Distributed parameter estimation method for unmanned aerial vehicle swarm based on adaptive data selection

Abstract

The application provides a distributed unmanned aerial vehicle cluster parameter estimation method based on adaptive data selection, a cross matching mechanism is established based on fusion data transmitted at the last time, and then a new data selection factor lambda is obtained k (t) generating a strategy and determining whether unmanned aerial vehicle k exchanges data with neighbor unmanned aerial vehicles based on the new lambda k (t) The application improves the estimation accuracy while reducing the calculation cost and communication cost and saving the energy consumption of the unmanned aerial vehicle cluster. The distributed ADS-DLMS algorithm is designed based on the cross matching mechanism, which not only improves the parameter estimation accuracy of the distributed unmanned aerial vehicle cluster, but also reduces the calculation cost and communication cost of the unmanned aerial vehicle node, and provides a new information fusion technical scheme for the distributed unmanned aerial vehicle cluster.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of unmanned aerial vehicles, in particular to a parameter estimation method for unmanned aerial vehicle clusters, and specifically designs an adaptive data selection scheme with a cross-matching mechanism, which improves the parameter estimation accuracy of unmanned aerial vehicles while reducing the computational cost and communication cost of unmanned aerial vehicles in the cluster, and provides a new information fusion technology support for distributed unmanned aerial vehicle clusters. BACKGROUND

[0002] Distributed theory is widely used in the fields of machine learning, smart city, wireless sensor network, artificial intelligence, Internet of Things, etc. Compared with centralized, distributed has many advantages such as scalability, reliability and robustness. In distributed parameter estimation, the distributed diffusion least-mean-square algorithm (Diffusion Least-Mean-Square, DLMS) with diffusion strategy can provide excellent adaptability and flexibility in terms of distributed node cooperation, and therefore has attracted extensive attention. Although the research on distributed parameter estimation has been extended to a very wide field, when its theory is applied to unmanned aerial vehicle clusters, since the unmanned aerial vehicles can only carry limited energy when flying, if each unmanned aerial vehicle node processes all the sensing data, the node's computing load will be relatively heavy, which will reduce the overall performance of the distributed unmanned aerial vehicle cluster. Especially in the era of big data, the problem of node computing load becomes more obvious and prominent.

[0003] Therefore, in order to reduce the computational load, it is extremely important to introduce the data selection (DS) strategy into the distributed parameter estimation algorithm from the perspective of the UAV. Based on this, researchers have studied distributed DS algorithms (Ierardi C, Orihuela L, Jurado I. A distributed set-membership estimator for linear systems with reduced computational requirements [J]. Automatica, 2021, 132: 109802.; Flores A, de Lamare R C. Set-membership adaptive kernel NLMS algorithms: Design and analysis [J]. Signal Processing, 2019, 154: 1-14.; Ding D, Wang Z, Han Q L. A set-membership approach to event-triggered filtering for general nonlinear systems over sensor networks [J]. IEEE Transactions on Automatic Control, 2019, 65(4): 1792-1799.) that discard data exceeding a predefined threshold and select normal data that meets the DS criteria for information fusion. Bhotto MZA et al. proposed a distributed DS least-mean-square (LMS) algorithm with Newton's method (Bhotto MZA, Antoniou A. Improved data-selective LMS-Newton adaptation algorithms [C] / / 2009 16th International Conference on Digital Signal Processing. IEEE, 2009: 1-6), which selects data by combining a priori error with update step size. However, there is a common point in the above DS strategies: they all only select normal data that does not exceed the predefined threshold.To further reduce the computational load, researchers (Diniz P S. On data-selective adaptive filtering[J]. IEEE Transactions on Signal Processin, 2018, 66: 4239-4252; Tsinos CG, Diniz P S R. Data-selective lms-newton and lms-quasi-newton algorithms[C] / / ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019: 4848-4852;. M O K, Ferreira J O, Tsinos CG, et al. On fast converging data-selective adaptive filtering[J]. Algorithms, 2018, 12(1): 4) proposed a DS LMS (DS-LMS) algorithm, which not only selects normal data, but also selects data that provides more new information in the update iteration. Hua Y et al. considered the cooperation between distributed network nodes and proposed a DSDLMS (DS-DLMS) algorithm (Hua Y, Chen F, Duan S, et al. Distributed data-selective DLMS estimation under channel attacks[J]. IEEE Access, 2019, 7: 83863-83872), to enhance the estimation ability through cooperation and select normal and new information data.

[0004] However, the above DS-LMS algorithm and DS-DLMS algorithm are based on the selection factor in the selection rule of the general DS (General DS, GDS). This GDS scheme will cause a certain accuracy loss in distributed parameter estimation, and both the DS-LMS algorithm and the DS-DLMS algorithm with the GDS scheme only consider the computational cost of the unmanned aerial vehicle node, ignoring the extremely important communication cost problem between unmanned aerial vehicles. In the era of big data, the problem of communication load has become particularly prominent. SUMMARY

[0005] In order to overcome the prior art, the present application provides a kind of based on adaptive data selection's distributed unmanned aerial vehicle cluster parameter estimation method.The present application designs an adaptive DS (Adaption DS, ADS) scheme, based on the fusion data transmitted in last time, establishes a cross matching mechanism, then obtains a new data selection factor λ k (t) k (t) determines whether unmanned aerial vehicle k exchanges data with neighbor unmanned aerial vehicle in communication, while improving estimation accuracy, reaches the reduction of computing cost and communication cost, saves the energy consumption of unmanned aerial vehicle cluster.

[0006] The technical solution of the present application for solving its technical problems comprises the following steps:

[0007] Step one: based on cross matching mechanism generates data selection factor λ k (t)

[0008] In t time, the k unmanned aerial vehicle has obtained the communication data of neighbor unmanned aerial vehicle set in t-1 time, first defines intermediate estimation error variable α k (t)

[0009] Based on intermediate estimation error variable α k (t) with cross matching condition, data selection factor λ k (t) is set;

[0010] Step two: adaptive update iteration with ADS;

[0011] When λ k (t) =0, the estimation value w k,t-1 of last iteration time is more reliable than w k,t ; In combination with the data selection characteristics of selection factor λ k (t) ∈ {0,1}, the following adaptive update iteration with ADS is designed to reduce the computing cost:

[0012] ψ k,t = w k,t-1+λk( t ) + λ k (t) μ k u k,t e k (t)

[0013] Wherein, It is the estimation of k unmanned aerial vehicle to w° in t-1+ λ k (t) time, μ k Indicates update step, u k,t Indicates regression vector signal, d k(t) represents the scalar expected response, and w° represents the interest parameter that needs to be estimated for an unknown and distributed UAV swarm with dimension L×1.

[0014] Step 3: Communication and data exchange;

[0015] Each drone k is selected based on the selection factor λ in its ADS. k The value of (t) determines whether to send exchanged data ψ to neighboring drones. k,t When the selection factor λ k When (t) = 1, the update formula in step two is ψ k,t =w k,t +μ k u k,t e k (t), which indicates that the data ψ exchanged at this time k,t It has been updated, and the data fusion accuracy for other drones has improved after the exchange. Therefore, the drone k in λ k (t) = 1 will send the exchanged data ψ k,t However, when the selection factor λ k When (t) = 0, the update formula in step two becomes ψ. k,t =w k,t-1 The intermediate estimate at this point is the same as the estimate from the previous time step. Therefore, the drone k in λ k (t) = 1 will not send exchanged information;

[0016] In summary, under the ADS strategy, drones in a distributed drone swarm only communicate with their neighbors when the intermediate estimate is updated, thus reducing the communication cost between drones.

[0017] Step 4: Distributed UAV swarm data fusion;

[0018] According to step three, when At that time, drone k received a message from its neighboring drone. Communication data But when At that time, drone k will not receive signals from neighboring drones. Communication data The data fusion method for distributed drone swarms is as follows:

[0019]

[0020] in, This represents the set containing drone node k and its neighboring drone nodes. It is a fusion weight. It is the first A drone pair Time about The intermediate estimation, the update formula of ADS-DLMS algorithm is as follows:

[0021]

[0022] ADS not only selects the perception data of the network, but also selects the data according to the data selection factor λ k (t) value selection estimation and determine whether to communicate with the surrounding neighbor unmanned aerial vehicle; then each unmanned aerial vehicle carries out data iterative update, and carries out corresponding data fusion according to whether the information transmitted by the neighbor unmanned aerial vehicle is received; for the unmanned aerial vehicle, as long as it is judged whether the information of the neighbor unmanned aerial vehicle can be received, if received, it is directly fused, and if not received, the last information is replaced directly.

[0023] In step one, the intermediate estimation error variable α k (t) is:

[0024]

[0025] Where d k (t) is the scalar expected response, u k,t is the Lx1 regression vector signal, and the operator (·) * represents the complex conjugate transpose of the vector or matrix, ψ l,t-1 is the intermediate estimation of the lth unmanned aerial vehicle pair about w l,t-1 at t-1, w l,t-1 is the estimation of the lth unmanned aerial vehicle pair about the unknown parameter w° at t-1, and l satisfies the following cross matching condition:

[0026]

[0027] Wherein represents a set containing unmanned aerial vehicle node k and its neighbor unmanned aerial vehicle nodes;

[0028] The data selection factor λ k (t) is:

[0029]

[0030] Wherein and are error tolerance coefficients in the cross matching mechanism, is a judgment whether the difference between the data packet {d k (t), u k,t} is within the tolerance degree; is a judgment whether the data packet {d k (t), u k,t} has generated enough new information; therefore, is the GDS are different; while and are the same, the same strategy in GDS is adopted, i.e., the selected selection probability P k (t) and the value of v are inversely solved to obtain .

[0031] The present application has the beneficial effect that a cross-matching mechanism is designed, based on which the design rule of the data selection factor λ k (t) is improved, and then an adaptive data selection ADS scheme is proposed. In the scheme, each unmanned aerial vehicle adaptively decides which estimated value at which time to select in the update according to the selection factor λ k (t), and whether to exchange data with the neighbor unmanned aerial vehicle in the communication. Therefore, the present application designs a distributed ADS-DLMS algorithm based on the cross-matching mechanism, which not only improves the parameter estimation accuracy of the distributed unmanned aerial vehicle cluster, but also reduces the calculation cost and communication cost of the unmanned aerial vehicle node, and provides a new information fusion technical scheme for the distributed unmanned aerial vehicle cluster. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 is a network diagram of the distributed DS-DLMS algorithm with the NS mechanism.

[0033] Figure 2 is a network diagram of the distributed ADS-DLMS algorithm with the cross-matching mechanism.

[0034] Figure 3 is an input signal covariance, noise covariance, and signal-to-noise ratio SNR diagram of the distributed unmanned aerial vehicle node; wherein Figure 3 (a) is an input signal covariance diagram of the distributed unmanned aerial vehicle node, wherein Figure 3 (b) is a noise covariance diagram of the distributed unmanned aerial vehicle node, wherein Figure 3 (c) is a signal-to-noise ratio SNR diagram of the distributed unmanned aerial vehicle node.

[0035] Figure 4 is a transient network MSD diagram of the DLMS algorithm, the NCLMS algorithm, the DS-LMS algorithm, the DS-DLMS algorithm, the SA-DLMS algorithm, and the ADS-DLMS with cross-matching algorithm.

[0036] Figure 5 is a communication rate diagram of the DLMS algorithm, the DS-DLMS algorithm, the SA-DLMS algorithm, and the ADS-DLMS with cross-matching algorithm.

[0037] Figure 6 is the computational rate plot of DLMS, NCLMS, DS-LMS, DS-DLMS, SA-DLMS, and ADS-DLMS with cross-matching algorithms.

[0038] Figure 7 is the transient network MSD plot of ADS-DLMS under NS and ADS-DLMS with cross-matching algorithms.

[0039] Figure 8 is the communication rate plot of ADS-DLMS under NS and ADS-DLMS with cross-matching algorithms.

[0040] Figure 9 is the computational rate plot of ADS-DLMS under NS and ADS-DLMS with cross-matching algorithms. DETAILED DESCRIPTION

[0041] The application will be further described below in conjunction with the accompanying drawings and examples.

[0042] In the era of data explosion, with the application of distributed unmanned aerial vehicle cluster becoming more and more widely, the computational cost and communication cost often become the main factors of determining the distributed unmanned aerial vehicle cluster under the premise of ensuring the parameter estimation accuracy of the DLMS algorithm. The ADS scheme designed by the application can further improve the estimation accuracy while effectively reducing the computational cost and communication cost, thereby reducing the energy consumption of the distributed unmanned aerial vehicle cluster and prolonging the service life of the unmanned aerial vehicle cluster.

[0043] The technical solution adopted by the application to solve the technical problems comprises the following steps:

[0044] Step one: generating a data selection factor λ based on a cross-matching mechanism k (t);

[0045] At the kth unmanned aerial vehicle at time t, the communication data of the neighbor unmanned aerial vehicle set at time t-1 has been obtained, and an intermediate estimation error variable α k (t) is defined as:

[0046]

[0047] Where d k (t) is a scalar expected response, u k,t is an Lx1-dimensional regression vector signal, and the operator (·) * represents the complex conjugate transpose of a vector or matrix, ψ l,t-1Is the l-th UAV pair at time t-1 about w l,t-1 The intermediate estimate, w l,t-1 Let be the estimate of the unknown parameter w° by the l-th UAV at time t-1, and let l satisfy the following cross-matching condition:

[0048]

[0049] in It represents the set containing drone node k and its neighboring drone nodes;

[0050] Based on the intermediate estimation error variable α with cross-matching conditions k (t), data selection factor λ k (t) is:

[0051]

[0052] in and It is the error tolerance coefficient in the cross-matching mechanism. It is to determine the data packet {d k (t),u k,t Are the differences between them within a tolerable range? It is to determine the data packet {d k (t),u k,t Whether enough new information has been generated; therefore, In GDS They are different; and and It is the same, and is generated using the same strategy as in GDS, that is, selecting the selection update probability P. k Given the values ​​of t and ν, we can obtain the inverse solution. value;

[0053] Step 2: Adaptive update iteration with ADS;

[0054] The distributed DS-DLMS algorithm only discards the selection factor λ. k Data packet {d} under (t) = 0 k (t),u k,t However, observation error and selection factor λ k The generation method of (t) reveals that if in λ k When (t) = 0, it represents {d k (t),u k,t ,w k,t The three are either mismatched or meaningless, which means that the estimated value w generated at time t is... k,t In data packet {d k (t),uk,t The following is also not trusted; therefore, in order to obtain better estimation performance, this invention improves the distributed DS algorithm, in λ k When (t) = 0, there is reason to believe that the estimate w at the previous iteration time is... k,t-1 Compared to w k,t It is more reliable; combined with the selection factor λ k To reduce computational cost, we design an adaptive update iteration with ADS to address the data selection characteristics of (t)∈{0,1}:

[0055] ψ k,t =w k,t-1+λk( t ) +λ k (t)μ k u k,t e k (t)

[0056] Step 3: Communication and data exchange;

[0057] Each drone k is selected based on the selection factor λ in its ADS. k The value of (t) determines whether to send exchanged data ψ to neighboring drones. k,t When the selection factor λ k When (t) = 1, the update formula in step two becomes ψ. k,t =w k,t +μ k u k,t e k (t), which indicates that the data ψ exchanged at this time k,t It has been updated, and the data fusion accuracy for other drones has improved after the exchange. Therefore, the drone k in λ k (t) = 1 will send the exchanged data ψ k,t However, when the selection factor λ k When (t) = 0, the update formula in step two becomes ψ. k,t =w k,t-1 The intermediate estimate at this point is the same as the estimate from the previous time step. Therefore, the drone k in λ k (t) = 1 will not send exchanged information;

[0058] In summary, it can be observed that drones in a distributed drone swarm only communicate with their neighboring drones when the intermediate estimate is updated under the ADS strategy, thus reducing the communication cost between drones.

[0059] Step 4: Distributed UAV swarm data fusion;

[0060] According to step three, when At that time, drone k received a message from its neighboring drone. Communication data But when At that time, drone k will not receive signals from neighboring drones. Communication data Under the designed ADS strategy, when the following occurs At that time, drone k will take over the neighboring drone Communication data from the previous moment As the fusion volume for this update; therefore, based on the designed ADS strategy, the data fusion method for the distributed drone swarm is designed as follows:

[0061]

[0062] In summary, based on steps one through four, the update algorithm of the ADS-DLMS algorithm is obtained as follows:

[0063]

[0064] Based on the distributed ADS-DLMS algorithm under the ADS scheme, the following is obtained: Figure 2 The distributed system shown

[0065] Network diagram of the ADS-DLMS algorithm; in Figure 2 In the study, it was found that ADS not only selects data from the network's perception data, but also selects a factor λ based on the data. k (t) Select an appropriate estimate and decide whether to communicate with neighboring drones; then each drone iterates and updates its data, and performs corresponding data fusion based on whether it receives information from neighboring drones; for a drone, it is only necessary to determine whether it can receive information from neighboring drones. If it receives the information, it will directly fuse the data. If it does not receive the information, it will use the previous information to replace the data and fuse it directly. The fusion method adopts the ADS-DLMS algorithm in step two.

[0066] The following is an example:

[0067] (1) Data Model of Distributed Unmanned Aerial Vehicle Cluster

[0068] A distributed drone swarm consisting of N drones in a self-organizing manner can be simulated using the following linear data model at time t, where the k-th drone node is a regression vector signal u of dimension L×1. k,t With scalar expected response d k The relationship between (t):

[0069]

[0070] Where w° represents an unknown, distributed drone swarm of interest parameter of dimension L×1 that needs to be estimated, and the operator (·)* Represents the complex conjugate transpose of a vector or matrix, n k (t) indicates that the expression follows a pattern with mean 0 and variance . Additive white Gaussian noise.

[0071] To estimate the unknown parameter w°, a DLMS algorithm based on distributed theoretical optimization derivation can be obtained, as shown below:

[0072]

[0073] Where w k,t It is the estimate of w° by the k-th UAV at time t, ψ k,t Is the k-th UAV pair at time t with respect to w? k,t Intermediate estimate, μ k It's about updating the step size. It is a fusion weight. It represents the set containing drone node k and its neighboring drone nodes.

[0074] Fusion weights It is the weight matrix C of the th For each corresponding element, the following conditions must be met:

[0075]

[0076] in This represents an L×1 column vector consisting entirely of elements 1.

[0077] (2) Data selection model based on GDS

[0078] For the k-th UAV at time t, the data selection factor λ of the distributed DS algorithm based on the GDS scheme is... k (t) is designed in the following form:

[0079]

[0080] in and These are two undetermined DS parameters. From the above equation, it can be seen that λ k (t) is generated based on noise statistics (NS) information, so this method of generating selection factors is defined as the NS mechanism.

[0081] if The DS algorithm considers the data packet {d} to be valid. k (t),u k,t The packet may contain outliers or mismatches, so it needs to be discarded. Therefore, the selection factor λ of DS is...k (t) = 0. If (t) = 0. The DS algorithm considers the data packet {d} to be valid. k (t),u k,t Since no new information can be generated during the update iteration, this data packet also needs to be discarded. Therefore, λ k (t) = 0. Only data that meets the DS requirements will be selected for data fusion, i.e., λ k The data packet {d} corresponding to (t) = 1 k (t),u k,t}

[0082] In the GDS scheme, and The relationship between them is determined by the update probability P of the white Gaussian input signal. k (t) means that:

[0083]

[0084] Where the function Q(·) represents the complementary Gaussian cumulative distribution function, and is defined as:

[0085]

[0086] In the GDS scheme, the update probability P k (t) and and The following relationship exists between them:

[0087]

[0088] Therefore, the update probability P k (t) is transformed into the following form:

[0089]

[0090] in because Since the value is generally large, its function Q(·) value will be small, and its influence in the above formula is relatively small, so its corresponding term can be ignored. Then, select an appropriate update probability P. k inverse solution of (t) and ν value We can obtain:

[0091]

[0092] Q -1 (·) is the inverse function of the complementary Gaussian cumulative distribution function Q(·).

[0093] In summary, in the GDS scheme, as long as an appropriate update probability P is selected... k (t), ν value, and value, the selection factor λ k (t) to get the following distributed DS algorithm:

[0094] DS-LMS: w k,t+1 = w k,t + λ k (t) μ k u k,t e k (t)

[0095]

[0096] According to the distributed DS algorithm under the GDS scheme, the network diagram of the distributed DS-DLMS algorithm can be obtained as shown in FIG. 6. In FIG. 6, it can be found that the GDS only selects the perception data of the cluster network, then iteratively updates the data, and performs data fusion after all communications are completed. In FIG. 7, all communication links are disconnected to obtain the schematic diagram of the DS-LMS algorithm. Figure 1 Figure 1 Figure 1

[0097] In order to fully verify the superiority of the algorithm and the strategy therein, the following three indexes are introduced for measurement:

[0098] Index 1: Transient Network Mean-Square-Deviation (MSD), which is used to measure the deviation degree of the estimated value w k,t from the ideal true value w° at time t, and is defined as

[0099]

[0100] It can be found that the smaller the Transient Network MSD is, the higher the estimation accuracy is.

[0101] Index 2: Communication rate, which refers to the frequency of data transmission between unmanned aerial vehicle nodes in the entire distributed unmanned aerial vehicle cluster at each iteration. The communication rate is defined as

[0102]

[0103] Because the DLMS algorithm is a full-communication algorithm, it communicates with neighbor nodes after each iteration, so the DLMS algorithm has the most communication times. Therefore, the DLMS algorithm is taken as a reference value for the communication rate.

[0104] ​​​Metric 3: Computation rate, refers to the rate at which each drone node in the distributed drone swarm processes data packets {d} after each data perception in each iteration. k (t),u k,t The frequency at which updates are performed. The communication rate is defined as...

[0105]

[0106] Because the DLMS algorithm senses data packets {d} each time k (t),u k,t All of these will be used in adaptive iterative updates, so the DLMS algorithm has the highest computation rate for data packets. Therefore, the DLMS algorithm is used as a reference value for the computation rate.

[0107] In this experimental example, MATLAB is used as the simulation platform, considering a distributed drone swarm network consisting of N=25 drones. The network's parameter w° to be estimated is set as a random vector of length L=5. In the following experiment, all the following parameters are set uniformly and remain unchanged during the experiment: μ k =0.014; P k (t) = 0.5; ν = 5; All the experimental results below are based on 500 Monte Carlo simulations. In all experiments, the regression signal u at each node... k,t covariance and noise n k The covariance of (t) is as follows Figure 3 As shown. Furthermore, the signal-to-noise ratio (SNR) of each drone node is also presented. Figure 3 middle.

[0108] Experiment 1: Performance verification of the distributed ADS-DLMS algorithm based on cross-matching mechanism;

[0109] In this experiment, the comparison algorithms used mainly included DLMS, NCLMS (Non-Cooperation LMS), DS-LMS, and DS-DLMS. Since the SA-DLMS (Sign Adoption DLMS) algorithm also applies to data packets {d... k (t),u k,t The algorithm that performs symbolic operations is therefore also used for comparison.

[0110] like Figure 4The Transient Network MSD of the main verification algorithm is shown. It can be seen that the performance of the DS-LMS algorithm and the NCLMS algorithm remains consistent, and the performance of the SA-DLMS algorithm is slightly better than that of the SA-DLMS algorithm due to the communication cooperation between the unmanned aerial vehicles, but the performance of the SA-DLMS algorithm is worse than that of the DLMS algorithm and the DS-DLMS algorithm. From Figure 4 it can be found that the performance of the DS-DLMS algorithm is slightly better than that of the DLMS algorithm. Among all the algorithms, the ADS-DLMS algorithm with cross-matching designed by the application shows the best performance and can obtain the best accuracy.

[0111] Figure 5 The communication rate is investigated. In Figure 5 , since the DS-LMS algorithm and the NCLMS algorithm adopt a non-cooperative strategy to run the algorithm, they do not have communication. Although the SA-DLMS algorithm and the DS-DLMS algorithm perform different operations on the data packet, respectively, they still require 100% communication. From Figure 5 it can be seen that the communication rate of the ADS-DLMS with cross-matching algorithm designed by the application is only about 45%, which saves more than general communication resources.

[0112] Figure 6 The calculation rate is investigated. The NCLMS algorithm and the DLMS algorithm do not have a data selection strategy, so their calculation rates are 100%; and the SA-DLMS algorithm only performs a symbol operation on the data packet, but still needs to be used in step two, so its calculation rate is also 100%. Since the DS-LMS algorithm and the DS-DLMS algorithm both adopt DS based on the NS mechanism, the calculation amount can be reduced, and the calculation amount of both is about 50%. The ADS-DLMS algorithm designed by the application can obtain the minimum calculation amount since it adopts ADS obtained by the cross-matching mechanism.

[0113] From Figures 4 to 6 , whether it is the MSD performance, or the communication rate and the calculation rate, the ADS-DLMS algorithm with the cross-matching mechanism designed by the application performs very well, so it can be proved that the application can improve the estimation ability in the distributed unmanned aerial vehicle cluster, and can reduce the network resource consumption and enhance the service life of the unmanned aerial vehicle cluster.

[0114] Experiment two: performance comparison of two different data selection factors under the ADS scheme

[0115] In this experiment, the superiority of the data selection factor λ generated based on the NS mechanism and the cross-matching mechanism is compared and investigated. In the experiment, in order to ensure the fairness of comparison, the control variable method is adopted, and the data selection algorithm is based on the distributed DLMS algorithm under the ADS scheme for distributed estimation. k (t) are compared and investigated. In the experiment, in order to ensure the fairness of comparison, the control variable method is adopted, and the data selection algorithm is based on the distributed DLMS algorithm under the ADS scheme for distributed estimation.

[0116] Figure 7 The transient network MSD performance of the two algorithms is verified. From Figure 7 , it can be found that the estimation performance of the ADS-DLMS algorithm with the cross-matching mechanism is obviously better than that of the ADS-DLMS algorithm with the NS mechanism.

[0117] Figure 8 The communication rate is investigated. It can be seen that the selection performance of the ADS-DLMS algorithm with the NS mechanism is about 50%, which is the same as that of the DS-LMS algorithm and the DS-DLMS algorithm with the NS mechanism in Figure 5 . The communication rate of the ADS-DLMS algorithm with the cross-matching mechanism is obviously lower than that of the ADS-DLMS algorithm with the NS mechanism.

[0118] Figure 9 The calculation rate is investigated. It can be seen that the calculation rate of the ADS-DLMS algorithm under the NS mechanism is similar to the communication rate result, and is also higher than that of the ADS-DLMS algorithm under the cross-matching mechanism designed in the application.

[0119] Based on all the experiments, by comparing all the related algorithms with DS, it can be found that the novel cross-matching mechanism data selection factor generation method designed in the application has better selection performance; and the ADS scheme designed in the application not only improves the distributed estimation accuracy, but also reduces the communication rate and the calculation rate of the distributed unmanned aerial vehicle cluster, which provides technical support for the large-scale application of the unmanned aerial vehicle cluster with the distributed theory.

Claims

1. A method for estimating parameters of a distributed unmanned aerial vehicle (UAV) swarm based on adaptive data selection, characterized in that... Includes the following steps: Step 1: Generate data selection factors based on cross-matching mechanism ; exist The first moment One drone has been obtained Based on the communication data of the neighboring drone ensemble, first define the intermediate estimation error variable. ; Based on intermediate estimation error variables with cross-matching conditions Set data selection factor ; Step 2: Adaptive update iteration with ADS; exist At that time, the estimated value at the previous iteration time. Compare More reliable; combined with selection factors Based on the data selection characteristics, the following adaptive update iteration with ADS is designed to reduce computational cost: ； in, It is the first One drone pair exist Time estimation, Representing a dimension as The unknown and distributed drone swarms require estimation of interest parameters. Indicates the update step size. Represents the regression vector signal. , operator This indicates the complex conjugate transpose. Scalar expected response; Step 3: Communication and data exchange; Each drone Based on the selection factor in its ADS The value determines whether to send exchanged data to neighboring drones. When selection factor At that time, the update formula in step two is: This indicates that the data being exchanged at this time It has been updated, and the data fusion accuracy for other drones has improved after the exchange, therefore the drones exist It will send exchanged data to neighboring drones. But when the selection factor At that time, the update formula in step two becomes The intermediate estimate at this point is the same as the estimate from the previous time step, therefore the drone... exist No exchanged information will be sent; In summary, under the ADS strategy, drones in a distributed drone swarm only communicate with their neighbors when the intermediate estimate is updated, thus reducing the communication cost between drones. Step 4: Distributed UAV swarm data fusion; According to step three, when ( At that time, drones Received a neighbor's drone Communication data But when ( At that time, drones Will not receive neighbor's drone Communication data The data fusion method for distributed drone swarms is as follows: in, Representative includes drone nodes And the aggregation of its neighboring drone nodes, It is a fusion weight. It is the first A drone pair Time about Based on the intermediate estimates, the update formula for the ADS-DLMS algorithm is obtained as follows: ； ADS not only selects data from the network's sensing data, but also selects data based on selection factors. The system selects an estimated value and decides whether to communicate with neighboring drones. Then, each drone iterates and updates its data, and performs corresponding data fusion based on whether it receives information from neighboring drones. For a drone, it only needs to determine whether it can receive information from neighboring drones. If it receives the information, it merges the data directly; if it does not receive the information, it merges the data directly using the information from the previous iteration.

2. The distributed UAV swarm parameter estimation method based on adaptive data selection according to claim 1, characterized in that: In step one, the intermediate estimation error variable for: ； in For scalar expected response, Dimensions Regression vector signal, operator Represents the complex conjugate transpose of a vector or matrix. It is the first A drone pair Time about Intermediate estimate, It is the first A drone for unknown parameters exist Time estimation, The following cross-matching conditions must be met: ； in Representative includes drone nodes And the collection of its neighboring drone nodes.

3. The distributed UAV swarm parameter estimation method based on adaptive data selection according to claim 1, characterized in that: The data selection factor for: ； in , and It is the error tolerance coefficient in the cross-matching mechanism. It is to judge data packets Are the differences within acceptable limits? It is to judge data packets Has enough new information been generated? therefore, In GDS They are different; and and They are the same, generated using the same strategy as in GDS, i.e., selecting and updating probabilities. and The value is obtained by inverse solution. value.

Images