Unmanned aerial vehicle calculation unloading method based on conditional value-at-risk and whale optimization

Abstract

The invention discloses an unmanned aerial vehicle calculation unloading method based on conditional value at risk and whale optimization. The method comprises the following steps: constructing an air-ground link channel state information estimation error uncertainty set, and constructing a chance constraint model; unmanned aerial vehicle position deployment, unloading decision and resource allocation are jointly optimized, and an air edge calculation unloading model is established; determining an association scheme of the user and the unmanned aerial vehicle; constructing a distributed robust opportunity constraint of task time delay under the worst condition, performing conservative estimation on the distributed robust opportunity constraint based on a conditional value-at-risk theory, and converting the distributed robust opportunity constraint into a second-order cone constraint problem; decoupling the second-order cone constraint problem into a computing resource allocation problem and an unloading decision problem; and solving to obtain a resource allocation and unloading decision scheme for minimizing the energy consumption of the unmanned aerial vehicle under the condition of satisfying multi-dimensional constraints in the hierarchical air edge computing network. According to the method, the problem of network performance optimization under the condition that channel state information estimation errors exist in an air-ground communication link is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of drone edge offloading, and in particular to a drone computing offloading method based on conditional value at risk and whale optimization. Background Art

[0002] By deploying edge servers on the user terminal side, mobile edge computing technology can significantly reduce task response delays and ensure the high service quality requirements of 6G network computing-intensive tasks. At the same time, drones have the advantages of high mobility and on-demand deployment. By adjusting the location and number of drones, the capacity and coverage of the communication network can be enhanced. Therefore, drones are regarded as ideal carriers for edge computing servers. Combining the advantages of mobile edge computing technology and drones, drones equipped with edge servers can be used as air base stations to provide temporary communication and computing services for remote areas lacking infrastructure and areas that are difficult to cover by communication networks, thereby helping to achieve global coverage of wireless communication services and build a ubiquitous interconnected network that integrates the sky and the earth.

[0003] However, due to the limitations of drone size, weight and battery capacity, its payload and task processing capabilities are limited. In order to make up for the shortcomings of drones in terms of endurance and computing power, a high-altitude platform can be introduced to transfer part of the computing tasks from the drone to the high-altitude platform for processing, thereby reducing the burden on the drone and improving the overall performance of the aerial mobile edge computing system. However, due to the high dynamics of the air-to-ground communication link, the channel state information is imperfect, and there is an unpredictable estimation error between the channel gain and the ideal state, which affects the service quality of user tasks and the timely update and processing of data. In addition, the location deployment, resource allocation and offloading decisions of drones play a vital role in the overall utility of the system and the quality of user service. Therefore, in the aerial edge computing network, it is necessary to comprehensively consider the distribution of ground users and communication requirements, optimize the deployment and joint offloading strategies of drones and high-altitude platforms, so as to meet the user's service quality requirements and the multi-dimensional resource constraints of the system, reduce the overall energy consumption of the system, and extend the service time. Summary of the invention

[0004] Aiming at the problem of network performance optimization in the presence of channel state information estimation errors in air-to-ground communication links, the present invention proposes a UAV computing offloading method based on conditional value at risk and whale optimization.

[0005] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0006] A method for unloading calculations from a drone based on conditional value at risk and whale optimization, the method comprising the following steps:

[0007] Step 1: Construct a hierarchical air edge computing network framework to describe computing resource allocation and offloading scenarios;

[0008] Step 2: Based on the hierarchical air edge computing network framework constructed in step 1, combined with historical data statistical information, the uncertainty set of air-ground link channel state information estimation error is constructed, and the opportunity constraint model is further constructed for the user task delay constraint;

[0009] Step 3: Based on the hierarchical aerial edge computing network framework constructed in step 1 and the opportunity constraint model obtained in step 2, the multi-dimensional resource constraints of the drone swarm and the high-altitude platform are considered, and the drone location deployment, offloading decision and resource allocation are jointly optimized to establish an aerial edge computing offloading model;

[0010] Step 4: Based on the aerial edge computing offloading model proposed in step 3, the K-means algorithm is used to cluster ground users according to the user distribution location and the preset number of drones, and drones are deployed at the centroid of the user cluster to obtain the drone deployment location and determine the association scheme between users and drones;

[0011] Step 5: Based on the uncertainty set of the channel state information estimation error of the air-ground link obtained in step 2, a distributed robust chance constraint of the mission delay under the worst condition is constructed, and a conservative estimate of the distributed robust chance constraint is made based on the conditional value at risk theory, which is converted into a second-order cone constraint problem;

[0012] Step 6: Decouple the second-order cone constraint problem into a computing resource allocation problem and an offloading decision problem;

[0013] Step 7: Solve the computing resource allocation problem using a standard solver to obtain the CPU frequency allocation scheme for the UAV and the server on the high-altitude platform;

[0014] Step 8: Based on the allocation scheme obtained in step 7, the penalty term method is used to transform the unloading decision problem into an unconstrained model, and the binary whale optimization is used to search for a feasible solution to the problem;

[0015] Step 9: Repeat steps 7 and 8 to obtain the resource allocation and unloading decision-making scheme that minimizes the energy consumption of UAVs under multi-dimensional constraints in the hierarchical aerial edge computing network.

[0016] Step 1 further comprises:

[0017] Construct a hierarchical aerial edge computing network scenario, which includes M users, N drones, and 1 high-altitude platform. The drone collects computing tasks generated by ground users and unloads them on the drone or forwards them to the high-altitude platform for further processing based on the task requirements and the resource constraints of the drone.

[0018] Get the position w of the mth user based on the Cartesian three-dimensional coordinate system m =(x m ,y m ), the horizontal position v of the nth UAV n =(x n ,y n ), the flight altitude of all drones is set to a constant value z n , the position of the high-altitude platform is recorded as The user task is recorded as Where L m Indicates the amount of data for the task, c m Indicates the number of CPU cycles required per bit of data, Indicates the maximum delay that the task can tolerate; Indicates whether user m and drone n are associated. If It means that user m is associated with drone n. The user offloads the computing task to the drone, and vice versa. Indicates whether the task of user m is forwarded to the high-altitude platform via drone n. If yes, on the contrary Set f = {f m , m = 1, 2, ..., M} represents the CPU frequency assigned to the user task; v = {v n , n=1, 2, ..., N} represent the horizontal plane position coordinates of the UAV.

[0019] Step 2 further includes:

[0020] Step 2-1: Assume that the estimated error between the actual channel state information and the ideal channel state information under environmental disturbance is Δ m , estimated error Δ m Subordinate to the mean variance The unknown distribution The uncertain set is as follows:

[0021]

[0022] Step 2-2: Orthogonal frequency division multiple access technology is used to transmit the user task from the ground to the UAV associated with it. The transmission delay is:

[0023]

[0024] in is the user uplink transmission channel gain, is the channel gain under ideal conditions, is the channel gain when the distance is 1m, d m，n is the distance between the user and the drone, B uis the channel transmission bandwidth, p u is the transmission power of the user equipment, n0 is the power spectrum density of the channel additive white Gaussian noise; when the user task is offloaded locally on the drone, the calculation delay is expressed as:

[0025]

[0026] The transmission delay of the user task from the UAV to the high-altitude platform is:

[0027]

[0028] where p h is the UAV transmission power, is the UAV antenna transmission gain, B h is the transmission bandwidth of the UAV channel, is the free space path loss, L l is the total route loss, d n，h is the distance between the UAV and the high-altitude platform, v c is the speed of light, f c is the center frequency, k B is the Boltzmann constant, T0 is the system noise temperature; the computational unloading delay of the task on the high-altitude platform is:

[0029]

[0030] The delay constraints of the build task meet the following requirements:

[0031]

[0032] in is the total task delay; α m is the confidence factor, indicating that The total delay of the user task is minimized with probability α m Not greater than the maximum tolerable delay of the task

[0033] Step 3 further includes:

[0034] Step 3-1: Calculate the energy consumption when the user task is processed by the drone:

[0035]

[0036] Among them, ε n is the effective switching capacitance constant of the MEC server carried on the drone; the computing energy consumption when the user task is processed on the high-altitude platform is:

[0037]

[0038] Among them, εh is the energy consumption coefficient of the edge computing server carried on the high-altitude platform; the transmission energy consumption of the task forwarded from the drone to the high-altitude platform is:

[0039]

[0040] The total energy consumption of a single drone is:

[0041]

[0042] The total energy consumption of the aerial platform is:

[0043]

[0044] Step 3-2: Convert the drone location deployment, offloading decision and resource allocation problems in the hierarchical aerial edge computing network into a problem of minimizing the objective function, and model it as the following problem of minimizing the system energy consumption P0:

[0045]

[0046] in and are the maximum energies of the UAV and the high-altitude platform, respectively, and H is the maximum number of tasks that the high-altitude platform can handle simultaneously. and are the maximum CPU speeds of the UAV and the high-altitude platform, respectively, [X min , X max ] and [Y min , Y max ] are the horizontal and vertical boundaries of the region in the two-dimensional plane.

[0047] Step 4 further includes:

[0048] Step 4-1: Randomly select N points in the area as the starting coordinates v of the drone;

[0049] Step 4-2: Calculate the distance between the drone and the user according to the following formula:

[0050]

[0051] And reassign users to their closest clusters;

[0052] Step 4-3: Update the position of the drone according to the following formula:

[0053]

[0054] in Representation Cluster The number of users contained in;

[0055] Step 4-4: Repeat steps 4-1 to 4-3 until the results converge and the deployment position v of the UAV is obtained;

[0056] Step 4-5: Cluster The users in the are associated with the corresponding drones, and the user-drone association indicator variable δ is updated;

[0057] Step 4-6: Convert the problem P0 of minimizing system energy consumption into problem P1:

[0058]

[0059] Step 5 further includes:

[0060] Step 5-1: For the delay opportunity constraint model proposed in step 2, construct the worst-case distributed robust opportunity constraint as follows:

[0061]

[0062] in, is a random probability distribution The lower bound of the distribution of is the total delay;

[0063] Step 5-2: For the opportunity constraint model proposed in step 2, based on the first-order Taylor series expansion formula, construct the worst-case distributed robust opportunity constraint as follows:

[0064]

[0065] in

[0066]

[0067] In the formula, is a random probability distribution The lower bound of the distribution of

[0068] Step 5-3: Based on the conditional value at risk, the distributed robust chance constraint described in step 5-2 is transformed into a mixed integer second-order cone constraint:

[0069]

[0070] where β m 、e m ,q m 、z m and m is an auxiliary variable; problem P1 is transformed into a mixed second-order cone constraint problem P2:

[0071]

[0072] Among them, β, e, q, z and s are vector sets of corresponding auxiliary variables.

[0073] Step 6 further comprises:

[0074] The mixed second-order cone constraint problem P2 is decomposed into a continuous variable problem P3 and a binary variable problem P4:

[0075]

[0076] Step 8 further comprises:

[0077] Step 8-1: According to the constraints in binary variable problem P4, define the following functions respectively:

[0078]

[0079] Step 8-2: According to the objective function in binary variable problem P4, introduce the penalty term The fitness function Γ(λ) is constructed as follows:

[0080]

[0081] Where H(·) is the indicator function, which is defined as follows:

[0082]

[0083] Transform the binary variable problem P4 into the following problem P5:

[0084]

[0085] Step 8-3: For problem P5, the computational offloading decision variable λ is iteratively optimized based on the binary whale optimization algorithm to obtain the optimized binary offloading decision variable λ and fitness function Γ(λ).

[0086] Step 8-3 further includes:

[0087] Initialize the relevant parameters, calculate the fitness of each whale individual according to the fitness function Γ, and obtain the optimal individual corresponding to the current minimum fitness value Iterate, for each individual in the whale population, update the parameters A, C and a, and execute the spiral encirclement mechanism, encirclement contraction mechanism or exploration mechanism according to the probability and the value of the parameter |A|, calculate the distance vector, update the whale position and the fitness function value of each whale individual, and obtain the optimal individual in the population until the maximum number of iterations is reached, and the optimized binary unloading decision variable λ and fitness function Γ(λ) are obtained.

[0088] Compared with the prior art, the present invention has the following beneficial effects:

[0089] First, the drone computing offloading method based on conditional risk value and whale optimization of the present invention proposes a layered aerial edge computing network framework, which combines drones and high-altitude platforms to provide communication and computing services for ground users; constructs an opportunity constraint model based on the delay requirements of user tasks, adopts a clustering algorithm to pre-deploy the location of the drone, conservatively estimates the opportunity model based on the distributed robust optimization algorithm and conditional risk value theory, optimizes the computing resource allocation decision, and designs a binary whale optimization algorithm to optimize the drone offloading decision, realizing energy-saving deployment and decision optimization of the two-layer aerial edge computing system, and effectively solving the problem of network performance optimization when there is channel state information estimation error in the air-to-ground communication link.

[0090] Second, the drone computing offloading method based on conditional risk value and whale optimization of the present invention comprehensively considers the distribution of ground users and communication needs, optimizes the deployment and joint offloading strategy of drones and high-altitude platforms, meets the user's service quality requirements and the system's multi-dimensional resource constraints, reduces the overall energy consumption of the system, and extends the service time. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 This is a scene diagram of the aerial edge computing network model based on multiple drones and high-altitude platforms in the present invention;

[0092] Figure 2 This is a flow chart of the binary whale optimization algorithm proposed in the present invention;

[0093] Figure 3 This is a flow chart of the drone computing offloading method based on conditional risk value and whale optimization proposed in the present invention;

[0094] Figure 4 Comparison of optimization results between the algorithm proposed in the present invention and the traditional algorithm;

[0095] Figure 5 Comparison of system energy consumption between the algorithm proposed in the present invention and the ideal state;

[0096] Figure 6 It is a relationship diagram between the maximum tolerable delay of the task and the total energy consumption of the system in the present invention;

[0097] Figure 7 This is a relationship diagram between user transmission power and system total energy consumption in the present invention. DETAILED DESCRIPTION

[0098] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments of the specification. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0099] Figure 1 The invention provides an air edge computing network model based on multiple drones and high-altitude platforms, specifically, including M users, N drones and 1 high-altitude platform. After deployment, the drone hovers at a fixed height, collects computing tasks generated by ground users, and weighs whether to offload them on the drone or forward them to a high-altitude platform for further processing based on the task requirements and the drone's own resource constraints.

[0100] Figure 2 The binary whale optimization algorithm flow chart provided by the present invention, specifically, firstly initialize relevant parameters, and randomly set the position of each individual in the whale population, and calculate the objective function value of each whale individual in the current iteration; for each whale individual, update parameters A, C and a, and execute the spiral encirclement mechanism, encirclement contraction mechanism or exploration mechanism according to the probability and the value of parameter |A|, update the whale position, and recalculate the fitness function value of each whale individual, and update the optimal individual in the population; repeat the above process until the maximum number of iterations is reached and the algorithm terminates.

[0101] Figure 3 It is an algorithm architecture diagram provided by the present invention. Specifically, the present invention is divided into two stages: drone deployment and computational offloading. First, the drone position and drone-user association decision in problem P0 are solved based on the K-means algorithm; then, based on distributed robust optimization and conditional risk value, the opportunity constraints in problem P1 are conservatively estimated, and the mixed integer second-order cone optimization problem P2 is further decomposed into P3 and P4; finally, problem P4 is converted into an unconstrained problem P5 through the penalty term method, and the integer offloading strategy of P5 is obtained based on the binary whale optimization algorithm.

[0102] Figure 4 It is a relationship diagram between system energy consumption and scenario scale obtained based on different offloading algorithms provided by the present invention, which records the system energy consumption optimization results of different algorithms under different network scales. Specifically, the present invention uses the binary whale optimization algorithm to solve the binary offloading decision of the task, and compares and analyzes it with three other algorithms, including the brute force search algorithm, the greedy algorithm, and the simulated annealing algorithm. It can be seen that the result of the binary whale optimization algorithm is close to the optimal solution obtained by the brute force search algorithm, and is better than the results obtained by the greedy algorithm and the simulated annealing algorithm.

[0103] Figure 5This is a comparison of the system energy consumption obtained based on distributed robust optimization and conditional risk value provided by the present invention and the system energy consumption under ideal conditions. It can be seen that when the channel state estimation error exists, the total energy consumption of the system is higher than that in the ideal case. This phenomenon is due to the fact that in order to cope with environmental interference and channel random errors, the edge computing server needs to allocate more computing resources when processing tasks to meet the task delay requirements. This figure verifies the robustness of the proposed algorithm based on distributed robust optimization and conditional risk value.

[0104] Figure 6 It is a relationship diagram between the maximum tolerable delay of the task and the total energy consumption of the system provided by the present invention. The user task amount increases from 50Mbit, 55Mbit, 60Mbit, 65Mbit, and 70Mbit in sequence, and the total energy consumption of the system under the maximum tolerable delay of different tasks is recorded respectively. The results show that as the amount of user tasks increases, the energy consumption of the system increases. In addition, the energy consumption of the system decreases significantly with the increase of the maximum tolerable delay of the task. This is because when the maximum tolerable delay of the task increases, the edge server's computing and processing time for the task is extended, thereby reducing the number of CPU frequencies required to be allocated for the task, resulting in a reduction in the computing energy consumption of the drone and the high-altitude platform, and a reduction in the total energy consumption.

[0105] Figure 7 This is a relationship diagram between user transmission power and total system energy consumption provided by the present invention. The user task volume increases from 50Mbit, 55Mbit, 60Mbit, 65Mbit, and 70Mbit in sequence, and the total energy consumption of the system under different user transmission powers is recorded respectively. It can be seen that as the user transmission power increases, the total energy consumption of the air edge computing system decreases. This is because the user's transmission power increases, the user-drone transmission delay decreases, the edge server's processing time for the task is extended, the number of CPU frequencies required for the task is reduced, the computing energy consumption of the drone and the high-altitude platform is reduced, and the total energy consumption of the air edge computing system is reduced.

[0106] The method for position deployment, computation offloading and resource allocation of a hierarchical aerial network drone based on conditional value at risk and whale optimization of the present invention comprises the following steps:

[0107] Step 1: Construct a hierarchical air edge computing network framework to describe computing resource allocation and offloading scenarios;

[0108] Step 2: Based on the scenario constructed in step 1 and historical data statistics, an uncertainty set of channel state information estimation errors of the air-ground link is constructed, and an opportunity constraint model is further constructed for the user task delay constraint;

[0109] Step 3: Based on the scenario constructed in step 1 and the opportunity constraint model obtained in step 2, the multi-dimensional resource constraints of the drone swarm and the high-altitude platform are considered, and the drone location deployment, offloading decision and resource allocation are jointly optimized to establish an aerial edge computing offloading model;

[0110] Step 4: Based on the model proposed in step 3, the K-means algorithm is used to cluster ground users according to the user distribution location and the preset number of drones, and drones are deployed at the centroid of the user cluster to obtain the drone deployment location and determine the association scheme between users and drones;

[0111] Step 5: Based on the uncertainty set of the air-ground link channel state information estimation error obtained in step 2, a distributed robust opportunity constraint of the mission delay under the worst conditions is constructed, and a conservative estimate of the distributed robust opportunity constraint is made based on the conditional value at risk theory, and converted into a second-order cone constraint;

[0112] Step 6: Decouple the problem into computing resource allocation problem and offloading decision problem;

[0113] Step 7: Use the standard solver to solve the CPU frequency allocation scheme for the UAV and the server on the high-altitude platform;

[0114] Step 8: Based on the computing resource allocation scheme obtained in step 7, the penalty term method is used to transform the offloading decision problem model into an unconstrained model, and the binary whale optimization is used to search for a feasible solution to the problem;

[0115] Step 9: Repeat steps 7 and 8 to obtain the resource allocation and unloading decision-making scheme that minimizes the energy consumption of UAVs under multi-dimensional constraints in the aerial hierarchical edge computing network.

[0116] To optimize the above technical solutions, the specific measures taken also include:

[0117] The hierarchical aerial edge computing network scenario constructed in step 1 above includes M users, N drones and 1 high-altitude platform. The drone collects computing tasks generated by ground users and unloads them on the drone or forwards them to the high-altitude platform for further processing based on the task requirements and the resource constraints of the drone. The user's position w is obtained based on the Cartesian three-dimensional coordinate system. m =(x m ,y m ), the horizontal position v of the drone n =(x n ,y n ), the flight altitude of all drones is set to a constant value z n The position of the high-altitude platform is recorded as The user task is recorded as Where L mIndicates the amount of data for the task, c m Indicates the number of CPU cycles required per bit of data, Indicates the maximum delay that a task can tolerate. Indicates whether user m and drone n are associated. If It means that user m is associated with drone n, and the user offloads the computing task to the drone, otherwise they are not associated. Indicates whether the task of user m is forwarded to the high-altitude platform via drone n. If yes, on the contrary The servers carried by drones and high-altitude platforms can use dynamic voltage and frequency scaling technology to dynamically adjust the computing frequency assigned to user tasks. The set f = {f m , m = 1, 2, ..., M} represents the CPU frequency allocated to the user task. v = {v n , n=1, 2, ..., N} represent the horizontal plane position coordinates of the UAV.

[0118] The above step 2 includes:

[0119] Step 2-1: Δ m is the estimated error between the actual channel state information and the ideal channel state information under environmental disturbance. Based on historical data statistics, the random channel state information estimation error Δ is set by observing the channel. m Subordinate to the mean variance The unknown distribution The possible distribution sets are as follows:

[0120]

[0121] in is the uncertain parameter Δ m An uncertain set of .

[0122] Step 2-2: The communication link between the UAV and the ground user is modeled as a large-scale fading model. In the ideal state, the channel gain is expressed as

[0123]

[0124] in is the channel gain when the distance is 1m, d m，n is the distance between the user and the drone. Considering the high dynamics and time-varying nature of the air-to-ground link, the actual channel gain is expressed as:

[0125]

[0126] The user task is transmitted from the ground to the UAV associated with it using orthogonal frequency division multiple access technology, and the transmission delay is:

[0127]

[0128] Among them B u is the channel transmission bandwidth, p u is the transmission power of the user equipment, and b0 is the power spectrum density of the channel additive white Gaussian noise. When the user task is offloaded locally on the drone, the calculation delay is expressed as:

[0129]

[0130] The transmission delay of the user task from the UAV to the high-altitude platform is:

[0131]

[0132] where p h is the UAV transmission power, is the UAV antenna transmission gain, B h is the transmission bandwidth of the UAV channel, is the free space path loss, L l is the total route loss, d n，h is the distance between the UAV and the high-altitude platform, v c is the speed of light, f c is the center frequency, k B is the Boltzmann constant, T0 is the system noise temperature. The computational unloading delay of the task on the high-altitude platform is:

[0133]

[0134] The delay constraints of the build task meet the following requirements:

[0135]

[0136] in is the total task delay, α m is the confidence factor, indicating that The total delay of the user task is minimized with probability α m Not greater than the maximum tolerable delay of the task

[0137] The above step 3 includes:

[0138] Step 3-1: Calculate the energy consumption when the user task is processed by the drone:

[0139]

[0140] Among them, ε nis the effective switching capacitance constant of the MEC server carried on the drone. The computing energy consumption when the user task is processed on the high-altitude platform is

[0141]

[0142] where ε h is the energy consumption coefficient of the edge computing server on the high-altitude platform, which is related to the server chip structure. The transmission energy consumption of the task forwarded from the drone to the high-altitude platform is

[0143]

[0144] Based on the system model constructed in step 1, the total energy consumption of a single UAV is

[0145]

[0146] The total energy consumption of the aerial platform is

[0147]

[0148] Step 3-2: Based on the model in step 1, the problem of drone location deployment, offloading decision and resource allocation in the hierarchical aerial edge computing network is transformed into a problem of minimizing the objective function, which is modeled as the following problem of minimizing system energy consumption P0:

[0149]

[0150] in are the maximum energy of the UAV and the high-altitude platform, respectively, H is the maximum number of tasks that the high-altitude platform can handle simultaneously, and are the maximum CPU speeds of the drone and high-altitude platform, respectively. min , X max ] and [Y min , Y max ] are the horizontal and vertical boundaries of the region in the two-dimensional plane.

[0151] Step 4 above includes:

[0152] Step 4-1: Randomly select N points in the area as the starting coordinates v of the drone.

[0153] Step 4-2: Calculate the distance between the drone and the user according to the following formula:

[0154]

[0155] And reallocate users to their closest clusters.

[0156] Step 4-3: Update the position of the drone according to the following formula:

[0157]

[0158] in Representation Cluster The number of users contained in .

[0159] Step 4-4: Repeat the above process until the result converges and the deployment position v of the UAV is obtained.

[0160] Step 4-5: Cluster The users in are associated with the corresponding drones, and the user-drone association indicator variable δ is updated.

[0161] Step 4-6: According to the above steps, the deployment location v of the drone and the user-drone association indicator variable δ can be obtained. The problem P0 described in step 3 above is transformed into problem P1 as follows:

[0162]

[0163] The above step 5 includes:

[0164] Step 5-1: For the delay opportunity constraint model proposed in step 2, construct the worst-case distributed robust opportunity constraint as follows:

[0165]

[0166] in is a random probability distribution The lower bound of the distribution of .

[0167] Step 5-2: Since the channel state estimation information error is much smaller than its theoretical value, based on the first-order Taylor expansion formula, the approximate expression of the total delay is as follows:

[0168]

[0169] The distributed robust chance constraint described in step 5-1 above can be transformed into:

[0170]

[0171] in

[0172]

[0173]

[0174] Step 5-3: Conditional risk value is defined as the conditional expected value of losses exceeding a certain confidence level under a given probability distribution, which can be used to assess potential risk losses. The conditional risk value constraint can constitute a conservative estimate of the worst-case distributed robust chance constraint as follows:

[0175]

[0176] in is a random probability distribution The upper bound of the distribution of , φ(ξ) is the loss function of the random parameter ξ, represents the conditional risk value of φ(ξ) under the confidence level α. Based on the strong duality theorem, we can get the loss function φ(ξ)=Θξ+θ 0 , the worst-case conditional risk value can be expressed as a second-order cone programming as follows:

[0177]

[0178] e-θ 0 +β+q-Θμ-z>0,

[0179] e≥0,z>0,

[0180]

[0181] Where β, e, q, z and s are auxiliary variables, μ and σ are the mean and standard deviation of the random parameter ξ, respectively. Based on the conditional value at risk, the distributed robust chance constraint described in step 5-2 above can be transformed into a mixed integer second-order cone constraint:

[0182]

[0183] where β m , e m ,q m , z m and m is an auxiliary variable. According to the above steps, the problem P1 in step 4 above can be transformed into a mixed second-order cone constraint problem P2 as follows:

[0184]

[0185] Where β, e, q, z and s are vector sets of corresponding auxiliary variables.

[0186] In the above step 6, for the problem P2 obtained in the above step 5, it is originally decomposed into a continuous variable problem P3 and a binary variable problem P4, as follows:

[0187]

[0188] and

[0189]

[0190] In the above step 7, for the continuous variable problem P3 obtained in the above step 6, a computing resource allocation strategy f can be obtained using a standard solver.

[0191] The above step 8 includes:

[0192] Step 8-1: Based on the constraints in the binary variable problem P4 obtained in step 6 above, define the following functions respectively:

[0193]

[0194] Step 8-2: Based on the objective function of the binary variable problem P4 obtained in step 6 above, introduce the penalty term Solutions that violate the constraints will be given a larger penalty to avoid invalid solutions. The fitness function Γ(λ) is constructed as follows:

[0195]

[0196] in Set to 10 5 , H(·) is the indicator function, defined as follows:

[0197]

[0198] Furthermore, the problem P4 in step 6 above is transformed into the following problem P5:

[0199]

[0200] Step 8-3: For the above problem P5, the binary whale optimization algorithm is used to iteratively optimize the calculation of the offloading decision variable λ to obtain the optimal offloading strategy. Figure 2 As shown, the details are as follows:

[0201] Binary whale optimization is a population-based metaheuristic algorithm, where each whale individual represents a potential binary candidate solution. In the iterative process, by alternating between exploration and contraction, several whale individuals continuously update their individual positions until a satisfactory binary solution is found. In the binary whale optimization described in the present invention, each potential solution X(i) is a binary unloading decision variable λ, where i is the current iteration number, and the optimization goal is to minimize the above fitness function Γ(λ).

[0202] First, set the maximum number of iterations I max , and randomly set the initial position of the binary whale population, calculate the fitness of each whale individual according to the above fitness function Γ, and obtain the optimal individual corresponding to the current minimum fitness value

[0203] Iterate, and in each iteration, the binary whale optimization algorithm continuously adjusts the position of each candidate solution and updates the optimal solution found, so that the entire group gradually develops in a better direction until the number of iterations is reached, and the position of the optimal individual and the corresponding fitness function value are output. Among them, the iterative process includes the encirclement stage and the exploration stage. In the encirclement stage, the whale individual updates the current position based on the contraction encirclement mechanism or the spiral encirclement mechanism; in the exploration stage, the whale individual uses the foraging mechanism to search for new solution space and update the position. For each individual in the whale population, repeat the following iterative process until the maximum number of iterations I is reached max :

[0204] 1) Update parameters:

[0205] The parameter a is linearly reduced from 2 to 0 during the iterations:

[0206]

[0207] Where i is the current iteration number. Update parameters A and C as follows:

[0208]

[0209] and

[0210]

[0211] in and is a random vector in the range [0, 1].

[0212] 2) Generate a random number P rand ∈[0, 1].

[0213] 3) When P rand ≥0.5, the spiral encirclement mechanism is implemented. This mechanism simulates the behavior of humpback whales forming a bubble net around their prey, spiraling up and gradually shrinking to catch prey, so that individual whales continue to approach the current optimal solution, while maintaining a certain degree of exploration to prevent falling into the local optimum. The position update formula of individual whales is as follows:

[0214]

[0215] in is the position of the individual whale in the next iteration, is the position of the individual whale in the current iteration, and C(·) is the complement operation. BWOA ∈[0, 1] is a random value, τ su is the position update step size, which is used to control the probability of each bit flipping and is calculated as follows:

[0216]

[0217] in is the distance vector between the current whale individual and the current optimal solution.

[0218] 4) When P rand When |A|≥1, the exploration mechanism is executed. This mechanism imitates the behavior of whales in nature in searching for prey in the ocean. By randomly selecting a position in the solution space, the current solution is guided to move to a new area, thereby increasing the probability of finding the global optimal solution. The individual whale will randomly move to a position in the solution space, and the distance vector is calculated as follows:

[0219]

[0220] is the position of a random whale individual in the solution space. The position update formula is as follows:

[0221]

[0222] in

[0223]

[0224] 5) When P rand When |A|<0.5 and |A|<1, the shrinking encirclement mechanism is executed. This mechanism simulates the behavior of humpback whales gradually shrinking the encirclement when hunting prey. By continuously reducing the distance between individual whales and the current optimal solution, the whale group gradually approaches the optimal solution, thereby improving the local search ability of the algorithm. The distance vector is calculated as follows:

[0225]

[0226] The position update formula is as follows:

[0227]

[0228] in

[0229]

[0230] 6) Calculate the fitness function value Γ of each whale individual and update the current optimal individual

[0231] After the above iterative process is completed, the optimized binary unloading decision variable λ and fitness function Γ(λ) can be obtained.

[0232] In the above step 9, given the computing resource allocation strategy f obtained in step 7, step 8 is executed to optimize the binary offloading decision variable λ; given the binary offloading decision variable λ obtained in step 8, step 7 is executed to optimize the computing resource allocation strategy f. Repeat this process until the results converge, and obtain the resource allocation and offloading decision scheme that satisfies the multi-dimensional constraints in the aerial hierarchical edge computing network and minimizes the system energy consumption.

[0233] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0234] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for unloading drone computing based on conditional value at risk and whale optimization, characterized in that: The drone computing offloading method comprises the following steps: Step 1: Construct a hierarchical air edge computing network framework to describe computing resource allocation and offloading scenarios; Step 2: Based on the hierarchical air edge computing network framework constructed in step 1, combined with historical data statistical information, the uncertainty set of air-ground link channel state information estimation error is constructed, and the opportunity constraint model is further constructed for the user task delay constraint; Step 3: Based on the hierarchical aerial edge computing network framework constructed in step 1 and the opportunity constraint model obtained in step 2, the multi-dimensional resource constraints of the drone swarm and the high-altitude platform are considered, and the drone location deployment, offloading decision and resource allocation are jointly optimized to establish an aerial edge computing offloading model; Step 4: Based on the aerial edge computing offloading model proposed in step 3, the K-means algorithm is used to cluster ground users according to the user distribution location and the preset number of drones, and drones are deployed at the centroid of the user cluster to obtain the drone deployment location and determine the association scheme between users and drones; Step 5: Based on the uncertainty set of the channel state information estimation error of the air-ground link obtained in step 2, a distributed robust chance constraint of the mission delay under the worst condition is constructed, and a conservative estimate of the distributed robust chance constraint is made based on the conditional value at risk theory, which is converted into a second-order cone constraint problem; Step 6: Decouple the second-order cone constraint problem into a computing resource allocation problem and an offloading decision problem; Step 7: Solve the computing resource allocation problem using a standard solver to obtain the CPU frequency allocation scheme for the UAV and the server on the high-altitude platform; Step 8: Based on the allocation scheme obtained in step 7, the penalty term method is used to transform the unloading decision problem into an unconstrained model, and the binary whale optimization is used to search for a feasible solution to the problem; Step 9: Repeat steps 7 and 8 to obtain the resource allocation and unloading decision-making scheme that minimizes the energy consumption of UAVs under multi-dimensional constraints in the hierarchical aerial edge computing network.

2. The method for unloading drone computing based on conditional value at risk and whale optimization according to claim 1, characterized in that: Step 1 further comprises: Construct a hierarchical aerial edge computing network scenario, which includes M users, N drones, and 1 high-altitude platform. The drone collects computing tasks generated by ground users and unloads them on the drone or forwards them to the high-altitude platform for further processing based on the task requirements and the resource constraints of the drone. Get the position w of the mth user based on the Cartesian three-dimensional coordinate system m =(x m ,y m ), the horizontal position v of the nth UAV n =(x n ,y n ), the flight altitude of all drones is set to a constant value z n , the position of the high-altitude platform is recorded as The user task is recorded as Where L m Indicates the amount of data for the task, c m Indicates the number of CPU cycles required per bit of data, Indicates the maximum delay that the task can tolerate; Indicates whether user m and drone n are associated. If It means that user m is associated with drone n. The user offloads the computing task to the drone, and vice versa. Indicates whether the task of user m is forwarded to the high-altitude platform via drone n. If yes, on the contrary Set f = {f m , m = 1, 2, ..., M} represents the CPU frequency assigned to the user task; v = {v n , n=1, 2, ..., N} represent the horizontal plane position coordinates of the UAV.

3. The method for unloading drone computing based on conditional value at risk and whale optimization according to claim 2, characterized in that: Step 2 further includes: Step 2-1: Assume that the estimated error between the actual channel state information and the ideal channel state information under environmental disturbance is Δ m , estimated error Δ m Subordinate to the mean variance The unknown distribution The uncertain set is as follows: Step 2-2: Orthogonal frequency division multiple access technology is used to transmit the user task from the ground to the UAV associated with it. The transmission delay is: in is the user uplink transmission channel gain, is the channel gain under ideal conditions, is the channel gain when the distance is 1m, d m，n is the distance between the user and the drone, B u is the channel transmission bandwidth, p u is the transmission power of the user equipment, n0 is the power spectrum density of the channel additive white Gaussian noise; when the user task is offloaded locally on the drone, the calculation delay is expressed as: The transmission delay of the user task from the UAV to the high-altitude platform is: where p h is the UAV transmission power, is the UAV antenna transmission gain, B h is the transmission bandwidth of the UAV channel, is the free space path loss, L l is the total route loss, d n，h is the distance between the UAV and the high-altitude platform, v c is the speed of light, f c is the center frequency, k B is the Boltzmann constant, T0 is the system noise temperature; the computational unloading delay of the task on the high-altitude platform is: The delay constraints of the build task meet the following requirements: in is the total task delay; α m is the confidence factor, indicating that The total delay of the user task is minimized with probability α m Not greater than the maximum tolerable delay of the task 4. The method for unloading drone computing based on conditional value at risk and whale optimization according to claim 3 is characterized in that: Step 3 further includes: Step 3-1: Calculate the energy consumption when the user task is processed by the drone: Among them, ε n is the effective switching capacitance constant of the MEC server carried on the drone; the computing energy consumption when the user task is processed on the high-altitude platform is: Among them, ε h is the energy consumption coefficient of the edge computing server carried on the high-altitude platform; the transmission energy consumption of the task forwarded from the drone to the high-altitude platform is: The total energy consumption of a single drone is: The total energy consumption of the aerial platform is: Step 3-2: Convert the drone location deployment, offloading decision and resource allocation problems in the hierarchical aerial edge computing network into a problem of minimizing the objective function, and model it as the following problem of minimizing the system energy consumption P0: in and are the maximum energies of the UAV and the high-altitude platform, respectively, H is the maximum number of tasks that the high-altitude platform can handle simultaneously, and are the maximum CPU speeds of the UAV and the high-altitude platform, [X min , X max ] and [Y min , Y max ] are the horizontal and vertical boundaries of the region in the two-dimensional plane.

5. The method for unloading drone computing based on conditional value at risk and whale optimization according to claim 1, characterized in that: Step 4 further includes: Step 4-1: Randomly select N points in the area as the starting coordinates v of the drone; Step 4-2: Calculate the distance between the drone and the user according to the following formula: And reassign users to their closest clusters; Step 4-3: Update the position of the drone according to the following formula: in Representation Cluster The number of users contained in ; Step 4-4: Repeat steps 4-1 to 4-3 until the results converge and the deployment position v of the UAV is obtained; Step 4-5: Cluster The users in the are associated with the corresponding drones, and the user-drone association indicator variable δ is updated; Step 4-6: Convert the problem P0 of minimizing system energy consumption into problem P1:

6. The method for unloading drone computing based on conditional value at risk and whale optimization according to claim 1, characterized in that: Step 5 further includes: Step 5-1: For the delay opportunity constraint model proposed in step 2, construct the worst-case distributed robust opportunity constraint as follows: in, is a random probability distribution The lower bound of the distribution of is the total delay; Step 5-2: For the opportunity constraint model proposed in step 2, based on the first-order Taylor series expansion formula, construct the worst-case distributed robust opportunity constraint as follows: in In the formula, is a random probability distribution The lower bound of the distribution of Step 5-3: Based on the conditional value at risk, the distributed robust chance constraint described in step 5-2 is transformed into a mixed integer second-order cone constraint: where β m 、e m ,q m 、z m and m is an auxiliary variable; problem P1 is transformed into a mixed second-order cone constraint problem P2: Among them, β, e, q, z and s are vector sets of corresponding auxiliary variables.

7. The method for unloading drone computing based on conditional value at risk and whale optimization according to claim 6, characterized in that: Step 6 further comprises: The mixed second-order cone constraint problem P2 is decomposed into a continuous variable problem P3 and a binary variable problem P4:

8. The method for unloading drone computing based on conditional value at risk and whale optimization according to claim 1, characterized in that: Step 8 further comprises: Step 8-1: According to the constraints in binary variable problem P4, define the following functions respectively: Step 8-2: According to the objective function in binary variable problem P4, introduce the penalty term The fitness function Γ(λ) is constructed as follows: Where H(·) is the indicator function, which is defined as follows: Transform the binary variable problem P4 into the following problem P5: Step 8-3: For problem P5, the computational offloading decision variable λ is iteratively optimized based on the binary whale optimization algorithm to obtain the optimized binary offloading decision variable λ and fitness function Γ(λ).

9. The method for unloading drone computing based on conditional value at risk and whale optimization according to claim 1, characterized in that: Step 8-3 further includes: Initialize the relevant parameters, calculate the fitness of each whale individual according to the fitness function Γ, and obtain the optimal individual corresponding to the current minimum fitness value Iterate, for each individual in the whale population, update the parameters A, C and a, and execute the spiral encirclement mechanism, encirclement contraction mechanism or exploration mechanism according to the probability and the value of the parameter |A|, calculate the distance vector, update the whale position and the fitness function value of each whale individual, and obtain the optimal individual in the population until the maximum number of iterations is reached, and the optimized binary unloading decision variable λ and fitness function Γ(λ) are obtained.

Images