A method for scheduling of a UAV group's reconnaissance and calculation power based on cooperative calculation

Abstract

The application discloses a kind of unmanned aerial vehicle group reconnaissance power scheduling method based on collaborative computing, the method is directed to how to realize information intercommunication and cluster resource comprehensive collaborative allocation in the process of collaborative detection without relying on ground mobile base station and cloud computing server, establish cluster information collaborative model and form cluster collaborative computing alliance, through mutual communication between unmanned aerial vehicle, reasonably coordinate cluster detection and computing resource, and realize cluster reconnaissance power optimal method.The present application includes that the system is based on task search vector in the detection topology optimization layer, optimizes the detection position of unmanned aerial vehicle, in the resource optimization layer, the system alternates task unloading proportion and computing resource allocation proportion to give the reconnaissance power under the current detection topology, the system is constantly optimized detection topology optimization layer and resource allocation optimization layer according to the feedback of reconnaissance power, finally obtain the optimal detection topology and resource allocation scheme of cluster, realize efficient detection.

Description

TECHNICAL FIELD

[0001] The present application relates to swarm intelligence technology, UAV cluster communication, UAV cooperative computing, UAV cooperative perception, UAV resource allocation, and in particular to a detection and calculation force scheduling method in a UAV cluster cooperative computing and detection system. BACKGROUND

[0002] In disaster low-altitude detection, the application of UAV has become a common practice. The use of UAV can enable rescue personnel to quickly understand the damage situation in the disaster area and quickly launch search and rescue work through a combination of human and machine collection modes. However, the current UAV detection mode still mainly relies on a single UAV for non-intelligent detection operations. Detection data is usually transmitted back to the command center for processing through wireless network or optical fiber connection, and then the results are transmitted to the search and rescue personnel terminal. This mode limits the flexibility of UAV to some extent and lacks intelligent processing capability, making it difficult to fully meet the diversified needs in complex environments.

[0003] With the rapid development and innovation of UAV technology, the importance of UAV in disaster detection is increasingly prominent. In this context, due to the limited detection resources and computing resources of a single UAV, it is often difficult for a single UAV to handle large-area efficient detection and large-scale data processing tasks alone. Therefore, the detection pattern of UAV is undergoing changes, and UAV cluster cooperative detection based on cooperative computing has become an important trend in future UAV intelligent detection methods. How to determine the detection position of each UAV and allocate the computing resources of each UAV is of paramount importance for UAV clustering and intelligent detection. SUMMARY

[0004] The present application aims to overcome the defects and shortcomings of the prior art and proposes a UAV cluster group detection and calculation force scheduling method based on cooperative computing, which is applied to the future intelligent UAV cluster detection scene.

[0005] The technical solution adopted by the present application to solve its technical problems is: a UAV cluster group detection and calculation force scheduling method based on cooperative computing, which includes five overall steps of S1 to S5:

[0006] S1, establish an adaptive grid map based on initial prior information to obtain a task search vector

[0007] S2, analyze the composition of the UAV detection vector, including the UAV detection range model, the UAV probability detection model, the task search vector update model, and establish the UAV cluster information cooperation model.

[0008] S3, analyze the composition of the UAV computing vector, including the UAV cooperative computing attribute, the cluster cooperative computing alliance, the task offloading proportion matrix, and the computing resource allocation matrix.

[0009] S4, analyze the composition of the group detection capability, including the cluster detection capability and the cluster task completion delay, obtain the calculation method of the group detection capability.

[0010] S5, establish a system optimization model with the goal of maximizing the group detection capability, and optimize the unmanned aerial vehicle cluster detection position and cluster resource allocation through the detection machine swarm intelligent optimization algorithm.

[0011] The following will explain the above five steps one by one in detail:

[0012] S1 constructs an adaptive grid map, that is, the adaptive grid map is established based on prior information, and the steps are as follows:

[0013] S1.1, according to the detection data of low-orbit satellite, the geographic environment information M={x M ,y M ,z M} of the target area is obtained, the number of target points to be detected n target and the target existence position information R={x R ,y R ,z R} are obtained.

[0014] S1.2, adaptive grid map G={g1, g2,…,g i}: according to the detection weight, the task area is divided into grid maps with different grid sizes. The target area is divided into four grids, and the grid detection weight is calculated:

[0015]

[0016] Among them, represents the importance of the target to be detected in the grid, represents the terrain complexity, α w ,β w are the corresponding coefficients. The grid with a weight exceeding the threshold W th is further divided into four sub-grids, and the above process is repeated until it cannot be divided.

[0017] S1.3, task search vector The task search vector is composed of the grid target existence probability and the grid uncertainty, and the grid target existence probability represents the possibility of the target existing in the grid g i detected by the unmanned aerial vehicle j, and the grid uncertainty represents the uncertainty of the grid g i detected by the unmanned aerial vehicle j. The calculation methods of the grid target existence probability and the grid uncertainty will be introduced one by one.

[0018] (1) Grid Target Existence Probability: Initial Grid Target Existence Probability Grid Detection Weight W i Normalization:

[0019]

[0020] wherein, η represents the accuracy of prior information, η ∈ [0, 1].

[0021] (2) Grid Uncertainty is defined as the information entropy of the target existence probability in the grid:

[0022]

[0023] S2 UAV Detection Vector Calculation Method

[0024] UAV Detection Vector ZC j = {vg j , PD j , PF j , Y j}, wherein vg j represents the detectable grid set of UAV j, PD j represents the successful detection probability, PF j represents the false detection probability, and the successful detection probability and the false detection probability are collectively referred to as the UAV probability detection model; Y j represents the task search vector of UAV j. The calculation methods of each part are introduced in turn as follows.

[0025] S2.1, UAV Detection Range Model: The center point of the grid is set as the information point of the grid. If the grid information point is within the detection range of the UAV and is not blocked by obstacles such as mountains, then the grid is within the detectable grid set vg j of the UAV.

[0026] S2.2, UAV Probability Detection Model:

[0027] The successful detection probability represents the probability that the UAV system successfully judges the existence of the target when the target exists, and is defined as the exponential function of the square of the height difference between the UAV and the detectable grid:

[0028]

[0029] In the formula, ω s is the successful detection sensitivity of the UAV; κ s is the successful detection coefficient of the UAV; indicates whether there is actually a target in the grid g i , when This means that there is no actual target in the grid gi, when Represents grid g i There are actual goals in China; It indicated the specific results of the drone's detection, when This means that at time t, the drone j detects grid g. i There is a goal, when This represents the drone detecting the grid g. i There is no target; Indicates the relationship between the drone j and the grid g. i The height difference between them.

[0030] Error detection probability represents the probability that the UAV system incorrectly determines the existence of a target when the target does not exist:

[0031]

[0032] In the formula, ω f For the error detection sensitivity of UAVs, κ f This represents the error detection coefficient for the drone.

[0033] S2.3, Task Search Vector Update Model

[0034] Based on the information detected by the UAV (i.e., PD and PF), the Bayesian criterion is used to dynamically update the probability of target presence. To reflect the current distribution of the target.

[0035] When the drone detects the grid... i When a target exists, that is The probability of the existence of the target in this grid is now updated as follows:

[0036]

[0037] When the drone detects the grid... i When there is no target, that is The probability of the existence of the target in this grid is now updated as follows:

[0038]

[0039] Simultaneously, based on formula (3), the grid uncertainty is updated according to the new target existence probability. Let the grid g be... i The conditions for determining whether detection is complete are:

[0040]

[0041] Therefore, combining formula (3), the grid uncertainty threshold o is obtained. th =0.0808, when the uncertainty of the raster is less than 0.th Grid detection is completed.

[0042] S2.4, Cluster information coordination model

[0043] In the process of cluster cooperative detection, each UAV independently completes its own task search vector update. In order to quickly and accurately complete the detection of the cluster, and to avoid the UAVs repeatedly detecting the grid that has been detected, the UAVs in the cluster need to obtain global task search vector information based on the task search vectors of other UAVs. Therefore, a cluster information coordination model is established.

[0044] According to formulas (6) and (7), the simultaneous detection of k UAVs can be expressed as:

[0045]

[0046] In order to simplify the calculation, formula (9) is transformed as:

[0047]

[0048] Then, the logarithmic transformation of formula (10) is obtained as:

[0049]

[0050] Let From formula (11), it can be transformed as:

[0051]

[0052] Wherein,

[0053] According to formula (12), through the cluster information coordination model, the UAV only needs to receive the probability offset transmitted by other UAVs to obtain the global search vector information. Since the communication only involves transmitting the probability offset, the communication delay can be ignored.

[0054] S3, UAV computing vector

[0055] The UAV computing vector JS = {(xt, AC), λ, f r}, wherein xt represents the cooperative computing attribute of the UAV; AC represents the cooperative computing alliance matrix; λ represents the task offloading ratio matrix; f r is the computing resource allocation ratio matrix. The calculation method of each part is introduced below.

[0056] S3.1, Cooperative computing attribute xt

[0057] ​Non-Cooperative Unmanned Aerial Vehicle (NC-UAV): When the load of the non-cooperative unmanned aerial vehicle is high, it needs to unload part of the computing task to the cooperative unmanned aerial vehicle, and can unload to multiple cooperative unmanned aerial vehicles in a certain proportion, so that multiple cooperative unmanned aerial vehicles can improve the task completion speed through cooperative computing. The cooperative computing attribute of the non-cooperative unmanned aerial vehicle is set as xt=0.

[0058] Cooperative Unmanned Aerial Vehicle (C-UAV): When the task load of the cooperative unmanned aerial vehicle is low, it can receive the task data of the non-cooperative unmanned aerial vehicle and help the non-cooperative unmanned aerial vehicle to perform cooperative computing. The cooperative computing attribute of the cooperative unmanned aerial vehicle is set as xt=1.

[0059] S3.2, Cluster Cooperative Computing Alliance AC: Considering that there may be multiple non-cooperative unmanned aerial vehicles in the cluster, all cooperative unmanned aerial vehicles can join the computing alliance of multiple non-cooperative unmanned aerial vehicles to participate in task computing, so the formation method of the cooperative computing alliance AC is as follows:

[0060]

[0061] Among them, AC ij =1 represents that the unmanned aerial vehicle j joins the cooperative computing alliance of the task i.

[0062] S3.3, Task Unloading Proportion Matrix: Define the task unloading proportion matrix λ = [λ ij ] N×N (λ ij ∈[0,1]), λ ij represents the task proportion of the task i unloaded to the unmanned aerial vehicle j by the unmanned aerial vehicle i, if the unmanned aerial vehicle j is not in the cooperative computing alliance AC i of the task i, then λ ij =0, and it needs to be ensured that Therefore, the initialization method of the task unloading proportion matrix is as follows:

[0063]

[0064] In the formula, represents the number of unmanned aerial vehicles in the alliance of the task i.

[0065] S3.4, Computing Resource Allocation Matrix: Define the remaining computing resource of the unmanned aerial vehicle i as f i , and the computing resource allocation proportion matrix as Among them represents the proportion of the remaining computing resource f i allocated to the task j by the unmanned aerial vehicle i, which needs to satisfy To ensure that the computing resources are not over-allocated, the computing resource allocation proportion matrix initialization method is as follows:

[0066]

[0067] In the formula, represents the number of cooperative computing alliances in which the unmanned aerial vehicle j participates.

[0068] S4, group detection computing power calculation method

[0069] The group detection computing power is mainly composed of cluster detection power and cluster task completion delay. The calculation methods of each part are introduced in turn as follows.

[0070] S4.1, cluster detection power

[0071] The cluster detection power is related to the change of the uncertainty of the task area and the number of detected targets, and is expressed as:

[0072]

[0073]

[0074] Wherein, μ represents the weight factor of the uncertainty change, n F represents the number of detected targets by the cluster, n g represents the number of grids of the task area, O represents the initial uncertainty of the task area, represents the initial uncertainty of each grid of the task area, O' represents the uncertainty of the task area after detection is completed, represents the uncertainty of each grid of the task area after detection is completed.

[0075] S4.2, cluster task completion delay model:

[0076] The cluster task completion delay is mainly composed of local computing delay of each subtask and cooperative computing delay. The calculation methods of each delay are introduced in turn as follows.

[0077] Local computing delay of task i:

[0078]

[0079] In the formula, λ i,i represents the offloading proportion of the unmanned aerial vehicle i to unload the computing amount C i of task i to itself, represents the allocation proportion of the unmanned aerial vehicle i to allocate the remaining computing force f i of itself to itself.

[0080]

[0081] In the formula, φi denotes the CPU cycles required for computing 1 bit data of task i, I i denotes the data volume of task i, denotes the area of grid g k k ∈vg i ; denotes the total area of the visible grid set vg i of UAV i; denotes the number of grids in vg i data s denotes the data volume generated by one detection of UAV; denotes the detection times of grid g k by UAV i.

[0082] When UAV i cannot complete the computing process of task i alone, task i needs to be split into multiple sub-computing tasks and unloaded to the collaborative UAVs in the collaborative computing alliance for computing. The collaborative computing delay of task i includes the transmission delay of sub-tasks and the computing delay of sub-tasks:

[0083]

[0084] Communication rate Rate i,j is calculated by formula (21):

[0085]

[0086] wherein, denotes that UAV i evenly distributes the bandwidth to all collaborative UAVs in the task i alliance, denotes the transmission power of UAV i, N0 is the noise power spectral density, g0 is the channel power gain of the reference distance, d i,j denotes the distance between UAV i and the collaborative UAV j in the alliance. When d i,j → 0, Rate i,j → ∞, however, this situation is impossible in reality. Considering that the infinite approach of UAVs will cause safety accidents such as collision, the hovering distance of all UAVs should be greater than a certain threshold d safe . Therefore, the task i completion delay:

[0087] t i = max(t loc,i , t col,i ) (22)

[0088] All tasks adopt parallel computing mode, so the total task completion delay T:

[0089] T = max(t i ) (23) ​

[0090] S5, the system optimization model is established as follows:

[0091] The group detection power U can be expressed as:

[0092]

[0093] The optimization goal is to maximize the group detection power:

[0094] max U (25) combined with the constraint conditions can be obtained:

[0095]

[0096] In the formula, constraint conditions 1, 2 and 3 represent the unmanned aerial vehicle detection position constraint, which means that the unmanned aerial vehicle needs to detect in the task area; constraint condition 4 represents that the hovering distance duav of the unmanned aerial vehicle in the cluster should be greater than a certain threshold dsafe constraint conditions 5 and 6 respectively represent the unmanned aerial vehicle cooperative computing alliance offloading ratio constraint; constraint conditions 7 and 8 represent the computing resource allocation ratio constraint in the unmanned aerial vehicle cooperative computing alliance.

[0097] The key of the optimization of the application is to determine an unmanned aerial vehicle cluster detection scheme to make the group detection power highest, which contains three main parts: detection topology, namely cluster unmanned aerial vehicle detection position, task offloading and computing resource allocation. The following is the intelligent optimization algorithm of the detection machine:

[0098] S5.1, initialize algorithm parameters, including unmanned aerial vehicle population size pop, maximum iteration number max_iter, convergence factor a, balance factor l1, and parameter A.

[0099] S5.2, detection topology optimization layer, initialize unmanned aerial vehicle population position by using the best point set.

[0100] According to formula (27), the best point r=(r1, r2, …, r j ) is calculated:

[0101]

[0102] Wherein, k is the smallest prime number satisfying (k-3) / 2≥dim, and dim represents the dimension of the optimization problem.

[0103] The best point set P n (i) can be expressed as:

[0104] P n (i)={({r1i},{r2i},…,{r n i})},1≤i≤pop (28)

[0105] Wherein, {r n i} represents the decimal part.

[0106] Mapping the set of good points into the search space of the population:

[0107] x i (j) = (ub j -lb j ) · P n (i) + lb j (29)

[0108] where x i represents the position of all UAVs in the cluster in the population i, ub j and lb j represent the upper and lower bounds of the jth dimension.

[0109] S5.3, based on the current detection topology, detect all detectable grids in the cluster, use the cluster information interaction model to make the grid uncertainty of all detectable grids below the threshold, output the cluster detection force F and the load calculation amount C of each UAV i to the resource optimization layer.

[0110] S5.4, resource optimization layer, according to the system optimization model (26), when the detection force F is determined, the group detection calculation force depends on the task completion delay T, the system optimization model is simplified to an optimization sub-model (30):

[0111]

[0112] Based on the block coordinate descent method, the sub-model (30) is decomposed into a task offloading optimization sub-problem (31) and a computing resource allocation optimization sub-problem (32).

[0113]

[0114] Sub-problem P2.1 solving method: the calculation factor cf can be expressed as:

[0115]

[0116] Through the following formula, the optimal task offloading scheme can be obtained when the computing resource allocation scheme is fixed.

[0117]

[0118] Sub-problem P2.2 solving method:

[0119] The local calculation factor b is expressed as:

[0120]

[0121] The collaborative calculation factor c is expressed as:

[0122]

[0123] The optimal computing resource allocation scheme when the task offloading scheme is fixed can be obtained by the following formula.

[0124]

[0125] Assemble a collaborative computing alliance to alternately optimize sub-problems P2.1 and P2.2 until the task completion time T converges to the optimum, output the best task completion delay T under the current detection topology, and the task offloading proportion matrix λ and the computing resource allocation matrix f r .

[0126] S5.5, calculate the fitness of all populations according to formula (24), and update the top 3 populations with better fitness as the current α, β and δ UAVs.

[0127] S5.6, update all population positions and parameters a, A, and C according to formula (43).

[0128]

[0129] wherein, is a random vector in [0, 2], is a random integer in [1, 2], represents the current position of the α, β and δ UAVs, represents the current position of the UAV.

[0130]

[0131] wherein r is a random number in [0, 1], and the calculation formulas of the coefficients A1 and A3 are as follows:

[0132]

[0133] wherein r1 is a random number in [0, 1], and the calculation formulas of the convergence factor a and the balance factor l1 are as follows:

[0134]

[0135] S5.7, determine whether the number of iterations has reached the maximum number of iterations, if not, increase the iteration number by one and return to step 3 to continue execution; if so, exit the iteration process and output the calculated optimal UAV cluster topology, task offloading scheme and computing resource allocation scheme.

[0136] Advantages

[0137] 1. The application is suitable for various production applications, including daily inspection and detection of landforms such as forests, grasslands, pastures, wetlands, and emergency multi-UAV groups after unexpected disasters, and rapid detection and calculation of key areas.

[0138] 2. The application can quickly complete the task compared with the traditional single machine detection scheme because the intelligent UAV cluster composed of multiple UAVs cooperates to detect the task area.

[0139] 3. In the traditional UAV cluster detection mode, the UAV relies on the ground base station and the command center for interaction, but in the harsh environment of the disaster area, the ground base station may be damaged and disabled, therefore, in the case of missing base station, the application introduces cooperative computing technology in the cluster cooperative detection scene, and through reasonable optimization of cluster resources, compared with the traditional detection method, the key information of the disaster area can be quickly and comprehensively integrated. BRIEF DESCRIPTION OF DRAWINGS

[0140] Figure 1 is a schematic diagram of the patrol detection task area grid division;

[0141] Figure 2 is a schematic diagram of the UAV cooperative computing alliance formation;

[0142] Figure 3 is a system influence factor action diagram;

[0143] Figure 4 is a flowchart of the intelligent optimization algorithm of the detection UAV group;

[0144] Figure 5 is an iteration process diagram of the intelligent optimization algorithm of the detection UAV group. DETAILED DESCRIPTION

[0145] The application will be further described in detail below in conjunction with the drawings of the specification.

[0146] As Figures 1-5 described, the application provides a UAV group detection and calculation force scheduling method based on cooperative computing, which comprises the following steps:

[0147] (1) The adaptive grid map is established based on prior information, and the steps are as follows:

[0148] (1.1) According to the detection data of the low-orbit satellite, the geographical environment information M={x M ,y M ,z M} of the target area is obtained, the number of target points to be detected n target and the target position information R={x R ,y R ,z R}.

[0149] (1.2) Adaptive grid map G = {g1, g2,..., g i} : As shown in FIG. 2, the task area is divided into grid maps with different grid sizes according to the detection weight. The target area is divided into four grids, and the grid detection weight is represented as: Figure 1

[0150]

[0151] wherein, represents the importance of the target to be detected in the grid, represents the terrain complexity, α w ,β w are corresponding coefficients. The grid with a weight exceeding a threshold W th is further divided into four sub-grids, and the above process is repeated until it cannot be further divided.

[0152] (1.3) Task search vector The task search vector is composed of a grid target existence probability and a grid uncertainty. The grid target existence probability represents the possibility of the target existing in the grid g i , and the grid uncertainty represents the uncertainty of the grid g i . The calculation methods of the grid target existence probability and the grid uncertainty are introduced in turn as follows.

[0153] Grid target existence probability: The initial grid target existence probability is normalized by the grid detection weight W i , and can be represented as:

[0154]

[0155] wherein, η represents the accuracy of prior information, and η ∈ [0, 1].

[0156] The grid uncertainty is defined as the information entropy of the target existence probability in the grid:

[0157]

[0158] (2) UAV detection vector

[0159] (2.1) UAV detection range model: The center point of the grid is defined as the information point of the grid. If the grid information point is within the detection range of the UAV and is not blocked by obstacles such as mountains, the grid is within the visible grid set of the UAV. According to formula (4), the visible grid set vg j ​:

[0160]

[0161] where, represents the detection position of UAV j, represents the grid g i information point position, Δh j,gi represents the vertical distance between UAV j and grid g i , represents the detection range radius of UAV j at Δh j,gi , α u is the detection angle, z k represents the height of the sampling point (x k , y k , z k ) equally sampled from the line connecting UAV j and grid g i , ks represents the ground height at (x k , y k ).

[0162] (2.2) UAV probability detection model:

[0163] The success detection probability represents the probability that the UAV system successfully judges the existence of the target when the target exists, which is defined as the exponential function of the square of the height difference between the UAV and the detectable grid:

[0164]

[0165] In the formula, ω s is the success detection sensitivity of the UAV; κ s is the success detection coefficient of the UAV; indicates whether there is actually a target in grid g i . When , it represents that there is actually no target in grid gi; when , it represents that there is actually a target in grid g i . indicates the specific result of the UAV detection. When b j,gi (t) = 1, it represents that UAV j detects that there is a target in grid g i at time t; when , it represents that the UAV detects that there is no target in grid g i . represents the height difference between UAV j and grid g i .

[0166] The false detection probability represents the probability that the UAV system incorrectly judges the existence of the target when the target does not exist:

[0167]

[0168] where ω f is the error detection sensitivity of UAV, κ f is the error detection coefficient of UAV.

[0169] (2.3) Task search vector update model

[0170] Based on the information detected by UAV (i.e. PD and PF), the Bayesian rule is used to dynamically update the grid target existence probability to reflect the current distribution of the target.

[0171] When UAV j detects that there is a target in grid g i , i.e. At this time, the grid target existence probability is updated as:

[0172]

[0173] When UAV j detects that there is no target in grid g i , i.e. At this time, the grid target existence probability is updated as:

[0174]

[0175] At the same time, based on formula (3), the grid uncertainty is updated according to the new target existence probability. Set grid g i The determination condition of the existence of the target is:

[0176]

[0177] Therefore, combined with formula (3), the grid uncertainty threshold o th = 0.0808, when the uncertainty of the grid is lower than o th , it means that the detection of the grid is completed.

[0178] (2.4) Cluster information coordination model

[0179] In the process of cluster cooperative detection, each UAV independently completes the update of its own task search vector In order to quickly and accurately complete the detection of the cluster and avoid the repeated detection of the UAV on the grid which has completed the detection, the UAV in the cluster needs to obtain the global task search vector information based on the task search vector of other UAVs, therefore, the cluster information coordination model is established.

[0180] From formulas (7) and (8), it can be seen that the simultaneous detection of k UAVs can be represented as:

[0181]

[0182] To simplify the calculation, formula (10) is transformed as:

[0183]

[0184] Then, formula (11) is transformed as:

[0185]

[0186] Let Thus, formula (12) can be transformed as:

[0187]

[0188] Where,

[0189] According to formula (13), through the cluster information coordination model, the unmanned aerial vehicle only needs to receive the probability offset amount delivered by other unmanned aerial vehicles That is, the global search vector information can be obtained. Since the communication only involves the delivery of the probability offset amount, the communication delay can be ignored.

[0190] (3) Unmanned aerial vehicle computing vector

[0191] The unmanned aerial vehicle computing vector JS = {(xt, AC), λ, f r}, where xt represents the cooperative computing attribute of the unmanned aerial vehicle; AC represents the cooperative computing alliance matrix; λ represents the task offloading ratio matrix; f r Table Computing resource allocation ratio matrix. The calculation method of each part is introduced below.

[0192] (3.1) Cooperative computing attribute xt

[0193] Non-cooperative unmanned aerial vehicle (NC-UAV): when the load is high, it needs to offload part of the computing task to cooperative unmanned aerial vehicles, and can offload to multiple cooperative unmanned aerial vehicles according to a certain proportion, so that multiple cooperative unmanned aerial vehicles can improve the task completion speed through cooperative computing. The cooperative computing attribute of the non-cooperative unmanned aerial vehicle is set as xt = 0.

[0194] Cooperative unmanned aerial vehicle (C-UAV): when the task load is low, it can receive the task data of the non-cooperative unmanned aerial vehicle and help the non-cooperative unmanned aerial vehicle to perform cooperative computing. The cooperative computing attribute of the cooperative unmanned aerial vehicle is set as xt = 1.

[0195] (3.2) Cluster cooperative computing alliance AC: as Figure 2As shown, considering that there can be multiple non-cooperative unmanned aerial vehicles in the cluster, all cooperative unmanned aerial vehicles can simultaneously join the computing alliance of multiple non-cooperative unmanned aerial vehicles to participate in task computing, and therefore the method for establishing the cooperative computing alliance AC is as follows:

[0196]

[0197] wherein AC ij = 1 represents that the unmanned aerial vehicle j joins the cooperative computing alliance of the task i.

[0198] (3.3) Task offloading proportion matrix: define the task offloading proportion matrix λ = [λ ij ] N×N (λ ij ∈ [0, 1]), λ ij represents the task proportion of the unmanned aerial vehicle i to be offloaded to the unmanned aerial vehicle j, if the unmanned aerial vehicle j is not in the cooperative computing alliance AC i of the task i, then λ ij = 0, and it needs to be ensured that Therefore, the task offloading proportion matrix initialization method is as follows:

[0199]

[0200] In the formula, represents the number of unmanned aerial vehicles in the alliance of the task i.

[0201] (3.4) Computing resource allocation matrix: define the remaining computing resource of the unmanned aerial vehicle i as f i , and the computing resource allocation proportion matrix is wherein λ represents the proportion of the remaining computing resource f i of the unmanned aerial vehicle i allocated to the task j, and it needs to be ensured that so as to ensure that the computing resource is not over-allocated, and therefore the computing resource allocation proportion matrix initialization method is as follows:

[0202]

[0203] In the formula, represents the number of cooperative computing alliances participated by the unmanned aerial vehicle j.

[0204] (4) Group detection computing power calculation method

[0205] The group detection computing power is mainly composed of the cluster detection power and the cluster task completion time delay, and the calculation methods of the respective parts are described below.

[0206] (4.1) Cluster detection power

[0207] The cluster detection power is related to the change of the uncertainty of the task area and the number of detected targets, and is represented as:

[0208]

[0209]

[0210] where μ denotes the weight factor of uncertainty change, n F denotes the number of targets detected by the cluster, n g denotes the number of grids of the task area, O denotes the initial uncertainty of the task area, denotes the initial uncertainty of each grid of the task area, O' denotes the uncertainty of the task area after detection is completed, denotes the uncertainty of each grid of the task area after detection is completed.

[0211] (4.2) Cluster task completion delay model:

[0212] The cluster task completion delay is mainly composed of local computing delay and collaborative computing delay of each subtask. The calculation method of each delay is introduced as follows.

[0213] (4.3) Local computing delay of task i:

[0214]

[0215] where λ i,i denotes the computing amount C i of task i unloaded to UAV i by the unloading ratio of UAV i, denotes the remaining computing force f i of UAV i allocated to UAV i by the allocation ratio of UAV i.

[0216]

[0217] where φ i denotes the CPU cycle required for computing 1 bit of data of task i, I i denotes the data amount of task i, denotes the area of grid g k , g k ∈vg i ; denotes the total area of the visible grid set vg i of UAV i; denotes the number of grids in vg i ; data s denotes the data amount generated by UAV detection once; denotes the detection frequency of UAV i on grid g k .

[0218] (4.4) When the UAV i cannot complete the computing task i alone, the task i needs to be split into multiple sub-computing tasks and unloaded to the cooperative UAVs in the cooperative computing alliance for computing. The cooperative computing delay of task i includes the transmission delay of the sub-tasks and the computing delay of the sub-tasks:

[0219]

[0220] Communication rate Rate i,j Calculated by formula (22):

[0221]

[0222] Wherein, indicates that the UAV i evenly allocates the bandwidth to all cooperative UAVs in the task i alliance, indicates the transmission power of the UAV i, N0 is the noise power spectral density, g0 is the channel power gain of the reference distance, d i,j indicates the distance between the UAV i and the cooperative UAV j in the alliance. When d i,j → 0, Rate i,j → ∞, however, this situation is impossible in reality. Considering that the UAVs will collide and other safety accidents when approaching infinitely, the hovering distance of all UAVs should be greater than a certain threshold d safe . Therefore, the task i completion delay:

[0223] t i = max(t loc,i , t col,i ) (23)

[0224] All tasks use parallel computing, so the total task completion delay T is:

[0225] T = max(t i ) (24)

[0226] (5) The system optimization model is established as follows:

[0227] The system group detection computing power U can be represented as:

[0228]

[0229] The optimization goal is to maximize the system group detection computing power:

[0230] max U (26) Combined with the constraint conditions, we can get:

[0231]

[0232] In the formula, constraint conditions 1, 2 and 3 represent the UAV detection position constraints, indicating that the UAV needs to detect in the task area; constraint condition 4 represents that the hovering distance duav of the UAV in the swarm should be greater than a certain threshold dsafe; constraint conditions 5 and 6 represent the UAV cooperative computing alliance offloading proportion constraints; and constraint conditions 7 and 8 represent the computing resource allocation proportion constraints in the UAV cooperative computing alliance.

[0233] As can be seen from formula (25), the swarm reconnaissance computing power depends on the detection power and the task completion time delay. As shown in FIG. 5, the number of targets discovered under different detection topologies and the uncertain changes of the task area are different, and the swarm reconnaissance power and the task computing load of each UAV are also different. When the detection topology is determined, the task completion time delay is related to the task offloading in the swarm and the computing resource allocation in the swarm. Therefore, the key to be optimized by the system is to determine a UAV swarm reconnaissance scheme including the detection topology, i.e., the UAV swarm detection position, the task offloading and the computing resource allocation, so as to maximize the swarm reconnaissance computing power. The following describes a reconnaissance UAV swarm intelligent optimization algorithm: Figure 3

[0234] The working theory of the reconnaissance UAV swarm intelligent optimization (RUSIO) algorithm is as follows: in the process of executing the task, different reconnaissance UAV swarms have different swarm reconnaissance computing powers. The first three reconnaissance UAV swarms with the highest swarm reconnaissance computing power are denoted as α-uavs, β-uavs and δ-uavs, and the remaining reconnaissance UAV swarms are denoted as ω-uavs. The α-uavs, β-uavs and δ-uavs guide the remaining reconnaissance UAV swarms to explore in the task area, and the positions of the α-uavs, β-uavs and δ-uavs are iteratively updated to obtain the optimal solution.

[0235] As shown in FIG. 6, the specific steps of the reconnaissance UAV swarm intelligent optimization algorithm are as follows: Figure 4

[0236] (5.1) Initialize the algorithm parameters, including the UAV population size pop, the maximum number of iterations max iter, the convergence factor a and the balance factor l1.

[0237] (5.2) Initialize the UAV swarm positions by using the set of optimal points.

[0238] According to formula (28), the optimal points r = (r1, r2, …, r j ) are calculated:

[0239]

[0240] wherein k is the smallest prime number satisfying (k-3) / 2≥dim, and dim represents the dimension of the optimization problem.​​

[0241] R n (i), which can be expressed as:

[0242] R n (i) = {({r1i}, {r2i},..., {r n i})}, 1≤i≤pop (29)

[0243] where {r n i} represents the decimal part.

[0244] Map the good point set to the search space of the population:

[0245] x i (j) = (ub j -lb j )·R n (i) + lb j (30)

[0246] where x i represents the position of all drones in the cluster in the population i, ub j and lb j represent the upper and lower bounds of the jth dimension.

[0247] (5.3) Based on the current detection topology, detect all detectable grids in the cluster, use the cluster information interaction model to make the grid uncertainty of all detectable grids below the threshold, output the cluster detection force F and the load calculation amount C of each drone i to the resource optimization layer.

[0248] (5.4) The optimization steps of the resource optimization layer are as follows:

[0249] From the system optimization model (27), when the detection force F is determined, the system cluster detection calculation force depends on the task completion delay T, and the system optimization model is simplified to the optimization sub-model (31):

[0250]

[0251] Based on the block coordinate descent method, the sub-model (31) is decomposed into the task offloading optimization sub-problem (32) and the computing resource allocation optimization sub-problem (33).

[0252]

[0253] The optimization steps of the sub-problem P2.1: the calculation factor cf can be expressed as:

[0254]

[0255] The optimal task offloading scheme can be obtained when the computing resource allocation scheme is fixed by the following formula.

[0256]

[0257] The sub-problem P2.2 optimization step is:

[0258] The local computing factor b is represented as:

[0259]

[0260] The collaborative computing factor c is represented as:

[0261]

[0262] The optimal computing resource allocation scheme can be obtained when the task offloading scheme is fixed by the following formula.

[0263]

[0264] Assemble a collaborative computing alliance, alternately optimize sub-problems P2.1 and P2.2 until the task completion time T converges to the optimal, output the best task completion delay T under the current detection topology, and the task offloading proportion matrix λ and the computing resource allocation matrix f r .

[0265] (5.5) Calculate the fitness of all populations according to formula (25), and update α, β and δ drones to the top 3 populations with better fitness.

[0266] (5.6) Update all population positions and parameters a, A, C according to the following formula.

[0267]

[0268] In the formula, is a random vector in [0, 2], is a random integer in [1, 2], represents the current position of α, β and δ drones, represents the current position of the drone.

[0269]

[0270] In the formula, r is a random number in [0, 1], and the calculation formulas of coefficients A1 and A3 are as follows:

[0271]

[0272] In the formula, r1 is a random number in [0, 1], and the calculation formulas of convergence factor a and balance factor l1 are as follows:

[0273]

[0274] (5.7)Judge whether the number of iterations reaches the maximum iteration number, if not, then the number of iterations is increased by one, and return to step 3 to continue execution; if so, then jump out of the iteration process, and output the calculated optimal UAV swarm topology, task offloading scheme and computing resource allocation scheme.

[0275] (6) Simulation result analysis:

[0276] The above algorithm is simulated and analyzed by using MATLAB to prove the effectiveness of the reconnaissance UAV swarm intelligent optimization algorithm (Reconnaissance Uav Swarm Intelligent Optimization, RUSIO). Consider a detection area of 0.5 km x 0.5 km, containing 6 targets to be detected, 6 reconnaissance UAVs, the channel bandwidth of the UAV B = 10 MHz, and the transmission power Pt = 10 dB.

[0277] The remaining simulation parameters are set as follows: the safe hovering distance between UAVs dsafe = 5 m, the channel gain power of the reference distance g0 = -50 dB, the noise power spectral density N0 = -170 dBm / Hz, the detection angle of the UAV αu = 45°, the UAV probability detection model parameters: κf = 0.9, κs = 0.9; the computing power of the UAV f = 2 GHz.

[0278] As shown in Figure 5 Under the reconnaissance UAV swarm intelligent optimization algorithm, the swarm detection computing power becomes 2624.72 bit / s, and the calculated swarm detection power is 5.4648 x 10 4 bit, and the task completion delay is 20.8343 s. Compared with the traditional particle swarm optimization algorithm (Particle Swarm Optimization, PSO), the swarm detection computing power is increased by 254.6%.

Claims

1. A method for scheduling computing power for unmanned aerial vehicle (UAV) swarm reconnaissance based on collaborative computing, characterized in that, include: S1. Establish an adaptive grid map based on initial prior information and obtain the task search vector. Where j represents the UAV number and i represents the grid number; Let g be the probability of the presence of a grid target, representing the probability that UAV j detects grid g. i The probability of the existence of a target, grid uncertainty This indicates that the drone j detected the grid g. i Uncertainty surrounding the situation; Adaptive rasterized map G = {g1, g2, ..., g i The task area is divided into a grid map of varying grid sizes based on detection weights. The target area is divided into four grids, and the grid detection weights are expressed as follows: in, Indicates the importance of target detection within the grid. Indicates terrain complexity. α w ,β w For the corresponding coefficients, the weights exceeding the threshold W are... th The grid is further divided into four sub-grids, and the above process is repeated until it can no longer be divided; Grid target existence probability: Initial grid target existence probability Grid detection weight W i Normalization yields: in, η represents the accuracy of prior information, η∈[0,1]; Grid uncertainty is defined as the information entropy of the probability of a target existing within that grid: S2. Analyze the composition of UAV detection vectors, including UAV detection range, UAV probability detection model, task search vector update model, and establish UAV swarm information collaboration model. S3. Analyze the composition of UAV computational vectors, including UAV collaborative computing attributes, cluster collaborative computing alliance, task offloading ratio matrix and computing resource allocation matrix; S4. Analyze the composition of the cluster detection power, including the cluster detection power and the cluster task completion latency, and obtain the calculation method of the cluster detection power; S5. With the goal of maximizing the computing power of the swarm detection, a system optimization model is established, and the detection location and resource allocation of the UAV swarm are jointly optimized through the intelligent optimization algorithm of the detection swarm. The specific steps of step S4 are as follows: (a) Calculation method of cluster detection computing power: The cluster detection computing power mainly consists of cluster detection power and cluster task completion latency. The calculation methods of each part are explained below. (b) Cluster detection capability: Cluster detection capability is related to the changes in uncertainty in the mission area and the number of targets detected, and is expressed as: Where μ represents the weighting factor for the uncertainty variation, and n F n represents the number of targets detected by the cluster. g The number of grid cells in the task region is represented by Ο, which represents the initial uncertainty of the task region. O' represents the initial uncertainty of each grid cell in the mission area, and O' represents the uncertainty of the mission area after detection is completed. This represents the uncertainty of each grid cell in the mission area after detection is completed; (c) Cluster task completion latency model: The cluster task completion latency mainly consists of the local computation latency of each subtask and the collaborative computation latency. The calculation methods for each latency are described below: Local computation latency of task i: In the formula, λ i,i This indicates that the computational cost C for task i is borne by drone i. i The percentage of uninstallation performed on itself. This indicates that drone i will use its remaining computing power f i The proportion allocated to itself In the formula, φ i I represents the CPU cycles required to compute 1 bit of data for task i. i This represents the amount of data for task i. Represents grid g k The area, and g k ∈vg i , Vg represents the set of visible grids for drone i. i Total area, Indicates vg i Number of grid cells in the middle, data s This indicates the amount of data generated in a single drone detection operation. Indicates that the drone i is paired with the grid g. k The number of times detected; The collaborative computation latency of task i includes the transmission latency of subtasks and the computation latency of subtasks: Communication Rate i,j Calculated using formula (14): in, This means that drone i will distribute bandwidth evenly among all cooperating drones in the task i coalition. Let N0 be the transmit power of UAV i, g0 be the noise power spectral density, g0 be the channel power gain at the reference range, and d be the transmit power of UAV i. i,j This represents the distance between drone i and its collaborating drone j within the alliance; when d i,j When →0, Rate i,j →∞, however, this situation is unlikely to occur in reality; considering that drones approaching too closely could cause collisions and other safety accidents, the hovering distance of all drones should be greater than a certain threshold d. safe Therefore, the completion delay of task i is: t i =max(t loc,i ,t col,i ) (15) All tasks are computed in parallel, therefore the total task completion delay T is: T=max(t i ) (16)。 2. The method for scheduling computing power for UAV swarm reconnaissance based on collaborative computing according to claim 1, characterized in that, The specific steps of step S2 are as follows: (a) UAV detection vector ZC j ={vg j ,(PD j ,PF j ),Y j }, where vg j PD represents the set of detectable grids for drone j. j PF represents the probability of successful detection. j Y represents the probability of error detection; the probability of successful detection and the probability of error detection are collectively referred to as the UAV probabilistic detection model. j This represents the task search vector of drone j; (b) Probabilistic detection model for unmanned aerial vehicles: The probability of successful detection represents the probability that the UAV system successfully determines the presence of a target when the target is present. It is defined as an exponential function of the square of the altitude difference between the UAV and the detectable grid. In the formula, Defined as the probability of successful detection. Indicates grid g i Does the target actually exist in China? It indicated the specific results of the drone's detection, ω s For successful drone detection sensitivity, κ s The success rate of drone detection. Represents the relationship between drone j and grid g. i The height difference between them, vg j Let j represent the set of detectable grid cells for drone j; Error detection probability represents the probability that the UAV system incorrectly determines the existence of a target when the target does not exist: In the formula, ω f For the error detection sensitivity of UAVs, κ f The error detection coefficient for the drone; (c) Cluster Information Collaboration Model During the cluster-based collaborative detection process, each drone independently completes its own task search vector. To ensure the cluster can quickly and accurately complete detection and avoid repeated detection of already detected grid cells, drones in the cluster need to obtain global task search vector information based on the task search vectors of other drones. Therefore, a cluster information collaboration model is established: According to Bayes' theorem, the simultaneous detection by k drones can be represented as: To simplify the calculation, formula (6) is transformed as follows: make Therefore, formula (7) can be transformed into: As can be seen from formula (8), through the cluster information collaboration model, the UAV only needs to receive the probability offset transmitted by other UAVs. The global search vector information can be obtained. Since the communication only involves transmitting probability offsets, the communication latency can be ignored.

3. The method for scheduling computing power for UAV swarm reconnaissance based on collaborative computing according to claim 1, characterized in that, The specific steps of step S3 are as follows: (a) UAV calculation vector The UAV calculates the vector JS = {(xt, AC), λ, f}. r }, where xt represents the collaborative computing attribute of the UAV, AC represents the collaborative computing alliance matrix, λ represents the task offloading ratio matrix, and f r The table calculates the resource allocation ratio matrix; (b) Collaborative computing attribute xt Non-Cooperative Unmanned Aerial Vehicle (NC-UAV): When its own load is high, it needs to offload some of its computing tasks to cooperative drones. It can offload some of its computing tasks to multiple cooperative drones in a certain proportion, so that multiple cooperative drones can cooperate to improve the task completion speed. Set the cooperative computing attribute of non-cooperative drones to 0. Cooperative Unmanned Aerial Vehicle (C-UAV): When its own workload is low, it can receive task data from non-cooperative drones and help non-cooperative drones perform cooperative calculations. The cooperative calculation attribute of the cooperative drone is set to xt=1. (c) Cluster Collaborative Computing Alliance Formation: Considering that there may be multiple non-collaborative drones in the cluster, in order to improve the cluster resource utilization, all collaborative drones can dynamically join the computing alliance of multiple non-collaborative drones at the same time to participate in task computing, taking into account their own load and collaborative benefits.

4. The method for scheduling computing power for UAV swarm reconnaissance based on collaborative computing according to claim 1, characterized in that, S5 proposes an intelligent optimization algorithm for detection swarms, which optimizes the detection computing power of the UAV swarm by alternating between the detection topology optimization layer and the resource optimization layer. (a) Initialize algorithm parameters, including UAV swarm size pop, maximum number of iterations max_iter, convergence factor a, balance factor l1, and parameter A; (b) In the detection topology optimization layer, the intelligent optimization algorithm for the detection swarm uses a set of optimal points to initialize the positions of the UAV swarm, thereby improving the diversity of the swarm. The method for initializing the optimal point set is as follows, and the optimal points are calculated according to formula (17): In the formula, k is the smallest prime number that satisfies (k-3) / 2≥dim, and dim represents the dimension of the optimization problem; Best Points Collection P n (i) can be represented as: P n (i)={({r1i},{r2i},…,{r n i})},1≤i≤pop (18) where, {r n i} indicates taking the decimal part, and pop indicates the size of the drone swarm; Map the set of best points to the search space of the population: x i (j)=(ub j -lb j )·P n (i)+lb j (19) In the formula, x i ub represents the location of all drones in population i, i.e., the detection topology. j and lb j Indicates the upper and lower bounds of the j-th dimension; (c) Based on the current detection topology, detect all detectable grids in the detection cluster, use the cluster information interaction model to ensure that the grid uncertainty of all detectable grids is below the threshold, and output the cluster detection force F and the computational load C of each UAV. i To the resource optimization layer; (d) In the resource optimization layer, the intelligent optimization algorithm for the detection swarm adopts the idea of ​​block coordinate descent algorithm to decouple the original optimization problem into a task unloading optimization subproblem (20) and a computing resource allocation optimization subproblem (21), which are expressed as follows: The drone swarm forms a collaborative computing alliance and solves two sub-optimization problems alternately to obtain the optimal resource allocation scheme, minimizing task completion latency. The solutions to the sub-optimization problems are as follows: Solution method for subproblem P2.1: Define the computation factor cf as: The optimal task offloading scheme can be determined using the following formula when the computational resource allocation scheme is fixed: Solution method for subproblem P2.2: Define the local computation factor b as: Define the collaborative calculation factor c as: The optimal computational resource allocation scheme can be determined using the following formula when the task unloading scheme is fixed: Output the optimal task completion delay T, task offloading ratio matrix λ, and computational resource allocation matrix f under the current detection topology. r ; (e) According to the formula Calculate and update the fitness of all populations, and update the α, β, and δ UAVs to the top 3 populations with the best fitness; (f) In the detection topology optimization layer, based on the guidance of UAVs α, β, and δ, the positions of the remaining UAVs are updated. The update steps are as follows: In the formula, It is a random number within the range [0,2]. It is a random integer in the range [1, 2]. Indicates the current positions of the α, β, and δ UAVs. Indicates the current location of the drone; In the formula, r is a random number in the range [0,1], and coefficients A1 and A3 are used to balance the exploration and development of the algorithm, and are calculated as follows: In the formula, r1 is a random number in [0,1]. The formulas for calculating the convergence factor a and the equilibrium factor l1 are as follows: (g) Determine if the loop count has reached the maximum number of iterations. If not, increment the iteration count by one and return to step 3 to continue execution. If it has reached the maximum number of iterations, exit the iteration process and output the calculated optimal UAV cluster topology, task offloading scheme and computing resource allocation scheme.

Images