Unmanned aerial vehicle cluster grouping assisted air-ground collaborative computing network resource allocation method

Abstract

The application provides a UAV cluster grouping assisted air-ground cooperative computing network resource allocation method, which comprises the following steps: step 1, constructing a mathematical model relied on in a cluster grouping assisted air-ground cooperative edge computing network; step 2, setting the number of iteration algorithm cycles for updating the system computing efficiency; step 3, obtaining an optimization problem; step 4, setting the number of double-layer iteration optimization algorithm cycles for solving the optimization problem; step 5, using a relaxation algorithm to complete the design of cluster grouping decision; step 6, completing the design of resource allocation; step 7, completing the design of the UAV trajectory in the current cycle; and step 8, updating the system computing efficiency. The convergence speed of the Dinkelbach-based cycle iteration optimization algorithm is very fast, and the efficiency is very high. The application can provide high-reliable technical guidance for trajectory optimization of a UAV cluster grouping assisted communication system.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, and in particular relates to a method for allocating air-ground collaborative computing network resources with unmanned aerial vehicle (UAV) swarm grouping assistance. Background Technology

[0002] With the rapid development of IoT services and the fast growth of new broadband services, highly intelligent applications with intensive and complex computing, such as augmented reality, 3D games, and smart factories, have seen unprecedented development in future 5G (B5G) or 6G mobile networks. Despite the rapid development of IoT and B5G / 6G wireless technologies, relying solely on remote computing task loading methods based on cloud computing technology will face significant challenges in realizing these applications. Drone swarm-assisted air-to-ground collaborative edge computing networks, as an emerging mobile computing technology, not only reduce communication latency and improve computing efficiency through collaboration among drones, but also solve the problem of limited flight time for single drones, making it a current research hotspot. Drone swarm-assisted air-to-ground collaborative edge computing networks deploy edge computing servers on computing drones. Mobile users can offload some or all of their computationally intensive tasks to computing drones for edge computing services, shortening the computing task loading distance and improving computing efficiency. Furthermore, computing drones can forward tasks they cannot process in a timely manner through relay drones within the swarm, offloading them to remote ground base stations for further processing, thus increasing the network's throughput. On the other hand, drone swarms, through reasonable grouping and partitioning, exhibit better coordination and stability. This not only improves the computing performance of mobile users but also enhances the utilization efficiency of system resources. Furthermore, traditional resource optimization techniques for drone swarm-assisted computing networks overemphasize optimizing computational bits or energy consumption. However, for drones with limited computational load and flight time, computational efficiency holds greater research value and practical significance for mobile applications that continuously enjoy high-performance computing over extended periods.

[0003] In their paper "Resource optimized multi-armed bandit-based offload path selection in edge UAV swarms" (IEEE Internet of Things, vol. 6, no. 3, pp. 4889-4896, 2019), A. Mukherjee, S. Misra, VSP Chandra, and MSObaidat et al. first proposed an offload path selection scheme based on the multi-armed bandit-based algorithm in edge computing networks assisted by UAV swarms. This scheme provides proportional load distribution within the swarm and ensures greater energy savings to obtain the optimal multi-hop path from the source UAV to the target UAV. The shortcomings of this model are: it assumes that edge computing can fully meet the needs of computing nodes, while ignoring the limitations of computing and communication resources. Furthermore, the paper only considers user-local computing and mobile edge computing (MEC), neglecting air-to-ground collaborative scenarios and failing to fully utilize the system's wireless communication, energy, and computing resources.

[0004] In their paper "Online computation offloading and traffic routing for UAV swarms in edge-cloud computing" (IEEE Transactions on Vehicular Technology, vol. 69, no. 8, pp. 8777-8791, 2020), Liu, W. Zhang, W. Chen, H. Huang, and S. Guo et al. first studied the joint optimization problem of UAV swarm workflow allocation and multi-hop routing scheduling in a dynamic environment under an UAV-edge-cloud computing air-ground collaborative model, and proposed a joint optimization algorithm to handle workflow allocation and multi-hop routing scheduling. This algorithm applies Lyapunov optimization and Markov approximation to control UAV applications and task migrations caused by environmental dynamics in real time, obtaining a near-optimal solution. The limitation of this resource allocation method is that the algorithm only focuses on task allocation and routing schemes, neglecting offloading decisions, computational resource allocation, and UAV trajectory optimization in the edge-cloud collaborative computing system, and does not fully utilize the advantages of air-ground collaboration. Summary of the Invention

[0005] Objective: The technical problem this invention aims to solve is to address the shortcomings of existing technologies by providing a resource allocation method for air-to-ground collaborative computing networks assisted by UAV swarm grouping. Since the throughput of UAV swarm transmission tasks is related to swarm grouping decisions, offload power, and UAV trajectories, and the minimum computational load required by the system is related to the user's and UAV's offload time, offload power, and computation frequency, optimal allocation of swarm grouping decisions, task offload time, task offload power, task computation frequency, and UAV trajectories can improve the system's computational efficiency while ensuring the system meets minimum service quality requirements. This method can be applied to practical air-to-ground collaborative edge computing networks assisted by UAV swarm grouping.

[0006] Includes the following steps:

[0007] Step 1: Construct the mathematical model relied upon in the cluster-group-assisted air-ground collaborative edge computing network.

[0008] Step 2, set the iteration algorithm for updating the computational efficiency η of the system to have i = 1 loop iteration;

[0009] Step 3: Complete the mathematical model using the Dinkelbach algorithm. The transformation of the optimization problem yields the optimization problem.

[0010] Step 4, Set up the optimization problem to solve The number of iterations k in the two-level iterative optimization algorithm used;

[0011] Step 5: Given the task unloading time t, task computation frequency f, task unloading power p, UAV trajectory q, and computational efficiency η, construct a mathematical model. The relaxation algorithm is used to design the cluster grouping decision α and β.

[0012] Step 6: Given the group decision α, β, UAV trajectory q, and computational efficiency η, construct a mathematical model. The resource allocation design was completed using the method of introducing variables and Taylor expansion.

[0013] Step 7: Given the group decision α, β, task unloading time t, task computation frequency f, task unloading power p, and computational efficiency η, construct a mathematical model. The design of the UAV trajectory q in this iteration is completed using the variable introduction method, Taylor expansion, and continuous convex approximation method.

[0014] Step 8: Given the group decision α, β, task unloading time t, task computation frequency f, task unloading power p, and UAV trajectory q, update the system computation efficiency η.

[0015] Furthermore, in step 1, the system parameters are set as follows:

[0016] Suppose there are U users, M computing drones, and S relay drones. The relay drones fly at an altitude of H, and the computing drones fly at an altitude of h. Let the user's label be u, the computing drone's label be m, the relay drone's label be s, and the ground base station's label be BS. The task execution time T is divided into N time slots, and the frame length of each time slot n is sufficiently small.

[0017] Step 1 includes:

[0018] Step 1-1: The channels between the drone and ground equipment, between drones, and between the drone and the ground base station are all line-of-sight channels. Establish a channel model between the user and the computing drone (i.e., the drone carrying the edge computing server):

[0019]

[0020] Where u represents the user index, ranging from 1, 2, ..., U; m represents the computational UAV index, ranging from 1, 2, ..., M; n represents the time slot index, ranging from 1, 2, ..., N; h u,m (n) represents the channel gain between user u and UAV m in the nth time slot; Υ0 is the channel gain when the distance between the user and the computing UAV is 1m. This represents the Euclidean distance between user u (labeled u) and computational drone m (labeled m) in time slot n, where h is the flight altitude of the computational drone. This means that q is equivalent to u The user position vector with label u is represented by q. m (n) represents the position vector of a computational UAV with index m in time slot n, and ||·|| represents the vector norm;

[0021] The channel model between the computational UAV and the relay UAV is as follows:

[0022]

[0023] Where s represents the index of the relay drone, ranging from 1, 2, ..., S; h m,s (n) represents the channel gain between UAV m and UAV s in the nth time slot. q represents the Euclidean distance between the computational UAV labeled m and the relay UAV labeled s in time slot n, where H is the altitude of the relay UAV and q is the distance between them. s (n) represents the position vector of the relay UAV with subscript number s in time slot n;

[0024] The channel model between the relay drone and the ground base station is as follows:

[0025]

[0026] Among them, h s (n) represents the channel gain between the UAV s and the edge base station in the nth time slot. q represents the Euclidean distance between the computational UAV with subscript s in time slot n and the ground base station BS. BS Represents the location vector of a ground base station (BS).

[0027] Steps 1-2, the computational load for local user calculations is:

[0028]

[0029] in, This represents the local computational load of user u in time slot n, where T is the total task execution cycle, and f is the local computational load. u (n) represents the computing power of ground user u in the nth time slot; L u This represents the number of CPU cycles required for a ground user to process one bit of computational task.

[0030] The computational cost of edge computing for drones is:

[0031]

[0032] in, f represents the edge computation cost of drone m in time slot n. m (n) represents the computing power of the computational UAV m in the nth time slot; L m This represents the number of CPU cycles required for a computational drone to process one bit of computational task.

[0033] Steps 1-3: Assume that the user, drone, and base station are all equipped with a single antenna. The computational task is offloaded from the user to the computing drone using TDMA (Time Division Multiple Access) technology. The following model is used to model the task offloading rate received by the computing drone from the user:

[0034]

[0035] Among them, R u,m (n) represents the total unloading amount from user u to drone m in time slot n, σ 2 p represents the noise power spectral density. u (n) represents the offload power of user u in the nth time slot; W is the spectrum bandwidth used by each UAV, and τ represents the duration of a time slot; α u,mThe variable t is a binary variable representing the unloading decision from user u to the computational drone m; u (n) is the proportion of time slots allocated to user u for task offloading by the computational UAV in the nth time slot, and log2(·) represents the logarithmic operation with base 2;

[0036] The following model is used to model the mission offload rate of a relay UAV receiving data from a computational UAV:

[0037]

[0038] Among them, R m,s (n) represents the total unloading amount from the computing drone m to the relay drone s in time slot n, p m (n) represents the unloading power of the computational UAV m in the nth time slot; β m,s Let t be a binary variable representing the unloading decision from drone m to relay drone s. m (n) is the percentage of time slots allocated to the computing drone m to the relay drone within the group for task offloading in the nth time slot of each group;

[0039] The ground base station (BS) task reception rate model is as follows:

[0040]

[0041] Among them, R s (n) represents the task offloading rate received by the ground base station BS from the relay drone s, p s (n) represents the unloading power of the relay UAV s in the nth time slot;

[0042] Steps 1-4, the energy consumed by the user's local computation is:

[0043]

[0044] in, This represents the energy consumption of user u in local computation during time slot n, and κ represents the effective capacitance coefficient of the computing chip processor.

[0045] The energy consumption of user u unloading to drone m in each time slot is:

[0046] E u,m (n)=α u,m t u (n)τp u (n) (10)

[0047] Where E u,m (n) represents the energy consumed by user u in time slot n when unloading data to drone m;

[0048] The energy consumed by edge computing in computational drones is:

[0049]

[0050] in This represents the edge computing energy consumption of drone m in time slot n.

[0051] The energy consumption for the unloading task from the computational UAV m to the relay UAV s is:

[0052]

[0053] in This represents the energy consumption of drone m unloading data to drone s in time slot n;

[0054] Steps 1-5: The drone's flight trajectory satisfies the following constraints:

[0055] ||q j (n)-q j (n-1)||≤τv max (13)

[0056]

[0057] ||q j' (n)-q j (n)||≥d min (15)

[0058]

[0059] Where j and j' represent arbitrary UAV numbers, ranging from 1, 2, ..., M+S, q j (n) represents the horizontal coordinate of drone j, q j' (n) represents the horizontal coordinate of drone j', q m (n) represents the horizontal coordinate of the drone m, q s (n) represents the horizontal coordinate of the drone s, v max This is the maximum flight speed of the drone, measured in meters per second. It is the starting point for the drone's flight. It is the final destination of the drone's flight; d min It is the minimum safe distance during drone flight; It is the farthest distance that a computational drone can deviate from the relay drone within the same group, where a group consists of a relay drone and all the computational drones connected to it.

[0060] Steps 1-6 establish the following optimization problem of maximizing the system's computational efficiency function:

[0061]

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076]

[0077]

[0078]

[0079] Where Θ(n)={α u,m ,β m,s ,f u (n),f m (n),t u (n),t m (n),p u (n),p m (n),p s (n),q j Let η(n)} represent the set of all variables in the optimization problem, and let η represent the system computational efficiency.

[0080] R(Θ(n)) represents the total computational cost of the system:

[0081]

[0082] E(Θ(n)) represents the total energy consumption of the system:

[0083]

[0084] Step 3 includes:

[0085] Step 3-1, after transformation by the Dinkelbach algorithm, optimize the problem. Equivalent to an optimization problem

[0086]

[0087] In step 4, the outer loop iteration algorithm of the two-layer iterative optimization algorithm (steps 5-1 to 7-10) is set to have a loop count of k = 0.

[0088] Step 5 includes:

[0089] Step 5-1, based on the given... and η i In the k-th iteration, this refers to the unloading time of user u in time slot n, the unloading time of computational drone m in time slot n, the local computation frequency of user u in time slot n, the edge computation frequency of computational drone m in time slot n, the unloading power of user u in time slot n, the unloading power of computational drone m in time slot n, the unloading power of relay drone s in time slot n, the horizontal position of drone j in time slot n, and the computational efficiency. If k = 0 or i = 0, then random initialization is performed. and η 0 When performing group decision optimization individually, the optimization problem is transformed by the relaxation algorithm. Equivalent to an optimization problem

[0090]

[0091] st(17a)-(17b),(17e)-(17f),(17h),(17o)-(17q), (21a)

[0092]

[0093] Step 5-2: Solve the group decision problem in Step 5-1 using the standard linear programming method, and obtain the continuous variable α. u,m ,β m,s The optimal solution;

[0094] Step 5-3, the obtained continuous variable α u,m ,β m,sThe optimal solution is forcibly restored to a binary integer solution that satisfies the constraints of the optimization problem in step 5-1, based on the basic rounding strategy.

[0095] Step 5-4, obtain the optimal grouping decision Assigning values ​​to the grouping decision in the k+1th iteration of the loop.

[0096] Step 6 includes:

[0097] Step 6-1, introduce a new variable z u (n)=t u (n)p u (n), z m (n)=t m (n)p m (n);

[0098] Step 6-2, set up a new variable equation:

[0099]

[0100]

[0101] Step 6-3: Obtain the function g through a first-order Taylor expansion. m (t m (n),z m (n)) and g s (p s (n) in any Global upper bound and

[0102]

[0103]

[0104]

[0105] Step 6-4, based on the optimization obtained in step 5 And the given and η i If k=0 or r=0, then initialize randomly. and η 0 When optimizing resources in isolation, after a first-order Taylor expansion, the problem is decoupled into a convex optimization problem.

[0106]

[0107] st(17c),(17g),(17h),(17j),(17k),(26a)

[0108]

[0109]

[0110]

[0111]

[0112]

[0113] Step 6-5: Obtain the optimal solution for resource allocation using a standard convex optimization algorithm. And assign them to

[0114] Step 7 includes:

[0115] Step 7-1: Set the inner loop iteration count of the two-layer iterative optimization algorithm to l = 0;

[0116] Step 7-2, Introduce new variables and This represents the variables related to drone m, drone s, and time slot n. Let represent the variables related to the drone s and time slot n, and satisfy:

[0117]

[0118]

[0119] Step 7-3, set up a new variable equation:

[0120]

[0121]

[0122]

[0123] Step 7-4: Obtain the function Q through a first-order Taylor expansion. m (n), Q m,s (n) and Q s (n) in any UAV trajectory The global lower bound of the location and If l = 0, then initialize randomly.

[0124]

[0125]

[0126]

[0127] The constraint sub-formulas ||q in steps 7-5 and 1-6 of the optimization problem m (n)-q s (n)|| 2 ,||q s (n)-q BS || 2 ,||q j (n)-q j (n-1)|| 2 and||q j' (n)-q j (n)|| 2 The first-order Taylor expansion at a point within the given feasible region is as follows:

[0128]

[0129]

[0130]

[0131]

[0132] Steps 7-6 are based on the optimization obtained in step 5. and the optimization obtained in step 6 When optimizing the drone trajectory alone, the problem is decoupled into a convex optimization problem.

[0133]

[0134] st(17l),(17m),(17o), (39a)

[0135]

[0136]

[0137]

[0138]

[0139]

[0140]

[0141]

[0142] Step 7-7: Solve the convex problem using MATLAB's CVX toolbox. Get the (l+1)th iteration

[0143] Steps 7-8 complete the UAV trajectory calculation iteration for the inner loop in this two-layer iterative optimization algorithm, and set l = l + 1;

[0144] Steps 7-9 are repeated, from steps 7-3 to 7-8, until the preset conditions are met. The position of the drone calculated in this case was obtained. And assign to

[0145] Steps 7-10: Complete the calculation iteration of Θ(n) of the outer loop in this double-layer iterative optimization algorithm, and set k = k + 1;

[0146] Step 7-11: Repeat steps 5-1 to 7-10 until the preset condition |R(Θ) is met. k (n))-η i E(Θ k (n))|≤ξ, solve for Θ in this calculation. k (n), and assign it to Θ i (n);

[0147] Step 8 includes:

[0148] Step 8-1: Complete the system computational efficiency η in this iterative algorithm. i The update sets i = i + 1:

[0149]

[0150] Step 8-2: Repeat steps 3-1 to 8-1 until the preset condition |R(Θ) is met. i (n))-η i E(Θ i (n))|≤δ, thus obtaining the approximate optimal solution Θ of the optimization problem in steps 1-6. opt (n).

[0151] This invention can achieve optimal resource allocation and trajectory optimization with maximum computational efficiency under an air-ground collaborative computing model, based on user energy consumption constraints, UAV data transmission constraints, causal relationships of data offloading, and UAV flight constraints.

[0152] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0153] (1) This invention proposes for the first time a method for resource allocation and trajectory optimization in an air-ground cooperative mobile edge computing network based on UAV cluster grouping;

[0154] (2) Under conditions such as limited drone flight, limited device computing power, limited device communication power and limited device data transmission power, the present invention optimizes the system's computing efficiency by adjusting the drone's flight trajectory and the device's computing resources, communication resources and transmission power.

[0155] (3) The Dinkelbach-based iterative optimization algorithm of the present invention has a fast convergence speed and high efficiency.

[0156] (4) This invention can provide highly reliable technical guidance for trajectory optimization of communication systems assisted by UAV swarm groups. Attached Figure Description

[0157] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0158] Figure 1 This is a schematic diagram of the air-ground collaborative mobile edge computing network based on UAV cluster grouping in this invention.

[0159] Figure 2 This is a simulation diagram of the cluster partitioning and UAV flight trajectory results when the maximum system computing efficiency is achieved in this invention.

[0160] Figure 3 This is a graph showing the trend of system computation efficiency changing with the minimum QoS under different optimization strategies of this invention.

[0161] Figure 4 This is a graph showing the trend of system computation efficiency changing with the minimum QoS under different parameter strategies of this invention. Detailed Implementation

[0162] like Figure 1 As shown, the air-to-ground communication channel model of the air-to-ground cooperative mobile edge computing network based on UAV swarm grouping in this invention is described as follows:

[0163] This invention considers an air-ground cooperative mobile edge computing network based on UAV swarm grouping. It assumes that due to unforeseen circumstances such as natural disasters, ground communication equipment cannot provide communication services, thus requiring the deployment of UAV swarms as mobile base stations or mobile relays to assist ground communication. It is assumed that all communication nodes at the user, UAV, and base station are equipped with single antennas, and computational tasks are offloaded from the user to the computational UAV using TDMA technology. Without loss of generality, it is assumed that the base station and user locations are fixed on the ground, the relay UAV's flight altitude is H, and the computational UAV's flight altitude is h, where H and h are the minimum altitudes required for the UAV to adapt to terrain or avoid buildings. This model includes 1 base station, M computational UAVs, S relay UAVs, and U users. The task execution time T is divided into N time slots, each time slot n has a sufficiently small frame length, and all nodes are equipped with single antennas. It is assumed that there is no direct link between the base station and users; communication between the base station and users needs to be completed using UAV relay nodes. The channels between drones and ground equipment, between drones, and between drones and ground base stations are all line-of-sight (LoS) channels.

[0164] (1.1) Channel Model

[0165] In the uplink of this invention, the channels between the UAV and ground equipment, between UAVs, and between the UAV and the ground base station are all line-of-sight channels. The channel model between the user and the computing UAV is established as follows:

[0166]

[0167] The channel model between the computational UAV and the relay UAV is established as follows:

[0168]

[0169] The channel model between the relay drone and the ground base station is established as follows:

[0170]

[0171] (1.2) Computational Model

[0172] The computational load for local user calculations is:

[0173]

[0174] The computational cost of edge computing for drones is:

[0175]

[0176] (1.3) Reachable Rate Model

[0177] Assuming that the user, drone, and base station are all equipped with a single antenna, and that the computational task is offloaded from the user to the computational drone using TDMA technology, the task offloading rate model received by the computational drone from the user can be modeled as follows, based on the proposed user transmission model:

[0178]

[0179] The task offload rate model for relay UAV missions receiving tasks from computational UAVs can be modeled as follows:

[0180]

[0181] The ground base station (BS) task reception rate model is as follows:

[0182]

[0183] (1.4) Energy Consumption Model

[0184]

[0185] The energy consumption of user u unloading to drone m in each time slot is:

[0186] E u,m (n)=α u,m t u (n)τp u (n) (10)

[0187] The energy consumed by edge computing in computational drones is:

[0188]

[0189] The energy consumption for the unloading task from the computational UAV m to the relay UAV s is:

[0190]

[0191] (1.5) Flight Model

[0192] The flight trajectory of the drone satisfies the following constraints:

[0193] ||q j (n)-q j (n-1)||≤τv max (13)

[0194]

[0195] ||q j' (n)-q j (n)||≥dmin (15)

[0196]

[0197] (1.6) Problem Description

[0198] Based on constraints such as transmission causality, user decision-making and UAV grouping, communication and computing resource constraints between users and UAVs, and UAV flight constraints, the following optimization problem is established to maximize the system's computational efficiency function:

[0199]

[0200]

[0201]

[0202]

[0203]

[0204]

[0205]

[0206]

[0207]

[0208]

[0209]

[0210]

[0211]

[0212]

[0213]

[0214]

[0215]

[0216]

[0217] Where Θ(n)={α u,m ,β m,s ,f u (n),f m (n),tu (n),t m (n),p u (n),p m (n),p s (n),q j (n)},

[0218] R(Θ(n)) represents the total computational load of the system, which consists of three parts: local computation by the user, edge computation by the computational UAV, and ground relay computation.

[0219]

[0220] E(Θ(n)) represents the total energy consumption of the system, which consists of five parts: user local computing energy consumption, computing drone edge computing energy consumption, user offloading energy consumption, computing drone offloading energy consumption, and relay drone offloading energy consumption.

[0221]

[0222] (2) Set the iteration number of the algorithm for updating the computational efficiency η of the system in this invention to i = 1;

[0223] (3) The transformation design of the fractional optimization problem in the mathematical model of step (1.6) in this invention using the Dinkelbach algorithm is as follows:

[0224] (3.1) After transformation by Dinkelbach's algorithm, the optimization problem This can be equivalent to an optimization problem.

[0225]

[0226] (3) Set up a two-layer iterative optimization algorithm to solve the outer loop iterative algorithm of the Θ(n) optimization problem in step (1.6) with a loop count of k = 0;

[0227] (5) Given the task unloading time t, task computation frequency f, task unloading power p, UAV trajectory q, and computational efficiency η, construct a mathematical model. The design of cluster grouping decisions α and β using the relaxation algorithm is as follows:

[0228] (5.1) According to the given and η iIn the k-th iteration, this refers to the unloading time of user u in time slot n, the unloading time of computational drone m in time slot n, the local computation frequency of user u in time slot n, the edge computation frequency of computational drone m in time slot n, the unloading power of user u in time slot n, the unloading power of computational drone m in time slot n, the unloading power of relay drone s in time slot n, the horizontal position of drone j in time slot n, and the computational efficiency. If k = 0 or i = 0, then random initialization is performed. and η 0 When performing group decision optimization individually, the optimization problem is transformed by the relaxation algorithm. This can be equivalent to an optimization problem.

[0229]

[0230] st(17a)-(17b),(17e)-(17f),(17h),(17o)-(17q), (21a)

[0231]

[0232] (5.2) Solve the group decision problem in step (5.1) using the standard linear programming method, and obtain the continuous variable α. u,m ,β m,s The optimal solution;

[0233] (5.3) The obtained continuous variable α u,m ,β m,s The optimal solution is forcibly restored to a binary integer solution that satisfies the constraints of the optimization problem in (5.1) according to the basic rounding strategy;

[0234] (5.4) The obtained optimal grouping decision Assigning values ​​to the grouping decision in the k+1th iteration of the loop.

[0235] (6) Given the group decision α, β, UAV trajectory q and computational efficiency η, construct a mathematical model. The resource allocation design, using the method of introducing variables and Taylor expansion, is as follows:

[0236] (6.1) Introduce a new variable z u (n)=t u (n)p u (n), z m (n)=t m (n)p m (n);

[0237] (6.2) Set up new variable equations:

[0238]

[0239]

[0240] (6.3) The function g can be obtained through a first-order Taylor expansion. m (t m (n),z m (n)) and g s (p s (n) in any Global upper bound and

[0241]

[0242]

[0243]

[0244] (6.4) Optimized according to step (5) And the given and η i If k=0 or r=0, then initialize randomly. and η 0 When optimizing resources individually, after the first-order Taylor expansion described above, the problem is decoupled into a convex optimization problem.

[0245]

[0246] st(17c),(17g),(17h),(17j),(17k), (26a)

[0247]

[0248]

[0249]

[0250]

[0251]

[0252] (6.5) The optimal solution for resource allocation can be obtained through standard convex optimization algorithms. And assign them to

[0253] (7) Given the group decision α, β, task unloading time t, task computation frequency f, task unloading power p, and computational efficiency η, construct a mathematical model. The UAV trajectory q in this iteration is designed using the variable introduction method, Taylor expansion, and continuous convex approximation method as follows:

[0254] (7.1) Set up a two-layer iterative optimization algorithm to solve the step-by-step UAV trajectory optimization problem. The inner loop iterative algorithm has a loop count of l = 0.

[0255] (7.2) The UAV trajectory q in this iteration is completed using the variable introduction method, Taylor expansion, and continuous convex approximation method. j Design of (n):

[0256] (7.3) Introducing new variables and This represents the variables related to drone m, drone s, and time slot n. Let represent the variables related to the drone s and time slot n, and satisfy:

[0257]

[0258]

[0259] (7.4) Set up new variable equations:

[0260]

[0261]

[0262]

[0263] (7.5) The function Q can be obtained through a first-order Taylor expansion. m (n), Q m,s (n) and Q s (n) in any UAV trajectory The global lower bound of the location and If l = 0, then initialize randomly.

[0264]

[0265]

[0266]

[0267] (7.6) The constraint sub-formula of the optimization problem in step (1.6) ||q m (n)-qs (n)|| 2 ,||q s (n)-q BS || 2 ,||q j (n)-q j (n-1)|| 2 and||q j' (n)-q j (n)|| 2 The first-order Taylor expansion at a point within the given feasible region is as follows:

[0268]

[0269]

[0270]

[0271]

[0272] (7.7) Based on the optimization obtained in step (5) and the optimization obtained in step (6) When optimizing the drone trajectory alone, after the above steps, the problem is decoupled into a convex optimization problem.

[0273]

[0274] st(17l),(17m),(17o), (39a)

[0275]

[0276]

[0277]

[0278]

[0279]

[0280]

[0281]

[0282] (7.8) Solving convex problems using MATLAB's CVX toolbox Get the (l+1)th iteration

[0283] (7.9) Complete the UAV trajectory calculation iteration of the inner loop in this two-layer iterative optimization algorithm, and set l = l + 1;

[0284] (7.10) Repeat steps (7.4)-(7.9) until the preset conditions are met. The position of the drone calculated in this case was obtained. And assign to

[0285] (7.11) Complete the calculation iteration of Θ(n) of the outer loop in this double-layer iterative optimization algorithm, and set k = k + 1;

[0286] (7.12) Repeat steps (5.1)-(7.11) until the preset condition |R(Θ) is met. k (n))-η i E(Θ k (n))|≤ξ, solve for Θ in this calculation. k (n), and assign it to Θ i (n);

[0287] (8) Given the group decision α, β, task unloading time t, task computation frequency f, task unloading power p, and UAV trajectory q, update the system computation efficiency η as follows:

[0288] (8.1) Complete the system computational efficiency η in this iteration algorithm. i The update sets i = i + 1:

[0289]

[0290] (8.2) Repeat steps (3.1)-(8.1) until the preset condition |R(Θ) is met. i (n))-η i E(Θ i (n))|≤δ, thus obtaining the approximate optimal solution Θ of the optimization problem in step (1.6). opt (n).

[0291] The invention will be further described below with reference to simulation conditions and results. The common simulation parameters are set as follows: Consider a UAV swarm-assisted air-to-ground cooperative mobile edge computing system. The system operates in an area of ​​200*200m. 2Forty-five ground users are randomly distributed within a two-dimensional area, with the ground base station's horizontal coordinates at [1000, 0]. Twelve drones provide service, including nine computing drones and three relay drones. Furthermore, the maximum number of users served by a drone is five, and the maximum number of computing drones in a group is five. These drones take off from the horizontal coordinate [0, 0], taking 12 seconds (40 time slots) to reach the horizontal coordinate [200, 200] at a speed not exceeding 30 m / s. The flight altitudes of the computing drones and relay drones are set to 12m and 15m, respectively. To ensure cluster cooperation, the maximum distance between computing drones and relay drones within each group is set to 20m, and the safe distance between drones is 2m. The channel gain at a distance of 1m is set to -50dB, and the bandwidth W is 20MHz. The maximum energy constraint for users is 3 joules, and the maximum computational constraint for computing drones is 10 Mbits. The noise power density is -150dB / Hz, and the coefficient κ is 10. -28 The maximum CPU frequency cycles for users and computing drones are 10. 3 cycles / bit and 10 2 cycles / bit. To ensure the quality of service of the system, the minimum QoS is set to 1 Gbits. To illustrate the performance advantages of the joint optimization scheme in this invention, three benchmarks are defined for comparison below: 1) Nearest distance-based grouping scheme: Each computing drone is associated with the nearest relay drone, and each user is associated with the nearest computing drone. All computing drones associated with the same relay drone form the same group; 2) Fixed trajectory scheme: All drones fly along a prescribed trajectory without optimization, and the cluster grouping and resource allocation subproblems are optimized on this basis; 3) No-aerial computing scheme: The computing resources of each computing drone are set to f m =0, then the user's computing tasks are obtained by the user's local machine and the ground BS through the cloud server. Figure 2 This diagram illustrates the cluster partitioning and simulated drone flight trajectories to achieve maximum system computational efficiency. Drones belonging to the same group are depicted with lines of the same color, providing service to users marked with an asterisk (asterisk). Figure 2 As can be seen, in order to obtain better channel conditions and achieve as much data offloading as possible, the trajectories of each group are biased towards the relevant users. Meanwhile, within the same group, the relay drones are located close to each other. Furthermore, all users served by the same group are distributed throughout the trajectory, rather than concentrated in one area, to maintain a high level of computational efficiency for each time slot. Figure 3The graph shows the trend of system computational efficiency as a function of minimum computational bits (lowest QoS) under different optimization strategies. The computational efficiency of the proposed algorithm and the other three benchmark schemes was significantly improved in the first six iterations, and then gradually stabilized at a relatively fast pace. The performance of the proposed algorithm is superior to the other three benchmark schemes, which also reflects the advantages of the joint design of cluster partitioning, resource allocation, and UAV trajectory. Figure 4 The graph illustrates the trend of system computational efficiency with minimum computation bits (lowest QoS) under different parameter strategies. When the minimum computation bits (lowest QoS) increase, the computational efficiency of the algorithm proposed in this invention, encompassing all different parameter scenarios, first decreases significantly, then the downward trend gradually slows. This is because when the minimum QoS is small, the UAV's computational resources are sufficient, and task processing mainly involves local computation by the user and MEC computation within the UAV, which consumes very little energy. As the minimum computation bits (lowest QoS) increase, the UAV's computational resources gradually become insufficient. Therefore, more and more computation bits need to be relayed to remote base stations (BS) via relay UAVs for further processing, a process that consumes a lot of energy. Thus, computational efficiency shows a significant downward trend. However, when the minimum computation bits (lowest QoS) are large, the system mainly processes computation by offloading it to remote base stations (BS), and the system remains in a high-energy-consumption computational state, therefore the decrease in computational efficiency is smaller. It can also be observed that when M=9, the system's computational efficiency increases with the increase in the number of S groups. This is because as S increases, the number of computing drones associated with the same relay drone decreases, thereby increasing the offloading time allocated to each computing drone and improving the overall system's computational efficiency. It's important to note that when S=1, all computing drones can forward computational tasks through a single relay drone, making the competition for offloading time particularly fierce. Therefore, in this case, increasing the number of relay drones S significantly improves computational efficiency. Furthermore, when S=3, the system's computational efficiency also increases with the increase in the number of computing drones M. More computing drones mean more computational and communication resources, which is beneficial for increasing the total number of computing bits with low energy consumption. Moreover, when S=3 and M=9, the computational efficiency is higher than in other cases, demonstrating the importance of drone swarm grouping in drone swarm-assisted networks.

Claims

1. A method for allocating air-to-ground collaborative computing network resources with UAV swarm grouping assistance, characterized in that, Includes the following steps: Step 1: Construct the mathematical model relied upon in the cluster-group-assisted air-ground collaborative edge computing network. ; Step 2, Configure and update system computing efficiency The iterative algorithm iterates through the loop with i = 1. Step 3: Complete the mathematical model using the Dinkelbach algorithm. The transformation of the optimization problem yields the optimization problem. ; Step 4, Set up the optimization problem to solve. The number of iterations k in the two-level iterative optimization algorithm used; Step 5, specify the task unload time. Task calculation frequency Task offloading power Drone trajectory and computational efficiency Constructing a mathematical model The relaxation algorithm is used to complete the cluster grouping decision. Design; Step 6, given group decision Drone trajectory and computational efficiency Constructing a mathematical model The resource allocation design was completed using the method of introducing variables and Taylor expansion; Step 7, given group decision Task unloading time Task calculation frequency Task offloading power and computational efficiency Constructing a mathematical model The UAV trajectory in this iteration is completed using the variable introduction method, Taylor expansion, and continuous convex approximation method. Design; Step 8, given group decision Task unloading time Task calculation frequency Task offloading power and drone trajectory Update system computing efficiency ; Step 1 includes: Step 1-1: The channels between the drone and ground equipment, between drones, and between the drone and the ground base station are all line-of-sight channels. Establish a channel model between the user and the computing drone (i.e., the drone carrying the edge computing server): （1） Where u represents the user index, ranging from 1, 2, ..., U; m represents the computational UAV index, ranging from 1, 2, ..., M; n represents the time slot index, ranging from 1, 2, ..., N; This represents the channel gain between user u and drone m in the nth time slot; This is the channel gain when the distance between the user and the computing drone is 1m. This represents the Euclidean distance between user u (labeled u) and computational drone m (labeled m) in time slot n, where h is the flight altitude of the computational drone. This means that it is equivalent to, This represents the user's location vector labeled u. This represents the position vector of a computational UAV with index m in time slot n. Represents the vector norm; The channel model between the computational UAV and the relay UAV is as follows: （2） Where s represents the index of the relay drone, ranging from 1, 2, ..., S; This represents the channel gain between UAV m and UAV s in the nth time slot. H represents the Euclidean distance between the computational UAV labeled m and the relay UAV labeled s in time slot n, where H is the altitude of the relay UAV. This represents the position vector of the relay UAV with index s in time slot n; The channel model between the relay drone and the ground base station is as follows: （3） in, This represents the channel gain between the UAV s and the edge base station in the nth time slot. This represents the Euclidean distance between the computational UAV (subscript s) in time slot n and the ground base station BS. Represents the location vector of a ground base station (BS). Steps 1-2, the computational load for user-local calculations is: （4） in, This represents the local computational load of user u in time slot n, where T is the total task execution cycle. The computing power of ground user u in the nth time slot; This represents the number of CPU cycles required for a ground user to process one bit of computational task. The computational cost of edge computing for drones is: （5） in, This represents the computational cost of drone m at the edge of time slot n. Let m be the computing power of a computational unmanned aerial vehicle (UAV) in the nth time slot. This represents the number of CPU cycles required for a computational drone to process one bit of computational task. Steps 1-3: Assume that the user, drone, and base station are all equipped with a single antenna. The computational task is offloaded from the user to the computational drone using TDMA technology. The following model is used to model the task offloading rate received by the computational drone from the user: （6） in, This represents the total unloading amount from user u to drone m in time slot n. Represents the noise power spectral density. This represents the offload power of user u in the nth time slot; W is the spectrum bandwidth used by each drone. It represents the duration of a time slot; This is a binary variable representing the unloading decision from user u to the computational drone m; It represents the percentage of time slots allocated to user u for task offloading by the computational UAV in the nth time slot. This represents a logarithmic operation with base 2. The following model is used to model the mission offload rate of a relay UAV receiving data from a computational UAV: （7） in, This represents the total unloading volume from the computing drone m to the relay drone s in time slot n. This represents the unloading power of the computational UAV m in the nth time slot; This is a binary variable representing the unloading decision from drone m to relay drone s. It is the percentage of time slots allocated to the computing drone m in the nth time slot for task offloading to the relay drone within the group; The ground base station (BS) task reception rate model is as follows: （8） in, This represents the task offloading rate received by the ground base station (BS) from the relay drone (s). This represents the unloading power of relay drone s in the nth time slot; Steps 1-4, the energy consumed by the user's local computation is: （9） in, This represents the energy consumption calculated locally by user u in time slot n. The effective capacitance coefficient of the computing chip processor; The energy consumption of user u unloading to drone m in each time slot is: （10） in This represents the energy consumed by user u in unloading data to drone m in time slot n; The energy consumed by edge computing in computational drones is: （11） in This represents the edge computing energy consumption of drone m in time slot n; The energy consumption for the unloading task from the computational UAV m to the relay UAV s is: （12） in This represents the energy consumption of drone m unloading data to drone s in time slot n; Steps 1-5: The drone's flight trajectory satisfies the following constraints: （13） （14） （15） （16） Where j and j' represent arbitrary UAV numbers, ranging from 1, 2, ..., M+S. This represents the horizontal coordinate of drone j. This represents the horizontal coordinate of drone j'. This represents the horizontal coordinate of drone m. This represents the horizontal coordinate of the drone 's'. This is the maximum flight speed of the drone, measured in meters per second. It is the starting point for the drone's flight. It is the final destination of the drone's flight; It is the minimum safe distance during drone flight; This refers to the furthest distance that a computational drone within the same group can deviate from the relay drone within the same group. Steps 1-6 establish the following optimization problem of maximizing the system's computational efficiency function: ； （17a） （17b） （17c） （17d） （17e） （17f） （17g） （17h） （17i） （17j） （17k） （17l） （17m） （17n） （17o） （17p） （17q） in , represents the set of all variables in the optimization problem. Indicates the system's computational efficiency; Represents the total computational load of the system: （18） Represents the total energy consumption of the system: （19） Where max represents the maximization operation, s.t. represents the constraint condition, formula (17a) means that the total amount of computing tasks processed by the computing drone m before the Vth time slot cannot exceed the total amount of computing tasks successfully unloaded to drone m before the current time slot; formula (17b) means that the total number of computing tasks forwarded by the relay drone before the Vth time slot must be less than the total number of computing tasks successfully unloaded to the relay drone before the current time slot; formula (17c) means that within the task completion cycle, the total amount of computing tasks of the system is not less than the minimum computing bits required to guarantee the quality of computing service; formulas (17d) and (17e) mean that each computing drone can only associate with a maximum of Each relay drone can serve a maximum of [number] users. Each computing drone; Formula (17f) indicates that each user and each computing drone can only be served by one computing drone and one relay drone respectively; Formulas (17g) and (17h) mean that the total time for all users associated with the same computing drone or all computing drones associated with the same relay drone to unload their tasks cannot exceed the duration of each time slot; Formulas (17i), (17j), and (17k) are constraints on the range of variable values; Formula (17l) is the maximum flight speed constraint for the drone; Formula (17m) is the initial and final position constraint for the drone; Formula (17n) ensures that no collisions occur between drones during task completion; Formula (17o) ensures that computing drones in the same group do not deviate too far from the relay drone; Formula (17p) indicates that a user's energy consumption cannot exceed its maximum energy consumption. Formula (17q) represents the computational bit constraint, meaning that the total number of computational bits processed by each computational UAV in each time slot cannot exceed the maximum number of computational bits. .

2. The method according to claim 1, characterized in that, Step 3 includes: Step 3-1, after transformation by the Dinkelbach algorithm, optimize the problem. Equivalent to an optimization problem : （20）。 3. The method according to claim 2, characterized in that, In step 4, the number of iterations of the outer loop of the two-layer iterative optimization algorithm is set to k = 0.

4. The method according to claim 3, characterized in that, Step 5 includes: Step 5-1, based on the given... and That is, in the k-th iteration, the unloading time of user u in time slot n, the unloading time of computational drone m in time slot n, the local computation frequency of user u in time slot n, the edge computation frequency of computational drone m in time slot n, the unloading power of user u in time slot n, the unloading power of computational drone m in time slot n, the unloading power of relay drone s in time slot n, the horizontal position of drone j in time slot n, and the computational efficiency; if k=0 or i=0, then randomly initialize. , , , , , , , as well as When performing group decision optimization individually, the optimization problem is transformed by the relaxation algorithm. Equivalent to an optimization problem : （21a） （21b） Step 5-2: Solve the group decision problem in Step 5-1 using the standard linear programming method, and obtain the continuous variables. The optimal solution; Step 5-3, the obtained continuous variables The optimal solution is forcibly restored to a binary integer solution that satisfies the constraints of the optimization problem in step 5-1, based on the basic rounding strategy. Step 5-4, obtain the optimal grouping decision Assigning values ​​to the grouping decision in the k+1th iteration .

5. The method according to claim 4, characterized in that, Step 6 includes: Step 6-1, Introduce new variables , ; Step 6-2, set up a new variable equation: ， （22） ； （23） Step 6-3: Obtain the function through a first-order Taylor expansion. and In any Global upper bound and : ； ； ； ， （24） ； ， （25） Step 6-4, based on the optimization obtained in step 5 , and the given and If k=0 or r=0, then initialize randomly. and When optimizing resources in isolation, after a first-order Taylor expansion, the problem is decoupled into a convex optimization problem. : ； （26a） （26b） （26c） （26d） （26e） （26f） Step 6-5: Obtain the optimal solution for resource allocation using a standard convex optimization algorithm. , , , , , , and assign them to , , , , , , .

6. The method according to claim 5, characterized in that, Step 7 includes: Step 7-1: Set the inner loop iteration count of the two-layer iterative optimization algorithm to l = 0; Step 7-2, Introduce new variables and , This represents the variables related to drone m, drone s, and time slot n. Let represent the variables related to the drone s and time slot n, and satisfy: （27） （28） Step 7-3, set up a new variable equation: （29） （30） （31） Step 7-4: Obtain the function through a first-order Taylor expansion. , and In any drone trajectory The global lower bound of the location , and If l = 0, then initialize randomly. : ； ； ； （32） ； ； ； （33） ； ； ； （34） Constraints of the optimization problem in steps 7-5 and 1-6 , , and The first-order Taylor expansion at a point within the given feasible region is as follows: （35） （36） （37） （38） Steps 7-6 are based on the optimization obtained in step 5. and the optimization obtained in step 6 , , , , , , When optimizing the drone trajectory alone, the problem is decoupled into a convex optimization problem. : ； （39a） （39b） （39c） （39d） （39e） （39f） （39g） （39h） Step 7-7: Solve the convex problem using MATLAB's CVX toolbox. Get the (l+1)th iteration ; Steps 7-8 complete the UAV trajectory calculation iteration for the inner loop in this two-layer iterative optimization algorithm, and set l = l + 1; Steps 7-9 are repeated, from steps 7-3 to 7-8, until the preset conditions are met. The position of the drone calculated in this case was obtained. and assign it to ; Steps 7-10 complete the outer loop of this two-level iterative optimization algorithm. The calculation iteration is performed, with k = k + 1; Steps 7-11 are repeated, from steps 5-1 to 7-10, until the preset conditions are met. Solve for the calculated result and assign it to .

7. The method according to claim 6, characterized in that, Step 8 includes: Step 8-1: Improve the system computational efficiency in this iterative algorithm. For the update, set i = i + 1; Step 8-2: Repeat steps 3-1 to 8-1 until the preset conditions are met, and obtain the approximate optimal solution to the optimization problem in steps 1-6. .

8. The method according to claim 7, characterized in that, In step 8-1, the system computational efficiency in this iterative algorithm is calculated using the following formula. Update: （40）。 9. The method according to claim 8, characterized in that, In step 8-2, the preset condition is: .

Images