Low-altitude Internet of Things communication computing cache and trajectory collaborative optimization method

Abstract

The invention discloses a low-altitude Internet of Things communication computing cache and trajectory collaborative optimization method, which is suitable for an edge computing scene in which a high-altitude platform and multiple unmanned aerial vehicles are collaborative. The method comprises the following steps: firstly, constructing a system model containing task calculation complexity uncertainty, and establishing an optimization problem by taking maximization of successful unloading data volume as a target; the solution process is decomposed into a preprocessing stage and a real-time execution stage; in the preprocessing stage, an unmanned aerial vehicle track and a program caching strategy are optimized; and in the real-time execution stage, aiming at the calculation complexity uncertainty, the time delay opportunity constraint is converted into processable mixed integer second-order cone constraint by utilizing the distributed robust optimization and conditional value-at-risk theory, and the optimal real-time unloading strategy and unloading proportion are solved by defining an unloading mode utility function and adopting a tabu search algorithm. According to the method, resource collaboration of communication, calculation, caching and tracks is realized, and the throughput and the operation robustness of the system can be remarkably improved under the condition that the task complexity is uncertain.

Description

Technical Field

[0001] This invention relates to the field of edge offloading technology for low-altitude intelligent networks, specifically to a method for collaborative optimization of communication computing caching and trajectory in low-altitude intelligent networks. Background Technology

[0002] Low-altitude intelligent networks, as a new network form integrating communication, sensing, and computing, are gradually becoming a key infrastructure for unlocking the "low-altitude economy." Among them, unmanned aerial vehicles (UAVs), with their high maneuverability, flexible on-demand deployment capabilities, and superior line-of-sight communication conditions, can quickly reach mission areas, providing precise communication coverage and edge computing services to ground users. High-altitude platforms, based in the stratosphere, possess significant advantages such as wide coverage, long endurance, and strong carrying capacity, serving as stable aerial backbone nodes and powerful computing centers. By combining mobile edge computing technology with the low-altitude intelligent network architecture, cloud computing capabilities can be seamlessly extended to the user's near end, providing highly reliable, low-latency offloading services for computing tasks in scenarios such as smart logistics, emergency rescue, and urban management, effectively overcoming bottlenecks in user terminals regarding computing power, storage space, and battery life.

[0003] However, task offloading in the low-altitude intelligent network environment faces numerous challenges. First, in practical applications, the computational complexity of tasks is often difficult to predict precisely due to the randomness of input data. This inherent uncertainty can lead to reliability issues such as task timeouts and service interruptions when implementing offloading strategies based on deterministic models. Second, UAVs have limited cache capacity and computing resources, making it impossible to store all types of task processing programs or handle a large number of concurrent tasks. Therefore, designing efficient program caching strategies and task scheduling schemes based on user task requirements and UAV resource availability is crucial for improving system performance. Furthermore, the low-altitude intelligent network environment places higher demands on real-time task processing. However, the multi-dimensional resources such as communication, computing, caching, and trajectory are interconnected, with complex constraints, making it difficult to achieve efficient resource coordination and overall performance optimization while meeting strict latency constraints. Therefore, robust offloading algorithms need to be designed to effectively address the uncertainty of task computational complexity distribution and to achieve joint optimization of communication, computing, caching, and trajectory while ensuring system robustness, thereby improving the overall performance and reliability of low-altitude intelligent network edge computing services. Summary of the Invention

[0004] To address the uncertainty of computational complexity in heterogeneous tasks, and to solve problems such as secure communication, computation offloading strategy design, trajectory optimization, and handling of computational complexity uncertainty in multi-UAV assisted edge computing networks, this invention proposes a distributed robust optimization-based multi-UAV secure communication offloading method that can provide efficient computation offloading services for ground users based on task requirements.

[0005] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows:

[0006] This invention discloses a method for collaborative optimization of communication computing cache and trajectory in low-altitude intelligent networks, comprising the following steps:

[0007] S1 constructs a low-altitude intelligent network architecture that coordinates high-altitude platforms and multiple drones to provide task offloading services for mobile users;

[0008] S2, based on the low-altitude intelligent network framework, establishes a communication model, a computation model, a caching model and a UAV trajectory model that include uncertainty in task computational complexity, constructs an optimization problem with the goal of maximizing the amount of data successfully unloaded, and introduces task latency opportunity constraints.

[0009] S3 breaks down the entire process into a preprocessing stage and a real-time execution stage. The preprocessing stage optimizes the drone trajectory and program caching strategy, while the real-time execution stage optimizes the task unloading strategy and ratio.

[0010] S4. In the preprocessing stage, the drone trajectory of all time slots and the static correlation between mobile users and drones are optimized based on the K-means++ algorithm, guided by the user's location.

[0011] S5. In the preprocessing stage, a dynamic optimization algorithm is designed based on the clustering results to determine the set of pre-buffered task processing programs for the UAV.

[0012] S6. Based on the joint optimization model of low-altitude intelligent network communication computing cache, in view of the uncertainty of task computing complexity, a set of moments uncertainty is constructed. Based on the distributed robust optimization method, the task delay opportunity constraint is transformed into a distributed robust opportunity constraint. Then, using the conditional risk value theory, the distributed robust opportunity constraint is conservatively approximated and transformed into a mixed integer second-order cone constraint.

[0013] S7 introduces a utility function representing different unloading modes, reconstructs the original mixed integer nonlinear programming problem into an integer programming problem with the goal of maximizing the total utility of the system, and designs a tabu search algorithm to solve the reconstructed integer programming problem, thereby obtaining the optimal real-time unloading strategy and unloading ratio.

[0014] S8 is designed with a global algorithm that integrates the optimization results of the preprocessing stage and the real-time execution stage, and outputs the final UAV trajectory, program caching strategy, task unloading strategy and unloading ratio scheme to achieve robust and efficient operation of low-altitude intelligent networks under uncertain task complexity.

[0015] To optimize the technical solution, further improvements to this invention include:

[0016] Step S1 specifically includes:

[0017] Construct a hierarchical low-altitude intelligent network system, which includes a high-altitude platform, A drone equipped with a mobile edge computing server and For mobile users, using a discrete time slot model, the total system runtime is divided into: There are 3 equal-length time slots, each with a duration of 1. Each user generates a computing task per second per time slot. The task can be partially offloaded to its associated drone, or relayed via drone to a high-altitude platform for processing. For users With drones binary associative variables, when the user With drones When establishing a connection, ,otherwise ,

[0018] The coordinates of the aerial platform are fixed as follows: drones In the time slot The horizontal coordinate is The flight altitude is fixed at Mobile users In the time slot The horizontal coordinate is Mobile users With drones In the time slot The distance between them is drones With the high-altitude platform in the time slot The distance is .

[0019] Step S2 specifically includes:

[0020] To model the task, we assume that there exists Types of tasks, users In the time slot The generated tasks correspond one-to-one with the task type. Within each time slot, the matching relationship between the tasks generated by each user and the task type can be represented as follows: ,in Indicates user In the time slot The types of tasks generated. Each task type For a specific computational program, a task of that type can only be unloaded if and only if the server has pre-cached the program and database required for the task computation. The cache size is Mobile users In the time slot The generated task is represented as ,in For the amount of task data, For task type The corresponding computational complexity, considering actual uncertainties, is categorized as follows: The computational complexity of the task is modeled as ,in This is an estimated value. For those with unknown distributions Random error, assuming Some statistical information can be obtained through long-term observation, and its first moment and second moment It is known that, among which The mean, To account for variance, the system adopts a partial unloading mode, with the unloading ratio... Indicates user In the time slot Will The amount of data is offloaded to its associated drone. The task offloading between the drone and the high-altitude platform adopts a binary offloading strategy.

[0021] Communication between mobile users and drones uses a line-of-sight channel model with a channel gain of [value missing]. ,in For the reference channel gain at a distance of 1 meter, the user To drones The uplink transmission latency is time slot At that time, drones User The transmission latency of the unloaded task to the high-altitude platform is ,

[0022] For binary cache decision variables, if the drone Cache processing For programs of type [type] tasks, then ,otherwise binary decision variables and These represent the calculation location selection, when The time indicates the mission is in the drone Local computation, when The time indicates that the task is forwarded to the high-altitude platform for computation. When the task is processed by the UAV, the computation latency is... High-altitude platform for handling tasks The computation delay is ,

[0023] Task In the time slot Total execution latency Including transmission latency from mobile users to drones Transmission latency from UAV to high-altitude platform UAV computational latency and high-altitude platform calculation delay Total amount of data successfully processed by the system for

[0024] ,

[0025] Considering task latency constraints, cache capacity limitations, computational access limitations, and variable flow constraints, the drone trajectory is jointly optimized. User-Drone Relationship Drone program caching strategy Task unloading decision and and task uninstallation ratio To maximize system efficiency, the following optimization problem is established:

[0026]

[0027] in, Indicates based on uncertainty distribution Opportunity constraints For confidence level, Maximum cache capacity limit for a single drone. The user task matching relationship represents the user In the time slot The generated task belongs to type , Used to represent drones Has the user been pre-cached? In the time slot The corresponding libraries required for the generated tasks. and These represent the maximum number of tasks that drones and high-altitude platforms can handle simultaneously, respectively. It represents the minimum and maximum values ​​of the region boundary.

[0028] Step S3 specifically includes:

[0029] The overall optimization process is broken down into two stages: a preprocessing stage and a real-time execution stage. In the preprocessing stage, the drone trajectory is optimized. Mobile user-drone relationship Program caching strategy During the real-time execution phase, the task offloading strategy is optimized based on the infrastructure configuration determined in the preprocessing phase. and Task uninstallation ratio .

[0030] Step S4 specifically includes:

[0031] In the system initial time slot ,from The location of a user is randomly selected from among the users as the first cluster center. For each user Calculate the minimum distance from the selected cluster center. and calculate user The probability of being selected as the next cluster center Select the user with the highest probability as the new cluster center, and then repeat the process until the desired cluster is obtained. The initial cluster centers, in time slots Calculate each user To each cluster center The distance is used to reassign users to the nearest cluster, let... Indicates belonging to a cluster The user set, The cluster center represents the number of users in the cluster. The location update rules are as follows:

[0032] ;

[0033] Repeat the above allocation and update process until the results converge, and finally obtain... The cluster center location is set as the time slot of the drone. The initial deployment location is determined, each user is assigned to the nearest cluster, and a communication link is established with the corresponding drone. Then, the connection relationship is determined based on the clustering results. If the user... Belongs to cluster center Then set ,otherwise For each time slot ,based on The real-time distribution of users in the data center is used to recalculate the drones according to the above update rules. The trajectory path points, and ultimately, the sequence of UAV positions in each time slot constitutes the UAV trajectory for the entire period. At the same time, it obtained the connection relationship between mobile users and drones. .

[0034] Step S5 specifically includes:

[0035] Define each program In drones Cache efficiency Define a state variable as a weighted sum of the normalized data size and the computational complexity of normalization for each user task in each cluster. Indicates before consideration The program has a usable storage capacity of [number] programs. The task caching decision optimization problem P1, which aims to maximize the efficiency of program caching while satisfying the UAV's cache space constraints, can be viewed as maximizing the efficiency of program caching.

[0036] Initialize and set initial values ​​for all cached decision variables, and initialize the dynamic programming table. , Then, iterate through all drones. Consider each procedure in sequence, according to the procedure. storage requirements Compared with current capacity The relationship is filled into the dynamic programming table. After the dynamic programming table is completed, the backtracking method is used to determine the optimal caching scheme. Starting from the final state, the process is reversed to check the decision records of each program and finally generate the optimal binary caching decision variables for each UAV.

[0037] Step S6 specifically includes:

[0038] Determining the drone trajectory based on the preprocessing stage User association and caching strategies Subsequently, the real-time execution phase continues to optimize the binary variables related to the unloading strategy. , and continuous variables That is, to solve problem P2:

[0039]

[0040] Based on the available moment estimation information and Constructing an uncertain set The set contains All possible probability distributions Based on uncertain sets By employing a distributed robust optimization method, the chance constraints are transformed into distributed robust chance constraints, and the worst-case solution is sought.

[0041] ;

[0042] in express The lower bound of the possible distribution. For confidence level,

[0043] Based on conditional risk value, convex optimization is performed, and P2 is further transformed into a mixed integer second-order cone constraint problem P3:

[0044]

[0045] in

[0046] Auxiliary variable A set of.

[0047] Step S7 specifically includes:

[0048] Each task has three offloading modes per time slot: no offloading, drone-based offloading, and high-altitude platform-based offloading. For the association satisfies drones Task Substitute into and Solve optimization problem P4 to obtain... and As a task The maximum unloadable ratio in both drone unloading and high-altitude platform unloading scenarios.

[0049]

[0050] Introducing ternary decision variables The uninstallation mode is defined as follows:

[0051]

[0052] Further define the utility function to quantify tasks under different unloading modes. In the time slot The amount of data that can be successfully processed:

[0053]

[0054] P3 is equivalently transformed into integer programming problem P5, which aims to maximize the total utility of the system.

[0055]

[0056] Introducing a penalty function mechanism, the fitness function is defined as follows:

[0057] ;

[0058] in The penalty factor is set to a constant, and the solution to the optimization problem is encoded into a variable of length . vector Each element of the vector Corresponding to the task In the time slot The unloading decision corresponds to three modes: no unloading, drone unloading, and high-altitude platform unloading. First, an initial set of candidate solutions that satisfy the cache constraints is generated. Calculate the fitness value for each initial solution. Then, the solution with the highest fitness is selected as the current optimal solution. and the global optimal solution ,

[0059] In each iteration In the process, mutations can be performed through single-point compilation mutations or task-type-based mutations from the current solution. generate Given candidate neighbor solutions, the generated candidate solution set... Calculate each candidate solution fitness value The candidate solutions are sorted in descending order of fitness value. Each candidate solution is examined sequentially. If the move of a candidate solution is not in the tabu list and its fitness is better than the current solution, the candidate solution is accepted. If the move of a candidate solution is tabu, but its fitness is better than the global optimum, the candidate solution is accepted as an exception. If no candidate solution is accepted after a complete traversal, a backoff mechanism is activated, and the first non-tabulated candidate solution is accepted to ensure the continuity of the search process. In each iteration, the tabu list and the current solution are updated. And compare it with the global optimal solution. The fitness of the algorithm is used to update the global optimum. The algorithm reaches its maximum number of iterations. Then terminate and output the decoded optimal unloading strategy. , and the corresponding uninstallation ratio .

[0060] Step S8 specifically includes:

[0061] A two-stage collaborative global optimization framework is constructed, in which the preprocessing stage provides basic resource configuration for the system, and the real-time execution stage performs dynamic optimization and adjustment based on the preprocessing results. The preprocessing stage executes S4 to optimize the UAV trajectory. User-Drone Relationship Then, based on the obtained associations, S5 is executed to determine the program caching strategy. During system operation, based on the preprocessing stage... , and Collect network state information for the current time slot, and execute S6 and S7 to solve the real-time offloading strategy to obtain... , and This enables robust and efficient operation in environments with uncertain task complexity.

[0062] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0063] 1. This invention breaks down the barriers between communication, computing, caching, and trajectory resources by jointly optimizing UAV trajectory, user association, program caching, unloading decisions, and unloading ratios, achieving efficient collaboration of multi-dimensional resources. Simulation results show that, compared with traditional block coordinate descent and genetic algorithms, the proposed method performs best in terms of the core indicator of successful unloading data volume, maximizing system throughput and meeting the real-time processing needs of low-altitude intelligent networks for massive tasks.

[0064] 2. This invention addresses the challenge of uncertain computational complexity in practical tasks by introducing a moment uncertainty set to describe the error distribution and utilizing distributed robust optimization and conditional value-at-risk theory to make conservative approximations of opportunity constraints. This method only requires known information on the first and second moments of the uncertain parameters, without needing their precise distribution, to ensure that the system can still meet task latency requirements with a high confidence level even in the worst-case scenario, significantly enhancing the system's service reliability and robustness in dynamic and unpredictable environments.

[0065] 3. The hierarchical collaborative architecture of "high-altitude platform + multiple UAVs" constructed in this invention effectively integrates the dual advantages of the high-altitude platform's wide coverage and strong computing power with the UAVs' flexible deployment and low latency. Simulation verification shows that the successful offloading of data by this collaborative architecture is significantly higher than that of a single architecture using only UAVs or only a high-altitude platform, achieving complementary advantages in resources and capabilities and alleviating the resource bottleneck of a single node.

[0066] 4. This invention rationally breaks down complex mixed-integer nonlinear programming problems through a two-stage decomposition of "preprocessing + real-time execution." The preprocessing stage optimizes long-term configurations offline based on historical information; the real-time execution stage designs a tabu search algorithm based on utility function reconstruction to quickly solve online unloading strategies. This design ensures high solution quality while significantly reducing computation time compared to traditional alternating optimization algorithms, achieving a good balance between optimization accuracy and real-time decision-making efficiency, and meeting the timeliness requirements of low-altitude intelligent networks.

[0067] 5. This invention employs a dynamic programming-based program caching strategy. Under the constraint of limited UAV storage capacity, it intelligently selects the set of programs with the highest caching efficiency based on historical statistical characteristics of user task requirements, thereby effectively improving the hit rate and utilization efficiency of cached resources. The UAV trajectory is adaptively adjusted based on user mobility, ensuring the continuous effectiveness of network coverage and improving the stability and quality of communication links.

[0068] In summary, this invention not only provides a robust optimization method for dealing with the uncertainty of task complexity in theory, but also achieves a comprehensive improvement in the overall performance, reliability, real-time performance and resource efficiency of the low-altitude intelligent network system in practice through systematic architecture and algorithm design, providing reliable technical support for low-altitude economic applications. Attached Figure Description

[0069] Figure 1 This is a scenario diagram of the edge computing offloading network model based on the low-altitude intelligent network in this invention;

[0070] Figure 2 This is a global flowchart of the two-stage algorithm proposed in this invention;

[0071] Figure 3 This is a performance comparison chart of different optimization algorithms in this invention on the metric of successfully unloading data volume.

[0072] Figure 4 This is a comparison chart of the computation time of different optimization algorithms in this invention;

[0073] Figure 5 This is a comparison chart of the amount of data successfully unloaded by different network architectures in this invention;

[0074] Figure 6 This is a comparison chart of the amount of successfully unloaded data under different drone cache capacities in this invention;

[0075] Figure 7 This is a comparison chart showing the amount of successfully unloaded data under different levels of uncertainty in the computational complexity of the tasks in this invention. Detailed Implementation

[0076] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0077] This invention discloses a method for collaborative optimization of communication computing cache and trajectory in low-altitude intelligent networks, the method comprising the following steps:

[0078] S1 constructs a low-altitude intelligent network architecture that coordinates high-altitude platforms and multiple drones to provide task offloading services for mobile users;

[0079] S2, based on the low-altitude intelligent network framework, establishes a communication model, a computation model, a caching model and a UAV trajectory model that include uncertainty in task computational complexity, constructs an optimization problem with the goal of maximizing the amount of data successfully unloaded, and introduces task latency opportunity constraints.

[0080] S3 breaks down the entire process into a preprocessing stage and a real-time execution stage. The preprocessing stage optimizes the drone trajectory and program caching strategy, while the real-time execution stage optimizes the task unloading strategy and ratio.

[0081] S4. In the preprocessing stage, the drone trajectory of all time slots and the static correlation between mobile users and drones are optimized based on the K-means++ algorithm, guided by the user's location.

[0082] S5. In the preprocessing stage, a dynamic optimization algorithm is designed based on the clustering results to determine the set of pre-buffered task processing programs for the UAV.

[0083] S6. Based on the joint optimization model of low-altitude intelligent network communication computing cache, in view of the uncertainty of task computing complexity, a set of moments uncertainty is constructed. Based on the distributed robust optimization method, the task delay opportunity constraint is transformed into a distributed robust opportunity constraint. Then, using the conditional risk value theory, the distributed robust opportunity constraint is conservatively approximated and transformed into a mixed integer second-order cone constraint.

[0084] S7 introduces a utility function representing different unloading modes, reconstructs the original mixed integer nonlinear programming problem into an integer programming problem with the goal of maximizing the total utility of the system, and designs a tabu search algorithm to solve the reconstructed integer programming problem, thereby obtaining the optimal real-time unloading strategy and unloading ratio.

[0085] S8 is designed with a global algorithm that integrates the optimization results of the preprocessing stage and the real-time execution stage, and outputs the final UAV trajectory, program caching strategy, task unloading strategy and unloading ratio scheme to achieve robust and efficient operation of low-altitude intelligent networks under uncertain task complexity.

[0086] The following simulation results further illustrate the effectiveness of this invention:

[0087] Figure 1 This invention provides a scenario diagram of an edge computing offloading network model based on low-altitude intelligent network, which includes one high-altitude platform, several low-altitude drones and multiple ground mobile users. Each node is connected through a line-of-sight channel, and user tasks are either processed locally by the associated drones or forwarded to the high-altitude platform for processing.

[0088] Figure 2This is the global flowchart of the two-stage algorithm provided by this invention. In the preprocessing stage, the user's initial position is used as input. The K-means++ algorithm determines the association between the drone deployment and the user, and adjustments are made based on the user's position to obtain the drone's full-time-slot trajectory. Then, a dynamic programming algorithm is used to generate a drone pre-caching strategy. In the real-time execution stage, based on the preprocessing results and the current time-slot state, an uncertain set of task complexity moments is constructed. Distributed robust optimization and conditional value-at-risk theory are used to conservatively estimate the opportunity constraints. After introducing a utility function, a tabu search algorithm is used to solve for the unloading strategy and proportion. Finally, the results of the two stages are integrated to output a complete solution.

[0089] Figure 3 This is a comparison chart of the successful data unloading volume of different optimization algorithms provided by this invention. The horizontal axis represents the number of mobile users (25, 28, 31, 34, 37, 40), and the vertical axis represents the successful data unloading volume. This invention designs a tabu search algorithm to solve the second-stage problem and compares it with block coordinate descent, differential evolution, and genetic algorithms. The block coordinate descent method decomposes the second-stage problem into a continuous variable optimization problem and a 0-1 variable optimization problem, and iteratively solves the subproblems until convergence by fixing one block variable and solving the other. The continuous optimization problem is solved by CVX, and the 0-1 variable problem is solved by Mosek. The results show that as the number of users increases, the successful data unloading volume of each algorithm increases, and the tabu search algorithm consistently performs best, verifying the superiority of this algorithm in improving data unloading efficiency.

[0090] Figure 4 This graph compares the computation time of different optimization algorithms provided by this invention. The horizontal axis represents the number of mobile users (25, 28, 31, 34, 37, 40), and the vertical axis represents the computation time. As the number of users increases and the network size grows, the computation time of each algorithm increases. The Block Coordinate Descent (BCD) method, in particular, requires alternating calls to the CVX and Mosek solvers and iterating repeatedly between continuous and discrete variables, resulting in a rapid increase in computation time with the problem size. It can be seen that the tabu search algorithm proposed in this invention has a significantly lower computation time than other benchmark algorithms and provides better solution quality, proving that the proposed algorithm achieves an effective balance between solution quality and computational efficiency.

[0091] Figure 5This is a comparison chart of the successful data offloading volume under different network architectures provided by this invention. The horizontal axis represents the number of mobile users (25, 28, 31, 34, 37, 40), and the vertical axis represents the successful data offloading volume. Specifically, this invention proposes a UAV + high-altitude platform collaborative framework to provide users with offloading services, and compares the offloading data volume with that obtained by UAV-only architecture and high-altitude platform-only architecture. The results show that as the number of users increases, the successful data offloading volume of the collaborative architecture is consistently significantly higher than that of the two single architectures. The results demonstrate that the architecture proposed in this invention fully leverages the complementary advantages of the flexible coverage of UAVs and the computing power backup of high-altitude platforms, alleviating the resource bottleneck of single architectures.

[0092] Figure 6 This is a comparison chart of the amount of data successfully unloaded by different drone cache capacities provided by this invention. The horizontal axis represents the number of mobile users (25, 28, 31, 34, 37, 40), comparing the amount of data that can be unloaded by the drone under three different cache capacities: 2Gbits, 3Gbits, and 4Gbits. As the number of users increases, the amount of data under each cache capacity shows an increasing trend, and the larger the cache capacity, the higher the amount of data successfully unloaded.

[0093] Figure 7 This is a comparison chart of the amount of data successfully unloaded under different uncertainties in task computational complexity, as provided by this invention. The horizontal axis represents the number of mobile users (25, 28, 31, 34, 37, 40), and the vertical axis represents the amount of data successfully unloaded. Ideally, computational complexity is unpredictable, and the chance constraint degenerates into a deterministic linear constraint. However, when computational complexity uncertainty exists, the system's unloading performance decreases to some extent compared to the ideal situation. This phenomenon is expected: when considering random factors in real-world systems, to meet service reliability requirements, the system needs to adopt a relatively conservative unloading strategy to cope with complexity fluctuations in the worst-case scenario, leading to a performance trade-off. Simultaneously, as computational complexity uncertainty increases, the maximum amount of tasks that can be unloaded decreases. This indicates that in environments with increased uncertainty, the system needs to adopt a more conservative resource allocation strategy, further verifying the effectiveness and adaptability of the proposed distributed robust optimization method. The results show that the distributed robust optimization method adopted in this invention can still ensure that the system meets task latency requirements with a high probability under uncertain environments, verifying its effectiveness in improving system robustness.

[0094] The technical solution of the present invention is as follows:

[0095] Step 1: Construct a low-altitude intelligent network architecture that coordinates high-altitude platforms and multiple drones to provide task offloading services for mobile users;

[0096] Step 2: Based on the low-altitude intelligent network framework constructed in Step 1, establish a communication model, a computation model, a caching model, and a UAV trajectory model that include uncertainty in task computational complexity. Construct an optimization problem with the goal of maximizing the amount of data successfully unloaded, and introduce task latency opportunity constraints.

[0097] Step 3: Based on the low-altitude intelligent network framework constructed in Step 1 and the problem model proposed in Step 2, the entire process is decomposed into a preprocessing stage and a real-time execution stage. The preprocessing stage optimizes the UAV trajectory and program caching strategy, while the real-time execution stage optimizes the task unloading strategy and ratio.

[0098] Step 4: Based on the model proposed in Step 2, in the preprocessing stage, guided by the user location, the UAV trajectory of all time slots and the static correlation between mobile users and UAVs are optimized using the K-means++ algorithm.

[0099] Step 5: Based on the model proposed in Step 2 and the clustering results obtained in Step 4, design a dynamic optimization algorithm to determine the set of pre-buffered task processing programs for the UAV;

[0100] Step 6: Based on the model proposed in Step 2, the optimized UAV position and association decisions in Step 4, and the UAV caching decisions determined in Step 5, to address the uncertainty of task computational complexity, construct its moment uncertainty set. Based on the distributed robust optimization method, transform the task delay opportunity constraint into a distributed robust opportunity constraint. Then, using the conditional risk value theory, make a conservative approximation of the distributed robust opportunity constraint and transform it into a mixed integer second-order cone constraint.

[0101] Step 7: Based on the mixed integer second-order cone optimization problem transformed in Step 6, introduce a utility function representing different unloading modes, reconstruct the original mixed integer nonlinear programming problem into an integer programming problem with the goal of maximizing the total utility of the system, and design a tabu search algorithm to solve the reconstructed integer programming problem to obtain the optimal real-time unloading strategy and unloading ratio.

[0102] Step 8: Design a global algorithm to integrate the optimization results of the preprocessing stage and the real-time execution stage, and output the final UAV trajectory, program caching strategy, task unloading strategy and unloading ratio scheme to achieve robust and efficient operation of the low-altitude intelligent network under uncertain task complexity.

[0103] To optimize the above technical solution, the specific measures also include:

[0104] Step 1 above includes:

[0105] Step 1-1: Construct a hierarchical low-altitude intelligent network system, which includes a high-altitude platform, A drone equipped with a mobile edge computing server and Mobile users. The set of drones is represented as... The user set is represented as To characterize the dynamic process, a discrete time-slot model is adopted, dividing the total system runtime into segments. There are 3 equal-length time slots, each with a duration of 1. Seconds. In this system architecture, the high-altitude platform hovers in a fixed position, possessing powerful computing capabilities and a complete program library, capable of handling all types of computational tasks. The drones fly in low-altitude airspace, with limited computing power and cache capacity, able to store only a portion of the task processing programs and execute a limited number of tasks. Mobile users move randomly on the ground, each user generating one computational task per time slot. Tasks can be partially offloaded to their associated drones or relayed to the high-altitude platform via drones for processing. To avoid additional switching overhead, each user can only communicate with one drone at a time. For users With drones Binary associative variables. When the user With drones When establishing a connection, ,otherwise .

[0106] Steps 1-2: Describe the positions of all nodes in the network using a three-dimensional Cartesian coordinate system. The coordinates of the aerial platform are fixed at... Drones In the time slot The horizontal coordinate is The flight altitude is fixed at Mobile users In the time slot The horizontal coordinate is Its position update formula is:

[0107]

[0108] in For users In the time slot movement speed, That is its maximum speed. This refers to the user's displacement direction angle. (Moving user) With drones In the time slot The distance between them is calculated as follows drones With the high-altitude platform in the time slot The distance is calculated as .

[0109] Step 2 above includes:

[0110] Step 2-1: Model the task. Assume there exists... There are several task types, and the set of task types is denoted as . .user In the time slot The generated tasks correspond one-to-one with the task type, and the mapping relationship is expressed as follows: Each task type For a specific computational program, a task can be unloaded if and only if the server has pre-cached the program and database required for the task's computation. The cache size is The corresponding task computational complexity is This refers to the number of CPU cycles required to process each bit of data. Therefore, mobile users... In the time slot The generated task can be represented as ,in For the amount of task data, For task type The corresponding computational complexity. Considering practical uncertainties, the type is... The computational complexity of the task is modeled as ,in This is an estimated value. For those with unknown distributions Random error. Assume... Some statistical information can be obtained through long-term observation, and its first moment and second moment It is known that, among which The mean, The variance is given. Furthermore, considering the divisibility of the task, the system adopts a partial offloading mode, introducing continuous variables. As an uninstallation percentage, it represents the user's uninstallation rate. In the time slot Will The amount of data is offloaded to its associated drone. A binary offloading strategy is used between the drone and the high-altitude platform, meaning that the drone must treat the received mission data as a whole and either process it all locally or forward it all to the high-altitude platform for processing.

[0111] Step 2-2: Model the communication model. Communication between the user and the drone, and between the drone and the high-altitude platform, uses Orthogonal Frequency Division Multiple Access (OFDMA) to avoid mutual interference within the communication range. Communication between the mobile user and the drone uses a line-of-sight channel model with a channel gain of [value missing]. ,in The channel gain is used as a reference distance of 1 meter. (User) To drones The uplink transmission rate is

[0112] ,

[0113] in For communication bandwidth, For user transmission power, This refers to the noise power spectral density. The task is assigned by the user. To drones The transmission delay is

[0114] .

[0115] Communication between the UAV and the high-altitude platform also uses a line-of-sight model, with a maximum transmission rate of [missing information].

[0116] ,

[0117] in For link bandwidth, For the drone's transmission power, For antenna gain, For path loss, At the speed of light, For the center frequency, For system line loss, Boltzmann's constant, System noise temperature. Time slot. At that time, drones User The transmission latency for the unloaded task to the high-altitude platform is:

[0118] ,

[0119] in For drones Should users The task is forwarded to the binary decision variables of the high-altitude platform.

[0120] Steps 2-3: Model the computation and caching model. Introduce binary caching decision variables. If drone Cache processing For programs of type [type] tasks, then ,otherwise The task processing method is determined by both the program cache state and system scheduling. For mobile users... The task type is And its associated drones The corresponding program has been cached, that is Then the task can be performed using a drone. Local computation or processing on a high-altitude platform; otherwise, it can only be done via drone. Forwarded to the high-altitude platform for calculation. Introducing binary decision variables. and These represent the calculation location selection, when The time indicates the mission is in the drone Local computation, when The time indicates that the task is forwarded to the high-altitude platform for computation. When the task is processed by the UAV, its computation latency is:

[0121] ,

[0122] in CPU frequency for handling a single task on a drone. Task processing on an aerial platform. The computation delay is

[0123] ,

[0124] CPU frequency for handling a single task on a high-altitude platform.

[0125] Steps 2-4: Construct the optimization problem P0. Based on the preceding analysis, the task... In the time slot Total execution latency Including transmission latency from mobile users to drones Transmission latency from UAV to high-altitude platform UAV computational latency and high-altitude platform calculation delay Its expression is

[0126] .

[0127] The low-altitude intelligent network has strict real-time requirements for tasks. Task data that has not been processed within the current time slot will be considered timed out and discarded. The processing delay of a task cannot exceed the length of the current time slot. Total amount of data successfully processed by the system for

[0128] .

[0129] Considering constraints such as task latency, cache capacity, computational access limitations, and variable flow constraints, the drone trajectory is jointly optimized. User-Drone Relationship Drone program caching strategy Task unloading decision and and task uninstallation ratio To maximize system efficiency, the following optimization problem is established:

[0130]

[0131] in, Indicates based on uncertainty distribution Opportunity constraints For confidence level, Maximum cache capacity limit for a single drone. The user task matching relationship represents the user In the time slot The generated task belongs to type , Used to represent drones Has the user been pre-cached? In the time slot The corresponding libraries required for the generated tasks. and These represent the maximum number of tasks that drones and high-altitude platforms can handle simultaneously, respectively. It represents the minimum and maximum values ​​of the region boundary.

[0132] Step 3 above further includes:

[0133] The overall optimization process is decomposed into two sequential optimization phases: a preprocessing phase and a real-time execution phase. The preprocessing phase, executed during system initialization, optimizes the UAV trajectory and program caching strategy, establishing a relatively stable network infrastructure configuration. The real-time execution phase, executed dynamically in each runtime slot, optimizes task offloading strategies and offloading ratios based on real-time network conditions, achieving dynamic resource adaptation. Specifically, in the preprocessing phase, optimization decisions are made based on the system's long-term statistical characteristics. The decision variables optimized in this phase include: the UAV trajectory. Mobile user-drone relationship Program caching strategy These decisions remain relatively stable throughout the entire runtime, providing the system with basic communication coverage and computing resource distribution. During the real-time execution phase, based on the infrastructure configuration determined in the preprocessing phase, rapid optimization is performed on the real-time state of each time slot. The decision variables optimized in this phase include task offloading strategies. and Task uninstallation ratio These decisions are dynamically adjusted based on real-time information such as the channel state and task characteristics of the current time slot, ensuring the system's robustness in environments with uncertain task computational complexity.

[0134] Step 4 above includes:

[0135] Step 4-1: Perform initial clustering of users based on K-means++ to establish the association between drones and users, and further adjust the drones based on the subsequent locations of mobile users to determine the drones' flight trajectories. Specifically, in the initial time slot of the system... , with the horizontal coordinates of all mobile users As the input dataset, the K-means++ clustering algorithm is executed from... The location of a user is randomly selected from among the users as the first cluster center. For each user Calculate the minimum distance from the selected cluster center: ,user The probability of being selected as the next cluster center is calculated as follows:

[0136] .

[0137] Based on the calculated probability Choose the user with the highest probability. As a new cluster center.

[0138] Step 4-2: Repeat the above process 4-1 until the desired result is obtained. 1. Obtain the initial cluster centers. After obtaining the initial cluster centers, execute the standard K-means algorithm. (In time slots...) Calculate each user To each cluster center The distance is calculated, and users are reassigned to the nearest cluster. Let... Indicates belonging to a cluster The user set, This indicates the number of users in the cluster. Cluster center. The location update rules are as follows:

[0139] ;

[0140] Repeat the above allocation and update process until the results converge, and finally obtain... The cluster center location is set as the time slot of the drone. The initial deployment location. Each user is assigned to the nearest cluster and establishes a communication link with the corresponding drone. Then, the connection relationship is determined based on the clustering results, if the user... Belongs to cluster center Then set ,otherwise .

[0141] Step 4-3: To adapt to the mobility of mobile users while maintaining system stability, an adjustment strategy is adopted to update the drone's location after the initial clustering is completed. For each time slot ,based on The real-time distribution of users in the data center is used to recalculate the drones according to the above update rules. The trajectory path points. Ultimately, the sequence of UAV positions in each time slot constitutes the UAV trajectory for the entire cycle. At the same time, it obtained the connection relationship between mobile users and drones. .

[0142] Step 5 further includes:

[0143] Step 5-1: Based on the user-drone association obtained in Step 4, for each drone Design a program caching strategy. The core objective of this strategy is to select the most suitable set of programs for caching within the limited storage capacity constraints of the drone, in order to maximize system performance.

[0144] First, for each program In drones Define cache benefit value This benefit value is derived from the data volume and computational complexity of the corresponding task using a weighted synthesis procedure.

[0145] ,

[0146] in

[0147]

[0148] The normalized data size of user tasks in each cluster is obtained by statistically analyzing the total amount of historical task data generated by associated users and then performing normalization.

[0149]

[0150] This represents the normalized computational complexity, taking into account an estimate of the task complexity. and uncertainty mean . Weighting coefficients This is used to balance the relative importance of data volume and computational complexity in caching decisions. In clustering The set of task types that appear in the text.

[0151] Step 5-2: Transform the program caching problem into a classic 0-1 knapsack problem model. Define state variables. Indicates before consideration The program has a usable storage capacity of [number] programs. The task caching decision optimization problem can be viewed as maximizing the program caching efficiency while satisfying the UAV's cache space constraints.

[0152]

[0153] P1 can be viewed as a 0 / 1 knapsack problem, where Considered a program In drones The value of the above Its weight.

[0154] Step 5-3: Solve P1 using dynamic programming, decomposing the original problem into overlapping subproblems, and solving them bottom-up by filling in tables. For each UAV... Starting with the first program, we gradually considered whether to add caching to each program, based on the program's storage space requirements. and benefit value Update state variables. First, initialize all cached decision variables: , Initialize the dynamic programming table , Then, iterate through all drones. Consider each procedure in sequence. Fill in the dynamic programming table. Specifically, according to the program... storage requirements Compared with current capacity The relationship is addressed in two separate cases. When... At that time, the program The storage demand exceeds the currently available capacity, so the program cannot be cached at this time. (Inherited from previous version) The program has capacity The optimal solution under the following conditions, i.e. .when At that time, the program It can be cached; we need to compare the benefits of caching versus not caching: (Effects of not caching:) Benefits of caching: Choose the option with the greater benefits:

[0155] .

[0156] Step 5-4: After constructing the dynamic programming table, use backtracking to determine the optimal caching scheme. Starting from the final state, trace backwards, examining the decision record of each program. If including the current program improves the overall efficiency, add the program to the cache set and deduct the corresponding storage space; otherwise, skip the program. In this way, the optimal caching scheme is ultimately determined for each drone. Generate optimal binary cache decision variables Ensure storage capacity Maximize caching efficiency under constraints.

[0157] Step 6 above includes:

[0158] Step 6-1: Determine the drone trajectory during the preprocessing stage User association and caching strategies Subsequently, the real-time execution phase continues to optimize the binary variables related to the unloading strategy. , and continuous variables That is, to solve problem P2:

[0159]

[0160] Due to the computational complexity of the task Medium random error Distribution Unknown, based on available moment estimation information and Constructing an uncertain set The set contains All possible probability distributions:

[0161] .

[0162] Then, based on uncertain sets By employing a distributed robust optimization method, the chance constraints are transformed into distributed robust chance constraints, and the worst-case solution is sought.

[0163] ;

[0164] in express The lower bound of the possible distribution. The confidence level.

[0165] Step 6-2: Introduce the conditional value-at-risk mechanism to address the lack of uncertain parameters. The distributed information's chance constraints are transformed into a tractable form. For uncertain parameters... loss function on Based on the definition of conditional value at risk, the Bruker opportunity constraint is equivalent to the worst-case conditional value at risk:

[0166] ;

[0167] in express The upper bound of the possible distribution.

[0168] Furthermore, for the loss function In the safety factor The worst-case conditional value-at-risk can be approximated by a second-order cone programming form:

[0169]

[0170] in and These are uncertain parameters. The mean and standard deviation, , , , , As an auxiliary continuous variable.

[0171] Step 6-3: Based on the above theory, P2 can be further transformed into a mixed-integer second-order cone constraint problem P3:

[0172]

[0173] in

[0174]

[0175] Uncertainty parameter-related variables are designed to address the error in computational complexity estimation, and are used to quantify the degree to which the actual computational complexity deviates from the ideal state. Auxiliary variable A set of.

[0176] Step 7 above includes:

[0177] Step 7-1: For the mixed-integer second-order cone programming problem P3 obtained in Step 6, due to continuous variables... Auxiliary variables Binary unloading decision , The strong coupling between these factors presents a significant challenge for direct solutions. Based on the inherent relationship between binary offloading decisions, a utility function can be introduced to decouple and reconstruct complex problems.

[0178] Specifically, when binary unloading decision , When determined, P4 is about and The convex optimization subproblem can be directly derived from the optimization solver, as follows:

[0179]

[0180] Each task has three offloading modes per time slot: no offloading, UAV-based offloading, and high-altitude platform-based offloading. Therefore, a corresponding utility function can be constructed by pre-calculating the maximum offloadable data volume for each task under different offloading modes. Specifically, in each time slot... For associated drones (satisfy ) task Substitute into and Solve optimization problem P4 to obtain... and As a task The maximum unloadable ratio in both drone unloading and high-altitude platform unloading scenarios.

[0181] Step 7-2: Introduce ternary decision variables The uninstallation mode is defined as follows:

[0182]

[0183] This variable represents the complete unloading status of each task in each time slot, combining the originally complex binary variables. and Transforming it into a single decision variable significantly simplifies the understanding of spatial structure. Based on Further define the utility function to quantify tasks under different unloading modes. In the time slot The amount of data that can be successfully processed:

[0184]

[0185] Step 7-3: By introducing unloading utility, P3 can be equivalently transformed into an integer programming problem P5 with the goal of maximizing the total system utility. Thus, the complex resource allocation and unloading decision problem is transformed into an integer programming problem with the goal of maximizing the total system utility.

[0186]

[0187] To further address the constraints in problem P5, a penalty function mechanism is introduced, and the fitness function is defined as follows:

[0188] ;

[0189] in The penalty factor is set to a constant.

[0190] Step 7-4: Design a heuristic algorithm based on tabu search, directly applying it to the ternary decision variables. The search optimization is performed on the above. The solution to the optimization problem, i.e., the offloading decisions of all tasks in all time slots, is encoded into a length of vector Each element of the vector Corresponding to the task In the time slot The unloading decision takes values ​​from a set. These correspond to three modes: no uninstallation, drone uninstallation, and high-altitude platform uninstallation.

[0191] During algorithm initialization, a set of initial candidate solutions that satisfy cache constraints is first generated. Specifically, for each decision variable in each solution vector. According to its associated drones Cache status Determine the feasible value space: If the required program is already cached, then Available Choose randomly from; otherwise, you can only choose from... Select from the options. Calculate the fitness value for each initial solution. Then, the solution with the highest fitness is selected as the current optimal solution. and the global optimal solution .

[0192] In each iteration In the middle, from the current solution, the following two mutation mechanisms are used. generate One candidate neighbor solution:

[0193] 1. Single-point mutation: Randomly select a position in the solution vector and change its corresponding decision variable. The value is modified to another random value within the same feasible region. This operation enables local fine-tuning of the solution, facilitating a fine search within the neighborhood of the current solution.

[0194] 2. Task-type-based mutation: First, randomly select a drone cluster and a task type, then mutate all drones of that type within the cluster. The task unloading decision is modified simultaneously. This mutation method can achieve significant changes in the solution, helping to escape local optima and enhancing the algorithm's global exploration capability.

[0195] All mutation operations strictly adhere to cache constraints, only when... hour, The value can be 1 (drone unloading mode), that is

[0196]

[0197] In the In the next iteration, the generated candidate solution set Calculate each candidate solution fitness value The candidate solutions are sorted in descending order of fitness value. Each candidate solution is examined sequentially. If all moves of a candidate solution are not in the tabu list and its fitness is better than the current solution, then the candidate solution is accepted. If some moves of a candidate solution are tabu, but its fitness is better than the global optimum, then the disregard criterion is invoked, and the candidate solution is accepted as an exception. If no candidate solution has been accepted after a complete traversal, then a backtracking mechanism is activated, accepting the first non-tabus candidate solution to ensure the continuity of the search process.

[0198] Then update the tabu list. The tabu list records the most recently accepted reverse operation, and each tabu item is represented by a triple. In the form of, This indicates the position of the decision variable in the solution vector. and These represent the values ​​before and after the move, respectively. All attempts to move the position... The value is determined by Transform into Operations that violate this rule will be rejected. When the tabu expression reaches the preset capacity, the earliest added tabu item is removed using a first-in, first-out (FIFO) strategy. This mechanism effectively prevents the algorithm from repeatedly accessing the same solution in the short term, promoting diversified exploration of the search space.

[0199] In each iteration, update the current solution. And compare it with the global optimal solution. The algorithm updates the global optimum as needed, adjusting its fitness. The algorithm reaches its maximum number of iterations. Then terminate and output the decoded optimal unloading strategy. , and the corresponding uninstallation ratio .

[0200] Step 8 above includes: based on the optimization results of the preceding stages, designing and implementing a complete global collaborative optimization algorithm to systematically integrate the optimization processes of the preprocessing stage and the real-time execution stage. First, a two-stage collaborative global optimization framework is constructed, where the preprocessing stage provides basic resource configuration for the system, and the real-time execution stage dynamically optimizes and adjusts based on the preprocessing results. The preprocessing stage executes step 4 to optimize the UAV trajectory. User-Drone Relationship Then, based on the obtained associations, step 5 is executed to determine the program caching strategy. During system operation, based on the preprocessing stage... , and Collect network status information for the current time slot, and execute steps 6 and 7 to solve the real-time offloading strategy to obtain... , and This enables robust and efficient operation in environments with uncertain task complexity, providing a complete communication-computing-caching-trajectory collaborative optimization solution for low-altitude intelligent networks.

[0201] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0202] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for collaborative optimization of communication computing cache and trajectory in low-altitude intelligent network, characterized in that, The method includes the following steps: S1 constructs a low-altitude intelligent network architecture that coordinates high-altitude platforms and multiple drones to provide task offloading services for mobile users; S2, based on the low-altitude intelligent network framework, establishes a communication model, a computation model, a caching model and a UAV trajectory model that include uncertainty in task computational complexity, constructs an optimization problem with the goal of maximizing the amount of data successfully unloaded, and introduces task latency opportunity constraints. S3 breaks down the entire process into a preprocessing stage and a real-time execution stage. The preprocessing stage optimizes the drone trajectory and program caching strategy, while the real-time execution stage optimizes the task unloading strategy and ratio. S4. In the preprocessing stage, the drone trajectory of all time slots and the static correlation between mobile users and drones are optimized based on the K-means++ algorithm, guided by the user's location. S5. In the preprocessing stage, a dynamic optimization algorithm is designed based on the clustering results to determine the set of pre-buffered task processing programs for the UAV. S6. Based on the joint optimization model of low-altitude intelligent network communication computing cache, in view of the uncertainty of task computing complexity, a moment uncertainty set is constructed. Based on the distributed robust optimization method, the task delay opportunity constraint is transformed into a distributed robust opportunity constraint. Then, using the conditional risk value theory, the distributed robust opportunity constraint is conservatively approximated and transformed into a mixed integer second-order cone constraint. S7 introduces a utility function representing different unloading modes, reconstructs the original mixed integer nonlinear programming problem into an integer programming problem with the goal of maximizing the total utility of the system, and designs a tabu search algorithm to solve the reconstructed integer programming problem, thereby obtaining the optimal real-time unloading strategy and unloading ratio. S8 is designed with a global algorithm that integrates the optimization results of the preprocessing stage and the real-time execution stage, and outputs the final UAV trajectory, program caching strategy, task unloading strategy and unloading ratio scheme to achieve robust and efficient operation of low-altitude intelligent networks under uncertain task complexity.

2. The method for low-altitude intelligent network communication computing cache and trajectory collaborative optimization according to claim 1, characterized in that, Step S1 specifically includes: Construct a hierarchical low-altitude intelligent network system, which includes a high-altitude platform, A drone equipped with a mobile edge computing server and For mobile users, using a discrete time slot model, the total system runtime is divided into: There are 3 equal-length time slots, each with a duration of 1. Each user generates a computing task per second per time slot. The task can be partially offloaded to its associated drone, or relayed via drone to a high-altitude platform for processing. For users With drones binary associative variables, when the user With drones When establishing a connection, ,otherwise , The coordinates of the aerial platform are fixed as follows: drones In the time slot The horizontal coordinate is The flight altitude is fixed at Mobile users In the time slot The horizontal coordinate is Mobile users With drones In the time slot The distance between them is drones With the high-altitude platform in the time slot The distance is . 3.The low-altitude intelligent networking communication computing cache and trajectory cooperative optimization method according to claim 2, characterized in that, Step S2 specifically includes: To model the task, we assume that there exists Types of tasks, users In the time slot The generated tasks correspond one-to-one with the task type. Within each time slot, the matching relationship between the tasks generated by each user and the task type can be represented as follows: ,in Indicates user In the time slot The types of tasks generated. Each task type For a specific computational program, a task of that type can only be unloaded if and only if the server has pre-cached the program and database required for the task computation. The cache size is Mobile users In the time slot The generated task is represented as ,in For the amount of task data, For task type The corresponding computational complexity, considering actual uncertainties, is categorized as follows: The computational complexity of the task is modeled as ,in This is an estimated value. For those with unknown distributions Random error, assuming Some statistical information can be obtained through long-term observation, and its first moment and second moment It is known that, among which The mean, To account for variance, the system adopts a partial unloading mode, with the unloading ratio... Indicates user In the time slot Will The amount of data is offloaded to its associated drone. The task offloading between the drone and the high-altitude platform adopts a binary offloading strategy. Communication between mobile users and drones uses a line-of-sight channel model with a channel gain of [value missing]. ,in For the reference channel gain at a distance of 1 meter, the user To drones The uplink transmission latency is time slot At that time, drones users The transmission latency of the unloaded task to the high-altitude platform is ; For binary cache decision variables, if the drone The processing was cached. For programs of type [type] tasks, then ,otherwise binary decision variables and These represent the calculation location selection, when The time indicates the mission is in the drone Local computation, when The time indicates that the task is forwarded to the high-altitude platform for computation. When the task is processed by the UAV, the computation latency is... High-altitude platform for handling tasks The computation delay is ; Task In the time slot Total execution latency Including transmission latency from mobile users to drones Transmission latency from UAV to high-altitude platform UAV computational latency and high-altitude platform calculation delay Total amount of data successfully processed by the system for ， Considering task latency constraints, cache capacity limitations, computational access limitations, and variable flow constraints, the drone trajectory is jointly optimized. User-Drone Relationship Drone program caching strategy Task unloading decision and and task uninstallation ratio To maximize system efficiency, the following optimization problem is established: ； in, Indicates based on uncertainty distribution Opportunity constraints For confidence level, The maximum cache capacity limit for a single drone, The user task matching relationship represents the user In the time slot The generated task belongs to type , Used to represent drones Has the user been pre-cached? In the time slot The corresponding program libraries required for the generated tasks and These represent the maximum number of tasks that drones and high-altitude platforms can handle simultaneously, respectively. It represents the minimum and maximum values ​​of the region boundary.

4. The method for low-altitude intelligent network communication computing cache and trajectory collaborative optimization according to claim 3, characterized in that, Step S3 specifically includes: The overall optimization process is broken down into two stages: a preprocessing stage and a real-time execution stage. In the preprocessing stage, the drone trajectory is optimized. Mobile user-drone relationship Program caching strategy During the real-time execution phase, the task offloading strategy is optimized based on the infrastructure configuration determined in the preprocessing phase. and Task uninstallation ratio . 5.The low-altitude intelligent networking communication computing cache and trajectory collaborative optimization method according to claim 4, wherein, Step S4 specifically includes: In the system initial time slot ,from The location of a user is randomly selected from among the users as the first cluster center. For each user Calculate the minimum distance from the selected cluster center. and calculate user The probability of being selected as the next cluster center Select the user with the highest probability as the new cluster center, and then repeat the process until the desired cluster is obtained. The initial cluster centers, in time slots Calculate each user To each cluster center The distance is used to reassign users to the nearest cluster. Indicates belonging to a cluster The user set, The cluster center represents the number of users in the cluster. The location update rules are as follows: ； Repeat the above allocation and update process until the results converge, and finally obtain... The cluster center location is set as the time slot of the drone. The initial deployment location is determined, each user is assigned to the nearest cluster, and a communication link is established with the corresponding drone. Then, the connection relationship is determined based on the clustering results. If the user... Belongs to cluster center Then set ,otherwise For each time slot ,based on The real-time distribution of users in the data center is used to recalculate the drones according to the above update rules. The trajectory path points, and ultimately, the sequence of UAV positions in each time slot constitutes the UAV trajectory for the entire period. At the same time, it obtained the connection relationship between mobile users and drones. . 6.The low-altitude intelligent networking communication computing cache and trajectory cooperative optimization method according to claim 5, wherein, Step S5 specifically includes: Define each program In drones Cache efficiency Define a state variable as a weighted sum of the normalized data size and the computational complexity of normalization for each user task in each cluster. Indicates before consideration The program has a usable storage capacity of [number] programs. The task caching decision optimization problem P1, which aims to maximize the efficiency of program caching while satisfying the UAV's cache space constraints, can be viewed as maximizing the efficiency of program caching. Initialize and set initial values ​​for all cached decision variables, and initialize the dynamic programming table. , Then, iterate through all drones. Consider each procedure in sequence, according to the procedure. storage requirements Compared with current capacity The relationship is filled into the dynamic programming table. After the dynamic programming table is completed, the backtracking method is used to determine the optimal caching scheme. Starting from the final state, the process is reversed to check the decision records of each program and finally generate the optimal binary caching decision variables for each drone. 7.The low-altitude intelligent networking communication computing cache and trajectory cooperative optimization method of claim 6, wherein, Step S6 specifically includes: Determining the drone trajectory based on the preprocessing stage User association and caching strategies Subsequently, the real-time execution phase continues to optimize the binary variables related to the unloading strategy. , and continuous variables That is, to solve problem P2: ； Based on the available moment estimation information and Constructing an uncertain set The set contains All possible probability distributions Based on uncertain sets By employing a distributed robust optimization method, the chance constraints are transformed into distributed robust chance constraints, and the worst-case solution is sought. ； in express The lower bound of the possible distribution. For confidence level, Based on conditional risk value, convex optimization is performed, and P2 is further transformed into a mixed integer second-order cone constraint problem P3: ； in Auxiliary variable A set of.

8. The method for low-altitude intelligent network communication computing cache and trajectory collaborative optimization according to claim 7, characterized in that, Step S7 specifically includes: Each task has three offloading modes per time slot: no offloading, drone-based offloading, and high-altitude platform-based offloading. For the association satisfies drones Task Substitute into and Solve optimization problem P4 to obtain... and As a task The maximum unloadable ratio in both drone unloading and high-altitude platform unloading scenarios. ； Introducing ternary decision variables The uninstallation mode is defined as follows: ； Further define the utility function to quantify tasks under different unloading modes. In the time slot The amount of data that can be successfully processed: ； P3 is equivalently transformed into integer programming problem P5, which aims to maximize the total utility of the system. ； Introducing a penalty function mechanism, the fitness function is defined as follows: ； in The penalty factor is set to a constant, and the solution to the optimization problem is encoded into a variable of length . vector Each element of the vector Corresponding to the task In the time slot The unloading decision corresponds to three modes: no unloading, drone unloading, and high-altitude platform unloading. First, an initial set of candidate solutions that satisfy the cache constraints is generated. Calculate the fitness value for each initial solution. Then, the solution with the highest fitness is selected as the current optimal solution. and the global optimal solution , In each iteration In the process, mutations can be performed through single-point compilation mutations or task-type-based mutations from the current solution. generate Given candidate neighbor solutions, the generated candidate solution set... Calculate each candidate solution fitness value The candidate solutions are sorted in descending order of fitness value. Each candidate solution is examined sequentially. If the move of a candidate solution is not in the tabu list and its fitness is better than the current solution, the candidate solution is accepted. If the move of a candidate solution is tabu, but its fitness is better than the global optimum, the candidate solution is accepted as an exception. If no candidate solution is accepted after a complete traversal, a backoff mechanism is activated, and the first non-tabulated candidate solution is accepted to ensure the continuity of the search process. In each iteration, the tabu list and the current solution are updated. And compare it with the global optimal solution. The fitness of the algorithm is used to update the global optimum. The algorithm reaches its maximum number of iterations. Then terminate and output the decoded optimal unloading strategy. , and the corresponding uninstallation ratio .

9. The method for low-altitude intelligent network communication computing cache and trajectory collaborative optimization according to claim 8, characterized in that, Step S8 specifically includes: A two-stage collaborative global optimization framework is constructed, in which the preprocessing stage provides basic resource configuration for the system, and the real-time execution stage performs dynamic optimization and adjustment based on the preprocessing results. The preprocessing stage executes S4 to optimize the UAV trajectory. User-Drone Relationship Then, based on the obtained associations, S5 is executed to determine the program caching strategy. During system operation, based on the preprocessing stage... , and Collect network state information for the current time slot, and execute S6 and S7 to solve the real-time offloading strategy to obtain... , and This enables robust and efficient operation in environments with uncertain task complexity.

Images