superimposed on the corresponding absolute time point of the expected trajectory sequence according to the timestamp to generate a closed-loop correction estimate of the current state of the target neighboring machine.

[0030] Specifically, each UAV continuously listens for and receives information pairs broadcast by its neighboring UAVs at a fixed frequency during each control cycle. Each information pair contains a predicted trajectory sequence and a deviation feedback sequence, and each data packet carries the sender's absolute timestamp. The UAV stores the received information pairs from each neighboring UAV in a local buffer according to the neighboring UAV identifier and absolute timestamp. The buffer maintains an independent circular queue for each neighboring UAV, with a queue length of 20 data packets, covering the historical data of the most recent 2 seconds. When a new information pair is received, if the queue of the corresponding neighboring UAV is full, the oldest data packet is overwritten.

[0031] After receiving and caching information pairs, the local machine determines the real-time classification result of the communication links between itself and each neighboring machine based on the link quality index generated in S1. This real-time classification result has been updated in S2. The local machine traverses all neighboring machines with established communication links and performs the following operations for each target neighboring machine. When the communication link with a target neighboring machine is classified as good, the local machine directly extracts the expected trajectory sequence from the latest information pair received by the target neighboring machine, and combines it with the absolute timestamp of the information pair to obtain the predicted state corresponding to the current moment through interpolation. This predicted state is used as the state estimate of the target neighboring machine and directly participates in the local machine's consistency control law calculation, without performing the closed-loop correction estimate generation operation. When the communication link with a target neighboring machine is classified as degraded or interrupted, the local machine extracts the expected trajectory sequence and deviation feedback sequence from the information pair broadcast by the target neighboring machine and cached locally based on the timestamp, and performs the closed-loop correction estimate generation operation. The specific generation method of the closed-loop correction estimate is as follows.

[0032] Suppose that the local machine performs a state estimation on the target neighbor j at the current time t. The local machine selects the most recently received information pair from the buffer of neighbor j, and assumes that the absolute timestamp of this information pair is . ,in The difference between the two is the communication delay. This information includes the time of neighbor j. The generated expected trajectory sequence and deviation feedback sequence.

[0033] Extract the expected trajectory sequence from this information pair. Let the time window covered by the expected trajectory sequence broadcast by neighboring machine j be... It contains N prediction points. For each prediction point, extract its predicted state vector at the corresponding absolute time: ; in, Indicates that neighboring machine j at time... Generated for future moments The predicted state includes three-dimensional position and three-dimensional velocity.

[0034] Extract the deviation feedback sequence from this information pair. The deviation feedback sequence is the error sequence between the expected trajectory sequence generated by the neighboring aircraft j in the previous broadcast cycle and its actual flight trajectory within the corresponding time period. Let the time window covered by the deviation feedback sequence be... The sequence contains the state error vector at each corresponding absolute time point: ; in, , indicating that the neighboring machine j at absolute time The error between the actual state and the predicted state at that moment in the previous period. T is the broadcast period, which is 100 milliseconds.

[0035] The deviation feedback sequence is superimposed onto the absolute time point corresponding to the timestamp in the expected trajectory sequence. The specific superposition method is as follows: for each predicted time point in the expected trajectory sequence... Find error terms with the same absolute time in the deviation feedback sequence. The error term is then added to the corresponding predicted state to obtain the state estimate after closed-loop correction: ; in, For the neighboring machine j at time after closed-loop correction The state estimate is given by i = 1, 2, ..., N. It should be noted that this is only true if the deviation feedback sequence contains a state that corresponds to the prediction time. The above superposition can only be performed when the absolute time coordinates are consistent and the error term for the deviation calculation time has been passed. For future times in the expected trajectory sequence that are ahead of the current time t and for which there is no corresponding deviation feedback, the superposition operation is not performed, and the predicted value in the expected trajectory sequence is directly used as the state estimate.

[0036] Based on the closed-loop calibrated state estimation sequence, this machine obtains the current state of the target neighbor at time t through interpolation, and generates a closed-loop calibrated estimate of the target neighbor's current state. The interpolation method is as follows: in the closed-loop calibrated state estimation sequence, the two estimation points that are closest in time to the current time t are selected, and linear interpolation is used to calculate the state estimate at the current time. Let the timestamps of the two closest estimation points be... and ,satisfy The corresponding state estimates are as follows: and Then the closed-loop correction estimate of the target neighbor j at the current time is: ; In the aforementioned closed-loop correction estimation process, the introduction of the deviation feedback sequence enables the local aircraft to utilize feedback from neighboring aircraft regarding their prediction errors to correct the open-loop prediction results. When a neighboring aircraft is subjected to external disturbances such as gusts or inter-aircraft airflow coupling, its actual flight trajectory will deviate from the expected trajectory sequence. This deviation will be accurately reflected in the deviation feedback sequence of the next cycle and compensated for in the local aircraft's closed-loop correction estimation, thereby effectively suppressing the accumulation of open-loop prediction errors.

[0037] At this point, the machine obtains closed-loop correction estimates of the current state of all target neighboring machines. These closed-loop correction estimates will be used for the consistency control law solution and compensation control decision in S4.

[0038] S4. Based on the real-time hierarchical results of the communication links between the aircraft and its neighboring aircraft, adjust the control parameters of the formation consistency control law, and enable or disable compensation control based on closed-loop correction estimation to generate the aircraft's flight control commands.

[0039] Specifically, within each control cycle, each UAV first acquires the real-time classification results of the communication links between itself and its neighboring UAVs, as well as the specific values ​​of the Link Quality Index (LQI) for each link. The real-time classification results have been updated in S2, including three states: good, degraded, and interrupted. The specific value of the Link Quality Index (LQI) is the normalized value output by the environment adaptive mapping model in S1, ranging from 0 to 1. Based on the specific value of the Link Quality Index, the coupling coefficient and speed damping coefficient of the consistency control law are determined on a preset continuous mapping function. In this embodiment, the consistency control law adopts a second-order consistency protocol in the following form. Let the UAV's number be i, and its desired formation position be... The actual location is The speed is The expected relative positional deviation between this machine and its neighboring machine j is: The actual relative position deviation is The consistency control law acceleration command of machine i. for: ; in, Let i be the set of its communication neighbors; To establish a link quality index with neighboring machine j The coupling coefficient determines the strength of the contribution of the position deviation to the control quantity; To establish a link quality index with neighboring machine j The velocity damping coefficient determines the strength of the contribution of the velocity difference to the control quantity.

[0040] Coupling coefficient and velocity damping coefficient All are based on the link quality index through a continuous mapping function. The specific values ​​are determined, rather than selected in segments based on the classification results. The continuous mapping function causes the coupling coefficient to decrease continuously as the link quality index decreases, and the speed damping coefficient to increase continuously as the link quality index decreases.

[0041] In this embodiment, the coupling coefficient The continuous mapping function takes the form of a linear mapping: ; in, This represents the maximum coupling coefficient, which is 2.0. The minimum coupling coefficient is 0.2. hour, The coupling strength corresponds to the optimal link quality; when hour, This corresponds to the minimum coupling strength when the link quality is at its worst.

[0042] Velocity damping coefficient The continuous mapping function takes the following form: ; in, This represents the maximum velocity damping coefficient, which is 3.0. This represents the minimum velocity damping coefficient, set to 1.0. When... hour, When the link quality is optimal, only a small speed damping is needed to ensure formation stability; when When it decreases, Continuously increasing the speed damping introduces stronger velocity damping to suppress formation oscillations caused by increased communication delay; when hour, It provides maximum speed damping.

[0043] The aforementioned continuous mapping mechanism ensures that when the link quality index fluctuates slightly near the grade boundary, the control parameters only undergo small, continuous changes, avoiding abrupt changes in control quantities caused by segmented parameter switching at the grade boundary.

[0044] When the communication link with the target neighbor is classified as interrupted, the coupling coefficient of the consistency control law corresponding to the target neighbor will be reduced. The position of the target neighbor is set to zero, and the neighbor no longer participates in the calculation of the position deviation term in the consensus control law. Simultaneously, the closed-loop correction estimate generated in S3 replaces the real-time state information of the target neighbor in the consensus control law solution. Specifically, in the consensus control law, the position of the target neighbor j is set to zero. and speed The position and velocity components are replaced with those in the closed-loop correction estimate so that the formation can still maintain coordination using the historical state information after the link is interrupted.

[0045] Within each control cycle, the aircraft determines whether to enable compensation control based on the real-time classification results of the communication links with each target neighbor. When the communication link classification with the target neighbor is good, compensation control is disabled. The aircraft directly generates its flight control commands based on the real-time status information broadcast by the target neighbor. The target neighbor's real-time status information extracts the expected trajectory sequence from its latest broadcast information pair and combines it with the absolute timestamp to obtain the current state estimate through interpolation. Under good link conditions, communication latency is low, and the accuracy of the real-time status information is sufficient to meet the requirements of consistent formation control, so compensation control is not required. When the communication link classification with the target neighbor is degraded or interrupted, compensation control is enabled. The aircraft no longer uses the target neighbor's real-time status information but instead uses the closed-loop correction estimate generated in S3 to replace the target neighbor's real-time status information to generate its flight control commands. The closed-loop correction estimate has corrected the open-loop prediction error through the deviation feedback sequence, providing a more accurate target neighbor state estimate under conditions of large communication latency or link interruption.

[0046] The activation and deactivation of compensation control are executed independently for each target neighbor. The aircraft may have a good link classification with some neighbors, thus disabling compensation control, while having a degraded or interrupted link classification with other neighbors, thus enabling compensation control. Both operate in parallel within the consistency control law. The real-time status information of neighbors with good links and the closed-loop correction estimates of neighbors with degraded or interrupted links jointly participate in the weighted summation of the consistency control law to generate comprehensive flight control commands.

[0047] Finally, the machine will calculate the acceleration command obtained from the consistency control law. As flight control commands, they are sent to the underlying flight controller. The underlying flight controller converts the acceleration commands into attitude control commands and motor speed commands, driving the UAV to perform the desired maneuvers and achieve formation keeping.

[0048] S5. When it is determined that the communication link with all neighboring aircraft is interrupted, a dynamic safety window is generated based on the drift rate output in real time by the local inertial measurement unit and the current flight status, and cooperative flight is carried out within the dynamic safety window based on the last received closed-loop correction estimate.

[0049] Specifically, within each control cycle, the local machine determines whether all communication links with neighboring machines are interrupted based on the real-time classification results of each communication link maintained in S2. When at least one communication link is not interrupted, it indicates that the local machine can still obtain neighboring machine information through that link, and the dynamic security window generation process is not triggered. When it is determined that all communication links with neighboring machines are interrupted, the following process is triggered.

[0050] The accelerometer zero-bias drift rate, output in real time by the local inertial measurement unit (IMU), is obtained. The accelerometer zero-bias drift rate characterizes the rate at which the accelerometer output changes slowly over time when the IMU is stationary, and is an important indicator for measuring the dead reckoning accuracy of the IMU. In this embodiment, the accelerometer zero-bias drift rate... The data acquisition method is as follows: During the ground stationary phase before each UAV takeoff, the output data of the inertial measurement unit accelerometer is collected within 30 seconds, and the linear fitting slope of its triaxial composite acceleration is calculated as the initial calibration value of the accelerometer zero-bias drift rate for the current flight. During flight, when the UAV is in a near-zero acceleration maneuvering state such as hovering or uniform straight flight, the estimated value of this drift rate is continuously updated. The update uses a first-order low-pass filter. ,in, The instantaneous drift rate is calculated under the current zero-acceleration maneuver state. This is the drift rate estimate for the previous period. This is the filter coefficient, with a value of 0.1.

[0051] The current flight status of the aircraft is obtained, which includes at least the current ground speed. In this embodiment, the current ground speed... The speed is obtained directly from the speed output of the local integrated navigation system, including the northward speed component. and eastward velocity component The current composite value of the ground speed is Current flight status also includes the current rate of change of altitude. The vertical velocity is obtained from the vertical velocity output of the integrated navigation system. Based on the accelerometer zero-bias drift rate and the current flight state, the maximum duration for which the aircraft can maintain its formation relative position during an interruption is calculated, generating a dynamic safety window. The duration of the dynamic safety window is... The calculation method is as follows: ; in, The preset maximum safe time is set to 5.0 seconds, which is used to prevent the calculated safe window from being too long under extremely ideal conditions and exceeding the actual physical limits. The maximum allowable relative position deviation for the formation is set to 1.5 meters. This value is preset according to the formation mission requirements and represents the maximum offset distance between the machine and the desired formation position without disrupting the formation configuration. The zero-bias drift rate of the accelerometer is expressed in meters per second squared (m² / s). g is the gravitational acceleration, with a value of 9.81 m² / s. This represents the cumulative rate of velocity estimation error caused by accelerometer zero-bias drift. It is proportional to the square of time and, after unit conversion, characterizes the rate of increase of position error. This is the ground speed error coefficient, with a value of 0.1. This represents the rate at which the positional uncertainty introduced by the current ground speed increases over time. The height change rate error coefficient is set to 0.05. This represents the rate at which the additional location uncertainty introduced by changes in altitude increases over time.

[0052] The physical meaning of the above calculation method is as follows: After all communication links are interrupted, the aircraft relies on the inertial measurement unit to perform dead reckoning to maintain the relative position of the formation. The position error of the dead reckoning accumulates with the square of time. The higher the accelerometer zero-bias drift rate, the greater the current ground speed, and the more drastic the altitude change, the faster the position error accumulates and the shorter the safety window. When the calculated position error is... If the time window is less than 0.5 seconds, it indicates that the inertial measurement unit (IMU) is currently insufficient to support short-term autonomous cooperative flight. The aircraft skips the shadow mode and directly enters the S6 step-by-step safety degradation strategy. The determination of the 0.5-second minimum safety window threshold is based on the following: For consumer-grade MEMS IMUs, their accelerometer zero-bias drift rate is typically in the range of 0.01 m / s² to 0.05 m / s². Under typical cruise ground speeds of 6 to 10 m / s, the time for the dead reckoning position error to reach the maximum allowable relative position deviation of 1.5 meters is approximately 1.0 to 2.0 seconds. When the calculated dynamic safety window is less than 0.5 seconds, it indicates that the position error accumulation rate under the current flight state has exceeded the allowable value. Continuing to rely on dead reckoning to maintain the formation's relative position will lead to severe disruption of the formation configuration. Therefore, 0.5 seconds is chosen as the dividing threshold between entering shadow mode and directly implementing the degradation strategy.

[0053] After generating a dynamic safety window, the local machine performs cooperative flight within the dynamic safety window based on the last received closed-loop correction estimate. The specific process is as follows.

[0054] Obtain the closed-loop correction estimates of each neighboring machine received last before the communication link with all neighboring machines was interrupted. These closed-loop correction estimates are generated in S3 and stored in the local buffer. Let the time of communication link interruption be t. For each neighboring machine j, extract from the buffer... The last generated and stored closed-loop correction estimate is denoted as... Including neighboring machine j at time Estimated location and estimated speed Based on closed-loop correction estimation, the current state of each neighboring aircraft before the abort is determined. Combined with the real-time state changes output by the aircraft's inertial measurement unit, flight control commands for maintaining the relative position of the formation are generated through dead reckoning.

[0055] During the interruption of this machine actual location and actual speed Provided by the airborne integrated navigation system. From time 10:00 onwards, the aircraft uses its inertial measurement unit (IMU) as the core to perform dead reckoning and obtain changes in its state. Let the current time be 10:00. ,in The duration of the interruption, and The machine's position estimate is updated by using the acceleration and angular velocity measured by the inertial measurement unit (IMU) and the strapdown inertial navigation system (SINS) to calculate and perform calculations. and speed estimation .

[0056] For each neighboring machine, since its new state information cannot be obtained after the communication link is interrupted, this machine assumes that each neighboring machine maintains its pre-interruption motion trend during the interruption. The calculated state of neighboring machine j at time t is: ; ; That is, during the interruption, the neighboring machine j moves at a constant speed at the estimated speed before the interruption, and its position is linearly extrapolated.

[0057] Based on the dead reckoning status of the aircraft and the reckoning status of neighboring aircraft, the flight control command for the aircraft at time t is generated. The flight control command adopts the same second-order uniform control law as in S4. Taking the position control command as an example: ; in, This is the set of neighboring machines that maintained a communication link with this machine before the interruption. The consistency coupling coefficient in shadow mode is set to 0.5 in this embodiment, which is more conservative than the coupling coefficient during normal flight to reduce the impact of dead reckoning errors on formation configuration. The speed damping coefficient in shadow mode is set to 2.0, which is larger than that in normal flight to enhance formation damping and suppress oscillations. The desired relative positional deviation is consistent with the definition in S4.

[0058] Finally, the aforementioned flight control commands are continuously executed within the dynamic safety window to maintain formation configuration with neighboring aircraft. The duration of the machine update interruption within each control cycle is as follows: Recalculate dead reckoning and flight control commands until... Beyond the dynamic security window If the communication link is restored, the local machine will receive the broadcast information pairs from the neighboring machine again. When the communication link is restored, the local machine exits shadow mode and returns to the normal control flow of S3 and S4.

[0059] S6. When the duration of the communication link interruption with all neighboring machines exceeds the dynamic security window, execute the preset step-by-step security degradation strategy.

[0060] Specifically, in S5, the local machine continuously monitors the duration of communication link interruptions with all neighboring machines. .when Beyond the dynamic security window When this occurs, it indicates that the aircraft, relying solely on dead reckoning via inertial measurement unit, is unable to maintain the relative position deviation of the formation within acceptable limits. Continuing to execute shadow mode would lead to formation configuration disruption or collision risk. At this point, the aircraft triggers a tiered safety degradation strategy.

[0061] First, the formation mission information stored before the interruption is retrieved. This information is transmitted from the ground station to each UAV via a data link during the mission loading phase before UAV takeoff and is stored in the onboard mission management module. The formation mission information includes at least a mission type identifier to distinguish different mission categories. In this embodiment, the mission type identifier is an enumerated value. The first type of mission corresponds to inspection and mapping missions. These missions are characterized by formation configuration being crucial to mission completion quality, and there are usually no strict altitude restrictions over the mission area. The second type of mission corresponds to emergency networking missions. These missions are characterized by each UAV constituting a node in the communication network, and maintaining the current airspace position being crucial for network coverage continuity.

[0062] Based on the mission type indicated by the formation mission information, select the corresponding security degradation strategy. When the formation mission information indicates a Category 1 mission, enter the Level 1 security degradation strategy. The specific operation of the Level 1 security degradation strategy is as follows: control the local machine to automatically climb to obtain communication visibility, and send a reconnection request to the backup communication frequency band or backup neighboring machine. The local machine climbs at a preset rate. Climbing upwards from the current altitude, rate of ascent The value is 2.0 meters per second. During the climb, the machine continuously broadcasts reconnection request messages at a frequency of 1 Hz on a preset backup communication band via a data radio. The reconnection request message includes the machine's identifier, current position, current altitude, and formation identifier. Simultaneously, the machine queries the preset backup neighbor identifier from the two-hop neighbor topology maintained in S2 and sends a reconnection request to the backup neighbor. If a response message is successfully received from any neighbor during the climb, the reconnection is successful, the machine stops climbing, exits the degradation strategy, and re-executes the normal formation control process from S1 to S5 according to the newly established communication link. In this embodiment, the preset reconnection waiting time is 10.0 seconds, starting from the moment the first-level security degradation strategy is entered.

[0063] When the formation mission information indicates a Category II mission, the second-level safety degradation strategy is initiated. The specific operation of the second-level safety degradation strategy is as follows: the aircraft maintains its current altitude and hovers within a preset airspace, continuously broadcasting its position information. The aircraft maintains a preset hovering radius. It will perform circular hovering flight around the position where it entered the degrading strategy, with a hovering radius of... The hovering distance is set to 10.0 meters, and the angular velocity is set to 0.5 radians per second. During the hovering process, the drone continuously broadcasts its location information at a frequency of 2 Hz on a preset backup communication band. The location information includes the drone's identifier, current position, current altitude, formation identifier, and mission type identifier. Maintaining the current altitude for hovering aims to preserve the drone's communication coverage within the existing network airspace. Simultaneously, by continuously broadcasting location information, other drones that may re-enter the communication range can detect the drone and actively establish a connection. If a response message or active connection request is successfully received from any neighboring drone during the hovering process, the reconnection is successful, the drone stops hovering, exits the degradation strategy, and re-executes the normal formation control procedures from S1 to S5 based on the newly established communication link. The reconnection waiting time for the second-level security degradation strategy is also set to 10.0 seconds.

[0064] If reconnection fails within the preset reconnection waiting time (i.e., if no response message or active connection request is received from a neighboring aircraft after 10.0 seconds from the moment the first or second level of security degradation strategy is entered), the third level of security degradation strategy is entered. The specific operation of the third level of security degradation strategy is as follows: the aircraft is controlled to perform a return to home or hover. If the aircraft's current remaining battery power is higher than the preset safety margin for the return to home, the aircraft is controlled to return to the takeoff point along the preset return route and land; if the aircraft's current remaining battery power is insufficient to support the return to home, the aircraft is controlled to hover at the current horizontal position and continuously broadcast its position information and low battery alarm information at a frequency of 1 Hz, waiting for manual intervention from ground operators or communication restoration. The preset safety margin for the return to home is 1.2 times the power consumed during the return to home.

[0065] Throughout the execution of the tiered security degradation strategy, the local machine continuously monitors the received signals of each communication frequency band. Once a response or connection request from a neighboring machine is successfully received at any stage, the local machine immediately exits the current degradation strategy, restores the communication link with the neighboring machine, and reacquires the information pairs broadcast by the neighboring machine based on the new link, returning to the normal formation control flow from S1 to S5, thereby restoring the formation configuration and continuing the execution of the task.

[0066] Example 2: This embodiment uses a formation of 5 quadcopter UAVs to perform a forest strip mapping task as a specific application scenario, further illustrating the complete implementation process of the distributed multi-UAV consensus formation control method for dealing with communication delays provided by the present invention.

[0067] The five UAVs are numbered UAV-1, UAV-2, UAV-3, UAV-4, and UAV-5. Before the mission begins, the ground station sends formation mission information to each UAV via data transmission link, with the mission type identified as Category 1, corresponding to a mapping mission. Simultaneously, the desired formation configuration parameters are sent: a rhombus shape, with UAV-1 at the front, UAV-2 and UAV-3 at the left and right rear sides respectively, UAV-4 at the left rear outer edge, and UAV-5 at the right rear outer edge. The desired relative positional deviations between adjacent UAVs are also specified. Based on the pre-set geometric relationship of the rhombic configuration, each UAV completes the ground static calibration of the inertial measurement unit before takeoff to obtain the initial calibration value of the accelerometer zero-bias drift rate.

[0068] Five UAVs were powered on in an open take-off and landing area, and the data radio self-organizing network established a communication network. Each UAV executed step S1, using a precise time synchronization protocol based on GNSS second pulses to synchronize the clock across the entire network under valid GNSS signal conditions. Subsequently, bidirectional communication latency, packet loss rate, and received signal strength with each neighboring UAV were measured at 100-millisecond intervals, and a normalized Link Quality Index (LQI) was generated using an environment adaptive mapping model. Due to the unobstructed signal propagation in the open environment, the LQI of each link remained stable above 0.85.

[0069] Each UAV executes step S2, dynamically determining a first threshold and a second threshold based on the statistical distribution characteristics of LQI. Due to the high mean and small standard deviation of LQI in open environments, the first threshold is approximately 0.82, and the second threshold is approximately 0.78. Currently, the LQI of each link is greater than the first threshold, and all are classified as good. Simultaneously, each UAV constructs and broadcasts information pairs at a fixed frequency of 10 Hz. These information pairs include a predicted trajectory sequence for a 1-second time window generated by a lightweight distributed model predictive controller, and a feedback sequence of the deviation between the predicted trajectory sequence of the previous period and the actual flight trajectory within the corresponding time period.

[0070] Each UAV executes step S3, receiving and buffering information broadcast by neighboring UAVs. Since all links are classified as good, the generation of closed-loop correction estimates is not triggered in S3; the received real-time state information is directly used in the control calculation.

[0071] Each drone executes step S4. Since all links are classified as good, the coupling coefficient in S4 is low. Based on the specific value of LQI (0.85), it is calculated to be 1.73 using a continuous mapping function, and the velocity damping coefficient is... The calculated value is 1.3. Simultaneously, in S4, the link classification is determined to be good, compensation control is disabled, and flight control commands are directly generated based on the received real-time status information of neighboring aircraft. Driven by the consistency control law, each UAV quickly converges to the desired diamond formation configuration and flies along the predetermined mapping route. Approximately 30 seconds after takeoff, the formation configuration stabilizes, and the relative deviation of each aircraft from the desired formation position is less than 0.1 meters.

[0072] After flying along the flight path for approximately two minutes, the formation entered forest airspace. The tree canopy gradually began to obstruct the radio signal. The communication link between UAV-2 and UAV-3 was the first to be affected; the two-way communication latency fluctuated from an initial 15 milliseconds to 45-80 milliseconds, the packet loss rate increased from 0 to 5-10%, and the received signal strength decreased by approximately 15 dBm. Each UAV executed step S1, and the environment adaptive mapping model, based on the updated statistical characteristics within the sliding window, gradually lowered the LQI output value from 0.85. The LQI of the link between UAV-2 and UAV-3 dropped to 0.55.

[0073] Each UAV executes step S2, and the LQI statistical distribution characteristics within the sliding window are updated according to environmental changes. The LQI mean decreases, the standard deviation increases, and the dynamic thresholds are adjusted accordingly. The first threshold is updated to 0.68, and the second threshold is updated to 0.58. The LQI of the link between UAV-2 and UAV-3 is 0.55, which is lower than the second threshold of 0.58, and it is classified as downgraded. The LQI of the remaining links remains between 0.75 and 0.82, and is still classified as good.

[0074] For links classified as degraded, each UAV executes step S3, triggering the generation of closed-loop correction estimates. Taking UAV-2's state estimation of UAV-3 as an example, UAV-2 extracts the latest broadcast information pair from the buffer, superimposes the deviation feedback sequence onto the corresponding absolute time point of the expected trajectory sequence based on the timestamp, and generates a closed-loop correction estimate of UAV-3's current state. This closed-loop correction estimate compensates for prediction deviations caused by increased communication latency and external disturbances.

[0075] Each UAV executes step S4. For links classified as "good," compensation control remains disabled in S4. For the degraded link between UAV-2 and UAV-3, the coupling coefficient in S4 is adjusted. Based on an LQI of 0.55, the value calculated using the continuous mapping function is 1.19, which is lower than that of a good link; the velocity damping coefficient... The calculated value is 2.1, which is higher than that of a good link. Simultaneously, in S4, the link is classified as degraded, and compensation control is activated. The closed-loop correction estimate generated in S3 replaces the real-time state information in the consistency control law calculation. Due to the continuous mapping adjustment of the coupling coefficient and velocity damping coefficient, the control parameters transition smoothly at the classification boundary, and the formation does not exhibit significant oscillations.

[0076] As the formation continued deeper into the forest, UAV-3 flew directly behind a large tree, and the communication links between it and UAV-2 and UAV-4 simultaneously deteriorated sharply. Two-way communication latency surged to over 150 milliseconds, packet loss exceeded 50%, and received signal strength dropped to near the receiver's sensitivity limit. Each UAV executed step S1, and the LQI of the links related to UAV-3 dropped to 0.15 to 0.2. Step S2 was executed; because the LQI fell below the dynamically updated second threshold of 0.58, the links between UAV-3 and UAV-2 and UAV-4 were classified as interrupted. Simultaneously, the LQI between UAV-3 and its neighboring UAV-1 and UAV-5 also dropped below 0.3 due to the greater distance and obstruction, and all were classified as interrupted.

[0077] The UAV-3 executes step S3. Since it is determined that the communication links with all neighboring machines are in a degraded or interrupted state, it performs a closed-loop correction estimation generation operation for each target neighboring machine. It extracts the last received information pair from the buffer of each neighboring machine, and superimposes the deviation feedback sequence onto the corresponding absolute time point of the expected trajectory sequence according to the timestamp to obtain and store the closed-loop correction estimate of each neighboring machine.

[0078] UAV-3 executes step S4, which involves setting the corresponding coupling coefficient for each interrupted link. The value is set to zero, and the real-time state information of each target neighbor is replaced by the closed-loop correction estimate in the consensus control law solution. Simultaneously, UAV-3 executes step S5. First, it determines that the communication links with all neighboring machines are interrupted, triggering the dynamic safety window generation process. Then, it acquires the real-time accelerometer zero-bias drift rate output by the local inertial measurement unit. The current ground speed is 0.015 meters per second squared. The rate of change of current altitude is 8.0 meters per second. The value is 0.2 meters per second. Based on the above parameters, the dynamic safety window is calculated as follows: ; The calculated dynamic safety window is approximately 1.6 seconds, which is greater than the minimum threshold of 0.5 seconds, so UAV-3 enters shadow mode. Within the dynamic safety window, UAV-3 acquires the closed-loop correction estimates of each neighboring aircraft received before the interruption, combines them with the real-time changes in its own state output by its inertial measurement unit, and generates flight control commands to maintain the relative position of the formation through dead reckoning, which are then continuously executed.

[0079] During the 1.6-second execution of the dynamic security window, UAV-3's topology awareness module queries the pre-configured backup neighbor in the two-hop neighbor topology graph. Since UAV-5 is marked as the optimal backup neighbor in UAV-3's two-hop topology view, UAV-3 simultaneously sends a reconnection request to UAV-5 during shadow mode execution. Because UAV-5 is located on the right outer edge of the diamond formation, the obstruction between it and UAV-3 is relatively minor. Approximately 0.3 seconds after UAV-3 sends the reconnection request, UAV-5 successfully receives the request and sends back a response message.

[0080] Within the dynamic safety window, UAV-3 successfully established a communication link with UAV-5, exited shadow mode, and returned to the normal formation control process from S1 to S5. After the new link was established, the communication neighbor set of UAV-3 was updated to include UAV-5, and the formation topology was adjusted from the original diamond shape to a form where UAV-3 and UAV-5 were directly connected. The overall formation configuration still approximately remained a diamond shape.

[0081] If UAV-3 fails to reconnect successfully within the dynamic safety window, proceed to step S6. (The interruption duration is not specified in the original text.) If the dynamic safety window is exceeded by 1.6 seconds, UAV-3 retrieves the formation mission information stored before the interruption, and the mission type is identified as Category 1. UAV-3 enters the first-level safety degradation strategy, automatically climbing at a climb rate of 2.0 meters per second, while simultaneously sending reconnection requests to the backup communication band and the pre-set backup neighbor UAV-5. If reconnection is unsuccessful within the 10.0-second reconnection waiting period, it enters the third-level safety degradation strategy, determining whether to return to base or hover based on the remaining battery power.

[0082] In this embodiment, UAV-3 successfully reconnected within the dynamic safety window. Throughout the entire forest crossing process, the maximum relative position deviation of the formation did not exceed 0.35 meters, and none of the UAVs experienced loss of control, separation, or breakaway from the formation. After leaving the forest area, the communication links of all five UAVs were restored to good condition, and the formation continued to complete the strip mapping task along the predetermined route.

[0083] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A distributed multi-UAV consensus formation control method for addressing communication latency, characterized in that, Each drone performs the following steps: S1. Synchronize with the execution time of neighboring machines in the communication network, and periodically acquire communication quality measurement values ​​with each neighboring machine, and generate a link quality index according to preset calculation rules; S2. Based on the link quality index, classify each communication link as good, downgraded, or interrupted, construct an information pair consisting of the expected trajectory sequence and the deviation feedback sequence generated for the same prediction time domain, and broadcast it at a fixed frequency; S3. Receive the information pairs broadcast by each neighboring machine. When it is determined that the communication link with a target neighboring machine is in a degraded or interrupted state, the received deviation feedback sequence is superimposed on the corresponding absolute time point of the expected trajectory sequence according to the timestamp to generate a closed-loop correction estimate of the current state of the target neighboring machine. S4. Based on the real-time hierarchical results of the communication links between the aircraft and its neighboring aircraft, adjust the control parameters of the formation consistency control law, and enable or disable compensation control based on the closed-loop correction estimate to generate the aircraft's flight control commands. S5. When it is determined that the communication link with all neighboring aircraft is interrupted, a dynamic safety window is generated based on the drift rate output in real time by the local inertial measurement unit and the current flight status, and cooperative flight is carried out within the dynamic safety window based on the last received closed-loop correction estimate. S6. When the duration of the communication link interruption with all neighboring machines exceeds the dynamic security window, execute the preset step-by-step security degradation strategy.

2. The distributed multi-UAV consensus formation control method for addressing communication delays according to claim 1, characterized in that, In step S1, a link quality index is generated according to a preset calculation rule, including: A two-level time synchronization mechanism is adopted to synchronize with the execution time of neighboring machines in the communication network; Periodically measure the bidirectional communication delay, packet loss rate, and received signal strength with each neighboring device, and use these as the communication quality measurement values. The communication quality measurement value is input into an environment adaptive mapping model to generate the normalized link quality index.

3. The distributed multi-UAV consensus formation control method for addressing communication delays according to claim 1, characterized in that, In S2, each communication link is classified as good, degraded, or interrupted based on the link quality index, including: Obtain the statistical distribution characteristics of the link quality index between the local machine and each neighboring machine over a historical period; A first threshold and a second threshold are dynamically determined based on the statistical distribution characteristics, wherein the first threshold is higher than the second threshold. The link quality index generated in the current period is compared with the first threshold and the second threshold to classify the corresponding communication link as good, downgraded, or interrupted.

4. The distributed multi-UAV consensus formation control method for addressing communication delays according to claim 1, characterized in that, In S2, an information pair consisting of an expected trajectory sequence and a deviation feedback sequence generated for the same prediction time domain is constructed and broadcast at a fixed frequency, including: A sequence of expected trajectories is generated at a fixed frequency, wherein the expected trajectory sequence is a sequence of predicted states for a future time window; A deviation feedback sequence is generated at the same fixed frequency. The deviation feedback sequence is the error sequence between the expected trajectory sequence generated in the previous period and the actual flight trajectory in the corresponding time period. The deviation feedback sequence and the expected trajectory sequence are generated for the same prediction time domain. The expected trajectory sequence and the deviation feedback sequence are constructed into the information pair and broadcast outward at the fixed frequency.

5. The distributed multi-UAV consensus formation control method for addressing communication delays according to claim 1, characterized in that, In step S3, a closed-loop correction estimate of the current state of the target neighbor is generated, including: Receive each neighboring machine broadcasts its respective information pairs at a fixed frequency, each information pair containing a expected trajectory sequence and a deviation feedback sequence; Based on the link quality index, the real-time classification result of the communication link between the local machine and each neighboring machine is determined. When it is determined that the communication link with a target neighboring machine is in a degraded or interrupted state, the expected trajectory sequence and the deviation feedback sequence are extracted from the information pair broadcast by the target neighboring machine according to the timestamp. The extracted deviation feedback sequence is superimposed onto the absolute time point corresponding to the timestamp in the expected trajectory sequence to generate a closed-loop correction estimate of the current state of the target neighboring machine.

6. The distributed multi-UAV consensus formation control method for addressing communication delays according to claim 1, characterized in that, In step S4, the control parameters of the formation consistency control law are adjusted based on the real-time hierarchical results of the communication links between the local machine and its neighboring machines, including: Obtain the real-time classification results of the communication links between the local machine and each neighboring machine, as well as the specific values ​​of the link quality index corresponding to each link; Based on the specific value of the link quality index, the coupling coefficient and speed damping coefficient of the consistency control law are determined on a preset continuous mapping function, such that the coupling coefficient decreases continuously as the link quality index decreases, and the speed damping coefficient increases continuously as the link quality index decreases. When the communication link with the target neighbor is classified as interrupted, the coupling coefficient of the consistency control law corresponding to the target neighbor is set to zero, and the real-time state information of the target neighbor is replaced by the closed-loop correction estimate to participate in the consistency control law solution in order to generate the flight control command.

7. The distributed multi-UAV consensus formation control method for addressing communication delays according to claim 1, characterized in that, In S4, compensation control is enabled or disabled based on the closed-loop correction estimate to generate flight control commands for the aircraft, including: Obtain the real-time hierarchical results of the communication links between the local machine and each neighboring machine; When the communication link with the target neighboring aircraft is classified as good, the compensation control is disabled, and the flight control command of the aircraft is generated directly based on the real-time status information of the target neighboring aircraft received. When the communication link with the target neighboring aircraft is classified as degraded or interrupted, the compensation control is activated, and the flight control command of the aircraft is generated based on the closed-loop correction estimation to replace the real-time status information of the target neighboring aircraft.

8. The distributed multi-UAV consensus formation control method for addressing communication delays according to claim 1, characterized in that, In S5, when it is determined that the communication links with all neighboring aircraft are interrupted, a dynamic safety window is generated based on the drift rate output in real time by the local inertial measurement unit and the current flight status, including: When it is determined that the communication link with all neighboring machines is interrupted, the accelerometer zero-bias drift rate output in real time by the local inertial measurement unit is obtained. Obtain the current flight status of the aircraft, which includes at least the current ground speed and the current rate of change of altitude; Based on the accelerometer zero-bias drift rate and the current flight state, the maximum duration for which the aircraft can maintain its relative formation position during an interruption is calculated, and the dynamic safety window is generated.

9. The distributed multi-UAV consensus formation control method for addressing communication delays according to claim 1, characterized in that, In S5, cooperative flight is performed within the dynamic safety window based on the last received closed-loop correction estimate, including: Obtain the closed-loop correction estimate of each neighboring machine that was last received before the communication link with all neighboring machines was interrupted; Based on the closed-loop correction estimation, the current state of each neighboring aircraft before the interruption is determined, and combined with the real-time output of the change in the aircraft's state from the inertial measurement unit, flight control commands for maintaining the relative position of the formation are generated through dead reckoning. The flight control commands are continuously executed within the dynamic safety window to maintain formation configuration with neighboring aircraft.

10. The distributed multi-UAV consensus formation control method for addressing communication delays according to claim 1, characterized in that, In step S6, a preset step-by-step security degradation strategy is executed, including: When the duration of the communication link interruption with all neighboring machines exceeds the dynamic safety window, retrieve the formation task information stored before the interruption. When the formation mission information indicates a first type of mission, the first level of security degradation strategy is entered, the local machine is controlled to automatically climb to obtain the communication field of view, and a reconnection request is sent to the backup communication frequency band or backup neighboring machine. When the formation mission information indicates a second type of mission, the second level of safety degradation strategy is entered, the machine is controlled to maintain the current altitude and hover in the preset airspace, and the machine's position information is continuously broadcast. If reconnection fails within the preset reconnection waiting time, the third-level security degradation strategy will be implemented, controlling the local machine to perform a return or hover.

Images