A load balancing method, device and computer readable storage medium

Abstract

The application provides a load balancing method and device for a space-air-ground integrated network and a computer readable storage medium. The method comprises the following steps: obtaining a load influence factor of each node in the space-air-ground integrated network, wherein the load influence factor comprises a current task amount, a computing capacity and / or a communication delay between nodes; determining a task allocation strategy according to the load influence factor of each node by using a Harris hawk algorithm, wherein a step length of a Levy flight used by the Harris hawk algorithm is controlled by an adjustment factor, and the adjustment factor is related to an iteration number; and updating data flow of each node according to the task allocation strategy. By adjusting the step length of the Levy flight by using the adjustment factor, a balance between strong global exploration in the early stage of iteration and adaptive jumping out of a local optimum in the later stage of iteration is achieved, so that a more optimal task allocation is obtained in a dynamic space-air-ground integrated network, data flow distribution is more balanced, and the processing capacity and response speed of the whole network are improved.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a load balancing method, apparatus, and computer-readable storage medium. Background Technology

[0002] Lévy flight is a random walk process with a step size following a Lévy distribution. This distribution has a heavy-tailed characteristic, which allows the search process to make long jumps in the solution space, thus helping to discover new search areas. Introducing Lévy flight can enhance the randomness of the algorithm. However, in the later stages of iteration, it can still get stuck in local optima. Summary of the Invention

[0003] To address the aforementioned technical problems, this application proposes a load balancing method, apparatus, and computer-readable storage medium for an integrated air-space-ground network.

[0004] To address the aforementioned technical problems, this application proposes a load balancing method for an integrated air-space-ground network. The method includes: obtaining the load impact factor of each node in the integrated air-space-ground network, wherein the load impact factor includes the current workload, computing power, and / or communication latency between nodes; using the Harris Eagle algorithm, determining a task allocation strategy based on the load impact factors of each node, wherein the step size of the Levy flight used in the Harris Eagle algorithm is controlled by an adjustment factor related to the current iteration number; and updating the data traffic of each node according to the task allocation decision.

[0005] Wherein, the adjustment factor includes an adjustment function with the current iteration number as the independent variable, and the adjustment function is any one of a sine function, a cosine function, or a linearly decreasing function; and / or, the adjustment factor first gradually increases and then gradually decreases with the increase of the current iteration number, or monotonically decreases with the increase of the current iteration number.

[0006] The distribution correction coefficient for the Levy flight step length is obtained by adjusting the original correction coefficient using the adjustment factor.

[0007] The Lévy step size is determined by the product of the Lévy step size numerator random number and the distribution correction coefficient, and the Lévy step size denominator random number. The Lévy step size numerator random number follows a normal distribution. The distribution correction coefficient is obtained by multiplying the adjustment factor and the original correction coefficient. The original correction coefficient is the standard deviation of the Lévy step size numerator random number. The Lévy step size denominator random number follows a standard normal distribution.

[0008] The step of using the Harris Eagle algorithm to determine the task allocation strategy based on the load influence factor of each node includes: generating multiple different task allocation schemes as an initial population based on the load influence factor of each node; repeating the following iterative process until a preset maximum number of iterations is reached; selecting the current optimal allocation scheme from the current population using the comprehensive performance index of the task allocation schemes; calculating the escape energy value based on the current iteration number, and determining different encirclement strategies based at least on the escape energy value to update each of the task allocation schemes based on the current optimal allocation scheme, wherein the different encirclement strategies include a dive encirclement strategy, which includes executing the Levi flight; and using the optimal allocation scheme obtained at the end of the iteration as the task allocation decision.

[0009] The comprehensive performance indicators include at least the overall time cost and the load balancing degree between nodes.

[0010] The method further includes: using chaotic mapping to generate multiple allocation schemes as the initial population.

[0011] The method of using the Harris Eagle algorithm to determine the task allocation strategy based on the load influence factor of each node further includes: increasing the adjustment factor in response to the current optimal allocation scheme failing to meet the improvement requirements in multiple consecutive iterations.

[0012] The process of updating the data traffic of each node includes: designating each node as the current node; taking the total amount of task data received by the current node from other nodes at the current moment as the received traffic of the current node; and configuring the sending traffic of the current node at the next moment to be equal to the received traffic, wherein the sending traffic of the current node at the next moment is the total amount of task data that the current node needs to forward to other nodes at the next moment.

[0013] The nodes include ground nodes, high-altitude communication nodes, UAV nodes, and satellite nodes. The ground nodes are used to process preset computing tasks. The high-altitude communication nodes are configured with lightweight mobile edge computing servers to process the unloading tasks of mobile terminals, and / or the high-altitude communication nodes forward the unloading tasks of mobile terminals to cloud data centers for processing. The satellite nodes are configured with lightweight mobile edge computing servers to process the unloading tasks of mobile terminals. The high-altitude UAV nodes in the UAV nodes are configured with mobile edge computing servers to process the unloading tasks of the low-altitude UAV nodes in the UAV nodes.

[0014] To address the aforementioned technical problems, this application proposes a load balancing device for an integrated air-space-ground network, comprising a processor and a memory connected to the processor, wherein the memory stores program instructions; the processor is used to execute the program instructions stored in the memory to implement the method described above.

[0015] To address the aforementioned technical problems, this application proposes a computer-readable storage medium storing program instructions that, when executed, implement the method described above.

[0016] Compared with existing technologies, the beneficial effects of this application are: by acquiring load influencing factors including current task load, computing power, and communication latency, it can comprehensively reflect the real-time status of network nodes, making the task allocation strategy more closely aligned with the actual network load. Utilizing the Harris Eagle algorithm combined with an adjustment factor related to the iteration count to control the Levy flight step size, the algorithm possesses strong global exploration capabilities in the early stages of iteration, enabling a broad search of the solution space. Simultaneously, by adjusting the step size amplitude in the later stages of iteration, the algorithm helps escape local optima traps, further improving convergence speed and solution accuracy. This adaptive search strategy enables the integrated air-space-ground network to achieve better task allocation when facing dynamically changing traffic and complex node relationships, resulting in a more balanced data traffic distribution and improved overall network processing capacity and response speed. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of an embodiment of the load balancing method for the integrated air-space-ground network provided in this application; Figure 2 This is a schematic diagram of the improved Levi flight trajectory with dynamic parameter control provided in this application; Figure 3 This is a structural diagram of the integrated air-space-ground network system provided in this application; Figure 4 This is a dynamic network topology diagram based on cellular automata provided in this application; Figure 5 This is the integrated space-air-ground cellular automaton network model provided in this application; Figure 6 This is a schematic diagram of the structure of an embodiment of the load balancing device for the integrated air-space-ground network provided in this application; Figure 7 This is a schematic diagram of the structure of an embodiment of the computer storage medium provided in this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0019] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] The subject of this application is a load balancing optimization scheme applied to an integrated air-space-ground network environment. Its protection boundary lies in dynamically adjusting task allocation based on multi-dimensional load influence factors using an improved Harris Eagle algorithm, thereby optimizing network traffic distribution. Please refer to [link to details]. Figure 1 , Figure 1 This is a schematic diagram of an embodiment of the load balancing method for the integrated air-space-ground network provided in this application.

[0021] Step S11: Obtain the load impact factor of each node in the integrated air-space-ground network. The load impact factor includes the current workload, computing power, and / or communication latency between nodes.

[0022] Specifically, the first step is to obtain the load impact factor of each node in the integrated air-space-ground network, which includes the current workload, computing power, and communication latency between nodes.

[0023] The load impact factors include at least the following three types of information: the amount of computing tasks currently pending on the node, the available computing power of the node, and the communication latency between nodes.

[0024] In a specific embodiment of this application, each node maintains a local task queue with a current workload. When a new task arrives or a current task is completed, the node updates the queue length in real time. For example, if a low-Earth orbit satellite node has 5 computational tasks in its pending task queue at the current moment, then the node's current workload is 5.

[0025] In a specific embodiment of this application, the computing power of each node can be determined by the processing power of the server configured for that node. The communication latency between nodes can be obtained through active measurement or link state estimation.

[0026] In a specific embodiment of this application, the acquisition of load impact factors does not depend on a single central node, but adopts a distributed approach. For example, each node only acquires its own and its neighboring nodes' load information, which is gradually spread to the entire network through information exchange between nodes.

[0027] Step S12: Using the Harris Eagle algorithm, determine the task allocation strategy based on the load impact factor of each node.

[0028] This application utilizes an improved Harris Eagle algorithm to solve the task allocation strategy. The core improvement of this algorithm lies in the step size control in its Levy flight mechanism.

[0029] The Lévy flight is a random walk process whose step size follows a Lévy distribution with heavy tails. This distribution allows the search process using the Harris Eagle algorithm to make long jumps in the solution space, which helps to discover new search areas.

[0030] To address the problem that traditional Lévy flight is prone to getting trapped in local optima in the later stages of iteration, this application introduces an adjustment factor related to the number of iterations. This allows for a larger step size in the early stages of iteration to enhance global exploration capabilities, while in the later stages of iteration, the step size amplitude is adjusted to achieve more refined local exploration or escape from local optima traps, thereby balancing global search and local convergence.

[0031] In this application, the step size of the Lévy flight is not fixed or controlled by a simple linear factor, but is controlled by an adjustment factor related to the current iteration number. For example... Figure 2 As shown, Figure 2 This is a schematic diagram of the improved Levi flight trajectory with dynamic parameter control provided in this application.

[0032] In one embodiment of this application, the adjustment factor includes an adjustment function with the iteration number as the independent variable, and the adjustment function is any one of a sine function, a cosine function, or a linearly decreasing function.

[0033] The adjustment factor is adopted by means of a cosine function or a specific nonlinear convergence formula. As the number of iterations increases, the value of the adjustment factor changes, thereby changing the step size distribution of the Lévy flight. This makes the algorithm more focused on the region near the current optimal solution in the later stages of the search, thus improving the convergence accuracy.

[0034] When the adjustment factor is a cosine function, the adjustment factor is larger in the early stage of iteration, fluctuates in the middle stage, and is smaller in the later stage. Thus, the step size is further compressed in the later stage of iteration through the tail characteristics of the cosine curve, achieving fine convergence.

[0035] When the adjustment factor is a sine function, the symmetry and fluctuation of the sine function are utilized to enable the algorithm to repeatedly adjust the step size amplitude in the middle of the iteration, thereby enhancing its ability to escape local optima.

[0036] When the adjustment function is a linearly decreasing function, the adjustment factor decreases uniformly with the number of iterations, so that the Lévy flight step size transitions smoothly from large to small.

[0037] In other embodiments of this application, the adjustment factor can also be other dynamic strategies, as long as it is related to the number of iterations and can control the amplitude of the Levy flight step, so as to achieve a smooth transition from coarse-grained exploration to fine-grained development.

[0038] By limiting the adjustment function to any one of the sine, cosine, or linearly decreasing functions, the change in the Levy flight step size has a clear mathematical pattern and is controllable, which is beneficial for the algorithm to adaptively adjust the search intensity according to the iteration stage.

[0039] In another embodiment of this application, the adjustment factor first gradually increases and then gradually decreases as the iteration number increases.

[0040] Specifically, when the adjustment factor first increases and then decreases with the number of iterations, a quadratic function or a piecewise function can be used in the implementation. For example, it can be linearly increased in the first half of the iteration and linearly decreased in the second half, so that the algorithm has the largest step size amplitude in the middle of the search, thereby jumping more significantly in the solution space and increasing the probability of finding the global optimal solution.

[0041] By limiting the adjustment factor to gradually increase and then gradually decrease with the number of iterations, the algorithm can obtain a larger step size amplitude in the middle of the iteration, which is conducive to making larger jumps in the solution space, increasing the diversity of the search, and thus reducing the risk of getting trapped in local optima.

[0042] In one embodiment of this application, the distribution correction coefficient of the Levy flight step is obtained by adjusting the original correction coefficient using an adjustment factor.

[0043] In one embodiment, the Lévy step size is determined by the product of the Lévy step size numerator random number and the distribution correction coefficient, and the Lévy step size denominator random number. The Lévy step size numerator random number follows a normal distribution, and the distribution correction coefficient is obtained by multiplying the adjustment factor by the original correction coefficient. The original correction coefficient is the standard deviation of the Lévy step size numerator random number, and the Lévy step size denominator random number follows a standard normal distribution.

[0044] The specific formula is as follows:

[0045] Where ξ is the adjustment factor, β is the Lévy exponent, and Γ is the gamma function. Let Lévy step size be the standard deviation of the random number in the numerator. The standard deviation of the random number in the denominator of the Levy step size is fixed at 1.

[0046] The Lévy step size is a displacement vector used to guide the update of an individual's position in the Harris Eagle algorithm; its magnitude determines the search stride. The random number in the Lévy step size numerator follows a normal distribution and is typically used to generate random fluctuations with Gaussian characteristics, while the random number in the Lévy step size denominator follows a standard normal distribution and is used to normalize or scale the numerator. Combining the two can construct a step size distribution that conforms to specific statistical properties.

[0047] The original correction coefficient is defined as the standard deviation of the random number of the Lévy step size numerator. The distribution correction coefficient is a dynamic value obtained by multiplying the adjustment factor by the original correction coefficient, where the adjustment factor changes with the number of iterations.

[0048] This application enables a large step size in the early stages of iteration, where a larger adjustment factor combined with the standard deviation produces a large step size. This facilitates extensive global exploration within the vast solution space of the integrated air-space-ground network, allowing for rapid location of potential optimal regions. In the later stages of iteration, a smaller adjustment factor compresses the step size amplitude, which is beneficial for fine-grained searches in local regions. This improves the convergence accuracy of the load balancing scheme, effectively avoiding the problem of traditional Levy flight easily getting trapped in local optima in the later stages of iteration, and enhancing the overall efficiency and reliability of load balancing.

[0049] In one embodiment of this application, if the current optimal allocation scheme does not meet the improvement requirements during multiple consecutive iterations, the adjustment factor is increased.

[0050] Specifically, when the overall performance index of the current optimal allocation scheme does not change significantly or does not meet the preset improvement threshold in multiple consecutive iterations, it is determined that the algorithm may be trapped in a local optimum or the search efficiency may be reduced.

[0051] At this point, the value of the adjustment factor is increased. Increasing the adjustment factor will directly lead to a larger amplitude of the Lévy flight factor, thus causing the searcher to make random jumps of a larger span in the solution space.

[0052] For example, the specific number of consecutive iterations is set to 5, and the improvement requirement is defined as the improvement of the objective function value of the current optimal allocation scheme is greater than 0.1%. When the improvement is not achieved in 5 consecutive iterations, the adjustment factor is multiplied by 1.2, thereby increasing the step size of Levy's flight.

[0053] By setting a mechanism to increase the adjustment factor when the current optimal allocation scheme does not meet the improvement requirements in consecutive iterations, the algorithm can automatically identify the search stagnation state and enhance the global exploration capability by increasing the jump amplitude of Levy's flight in this state. This helps to escape the local optimum trap and further improve the global optimization accuracy of the task allocation strategy and the convergence stability of the algorithm.

[0054] Furthermore, this application proposes an embodiment that further defines the specific implementation steps for determining the task allocation strategy using the Harris Eagle algorithm. The specific steps are as follows: Based on the load impact factor of each node, generate multiple different task allocation schemes as the initial population; repeat the following iterative process until the preset maximum number of iterations is reached; select the current optimal allocation scheme from the current population using the comprehensive performance index of the task allocation schemes; calculate the escape energy value based on the current number of iterations, and determine different encirclement strategies based at least on the escape energy value, so as to update each of the task allocation schemes based on the current optimal allocation scheme.

[0055] Among them, different encirclement strategies include a dive-encirclement strategy, which includes executing the Levi flight; and using the optimal allocation scheme obtained at the end of the iteration as the task allocation decision.

[0056] Specifically, the optimal allocation scheme is first selected by using comprehensive performance indicators, which can be a fitness function value composed of multiple objectives such as response time, throughput, and resource utilization.

[0057] Subsequently, the escape energy value is calculated based on the current iteration number. This escape energy value reflects the simulated prey's ability to escape, and different encirclement strategies are determined accordingly. When certain conditions are met, the algorithm enters a dive-encirclement strategy, which explicitly includes the step of performing a Lévy flight. A Lévy flight is a random walk process with a step size following a Lévy distribution, and its heavy-tailed characteristic allows the searcher to make long-distance jumps in the solution space.

[0058] In this step, the improved Levi flight described in any of the embodiments of this application is performed. By switching between dynamic strategies based on escape energy and Levy flight, the algorithm can achieve a balance between exploration and exploitation.

[0059] In a specific implementation of this application, the following steps are adopted: The Harris Hawks Optimization Algorithm (HHO) simulates the hunting behavior decisions of Harris Hawks under different mechanisms. Specifically, it can be divided into three phases: exploration, transition, and attack.

[0060] During the exploration phase, individuals will update their location using the following formula:

[0061]

[0062]

[0063] Where X is the position of the individual, and t is the current iteration number. The position of an individual is randomly selected from the population during iteration. Let r1-r4 and q be random numbers in the interval (0,1), K be a random position in the position space, and LB and UB represent the lower and upper bounds of the individual's position space. Indicates the location of the prey. This represents the average position of all individuals when iterating to generation t.

[0064] In this embodiment of the application, each individual represents a candidate allocation strategy, and the prey represents the current optimal allocation strategy among all candidate allocation strategies.

[0065] During the transition phase, the Harris Eagle algorithm selects different search behaviors based on the prey's escape energy.

[0066] like The algorithm enters the exploration phase, if The algorithm then enters the attack phase. The escape energy can be expressed by the following formula:

[0067] Where E represents the escape energy of the prey, E0 represents a random number in the interval (-1,1), which is the initial energy of the prey, and T is the maximum number of iterations.

[0068] During the attack phase, the algorithm will consider the prey's escape energy. Given the probability r of escaping, choose one of four attack strategies. Here, r is a random number in the interval (0,1).

[0069] when and When the Harris Eagle algorithm selects a soft-surround strategy, the position update is as follows:

[0070]

[0071]

[0072] in, The distance between the prey's location and the current individual's location is given by , J represents the random jump intensity of the prey during its escape, and r5 represents a random number in the interval (0,1).

[0073] when and The Harris Eagle algorithm selects a hard-surround strategy and updates the position as follows:

[0074] when and If the Harris Eagle algorithm selects a dive-encirclement strategy, it updates the individual's position; if the fitness is not improved, it performs a Levi flight; if it is still not improved, it uses the original position.

[0075]

[0076]

[0077]

[0078] in, , Follows a normal distribution. , .

[0079] , The definition is as follows:

[0080] Where β is the Lévy exponent and Γ is the gamma function. Let Lévy step size be the standard deviation of the random number in the numerator. The standard deviation of the random number in the denominator of the Levy step size is fixed at 1.

[0081] In this application, β is usually taken between [0,2], and the value of 1.5 is used as an example.

[0082] when and At that time, the Harris Eagle algorithm selects a dive-encirclement strategy to update the individual's position. If the fitness is not improved, it performs a Levi flight. If the fitness is still not improved, it uses the original position.

[0083]

[0084]

[0085] In this embodiment, the improved Lévy flight factor with nonlinear convergence factor is taken as an example. An adjustment factor ξ is added to control... The magnitude of the adjustment factor can be any one of the embodiments in this application. Here, a sine function is used as an example to control the amplitude of the Levy flight factor, as follows:

[0086] Where ξ is the adjustment factor, t is the current iteration number, and T is the maximum iteration number.

[0087] This is the bias constant.

[0088]

[0089] Where ξ is the adjustment factor, β is the Lévy exponent, and Γ is the gamma function. Let Lévy step size be the standard deviation of the random number in the numerator. The standard deviation of the random number in the denominator of the Levy step size is fixed at 1.

[0090] It should be noted that in this embodiment, b is a bias constant, used as the baseline value for the translational dynamic adjustment factor to adapt to the step size adjustment requirements of different optimization stages. The bias constant and the Lévy index β can be set to the same value, such as... Figure 2 The representation shown may differ, and this application does not impose specific limitations.

[0091] By generating multiple different task allocation schemes as the initial population, the initial diversity of the search space is increased, enabling the algorithm to cover a wider solution region and facilitating the discovery of globally optimal solutions. Using comprehensive performance metrics to select the current optimal allocation scheme ensures that the iteration direction always points to the task allocation state with better performance, improving the algorithm's convergence orientation. The escape energy value is calculated based on the current iteration number, and different encirclement strategies are determined accordingly. This allows the algorithm to dynamically adjust its search behavior by simulating changes in prey escape energy, focusing on global exploration in the early stages of iteration and local development in the later stages, thus adapting to dynamic changes in network load.

[0092] In one embodiment of this application, the comprehensive performance index includes at least the overall time cost and the degree of load balancing among nodes.

[0093] The total time cost refers to the total time spent from the time a task initiates a request to the time it completes computation and returns results at each node, including transmission latency, queuing latency, and processing latency.

[0094] The degree of load balancing among nodes refers to the measure of the load difference among all available computing nodes in the network. It is usually achieved by calculating the variance or standard deviation of the load rate of each node, which is used to reflect the uniformity of load distribution.

[0095] Because integrated air-space-ground networks exhibit dynamic topology changes and heterogeneous resources, simply focusing on time optimization often leads to overload of some hot nodes while leaving edge nodes idle, resulting in network congestion or even service interruptions. By simultaneously considering both time cost and load balancing, the algorithm can achieve low latency while avoiding extreme uneven resource allocation.

[0096] In terms of implementation, the overall performance index can be expressed as the weighted sum of the total time cost and the load balancing penalty, where the load balancing penalty can be the sum of the squares of the differences between the load rate of each node and the average load rate.

[0097] For example, in one embodiment, the overall time cost is estimated by collecting the signal strength and link bandwidth from the user to each access point, combined with the current CPU utilization of the MEC server, using a queuing theory model; the load balancing degree between nodes is measured by real-time monitoring of the task queue length and computing resource utilization of each node, and calculating its standard deviation as a metric.

[0098] In another embodiment, for services that are extremely sensitive to latency, the weight of the overall time cost can be set higher, while for background batch processing tasks that pursue system stability, the weight of the load balancing degree between nodes can be set higher.

[0099] In addition, the degree of load balancing can also be reflected as the difference between the load of the largest node and the load of the smallest node, or the proportion of the node with the highest load factor, which can be used to specifically reflect the requirements of load uniformity in various mathematical forms.

[0100] By including overall time cost in the comprehensive performance metrics, the algorithm's optimization direction can directly target reducing end-to-end communication and processing latency, thereby improving user service experience and network response speed. By using both overall time cost and load balancing among nodes as optimization objectives, the algorithm establishes an inherent trade-off mechanism between latency reduction and load balancing. This allows for the optimization of overall system performance in the dynamic and complex environment of an integrated air-space-ground network, avoiding problems such as localized resource congestion or increased overall latency caused by optimizing a single metric.

[0101] In one embodiment of this application, chaotic mapping is used to generate multiple allocation schemes as the initial population.

[0102] Specifically, in the Harris Eagle optimization algorithm, the distribution quality of the initial population directly affects the algorithm's global search capability and convergence speed.

[0103] Traditional random initialization leads to uneven distribution of the population in the solution space, resulting in clustering and reduced diversity of the initial population, making the algorithm prone to getting trapped in local optima. This application introduces chaotic mapping to ensure that the initial population is more evenly distributed within the feasible region, thereby covering more potential solution areas.

[0104] In this embodiment, the chaotic mappings that can be used include Logistic mapping, Tent mapping, Sine mapping, or Logistic-Tent-map composite mapping, etc.

[0105] Taking Logistic-Tent-map as an example, this mapping combines the advantages of Logistic mapping and Tent mapping, and has a wider range of chaotic parameters and better traversal.

[0106] The specific working method is as follows: first, the initial values ​​of the chaotic variables are determined; then, a series of chaotic sequences are generated through iterative formulas; finally, these chaotic sequences are mapped to the solution space of the task allocation scheme, that is, the load allocation ratio or path selection probability of each node, to form the initial population.

[0107] For example, the Logistic mapping formula x(n+1)=r*x(n)*(1-x(n)) can be used, where r is a control parameter. When the value is 4, the system is in a completely chaotic state, and the generated x value is mapped to the task allocation weight in the interval [0,1].

[0108] For example, the Tent map can be used, whose piecewise linear properties make the chaotic sequence more evenly distributed within the interval, making it suitable for generating a uniformly distributed initial load distribution scheme. In Example 3, a Logistic-Tent-map composite map can be used, which has the high chaos of the Logistic map in most intervals, while having the linear properties of the Tent map in some intervals, thereby improving search efficiency while maintaining diversity.

[0109] By utilizing chaotic mapping to generate multiple allocation schemes as the initial population, the initial population is more evenly distributed in the solution space, increasing population diversity. This allows the algorithm to cover a wider solution region in the early stages of the search, increasing the probability of finding the global optimum. Furthermore, the ergodicity of chaotic mapping avoids population clustering issues that may occur with traditional random initialization, thereby further improving the algorithm's global search capability and convergence stability.

[0110] Step S13: Update the data traffic of each node according to the task allocation decision.

[0111] Based on the task allocation decisions determined by the algorithm, the data traffic of each node is updated to achieve network load balancing.

[0112] The specific steps are as follows: each node is designated as the current node; the total amount of task data received by the current node from other nodes at the current moment is taken as the received traffic of the current node; the sending traffic of the current node at the next moment is configured to be equal to the received traffic, wherein the sending traffic of the current node at the next moment is the total amount of task data that the current node needs to forward to other nodes at the next moment.

[0113] In one embodiment of this application, the integrated air-space-ground network includes ground nodes, high-altitude communication nodes, unmanned aerial vehicle (UAV) nodes, and satellite nodes, such as... Figure 3 As shown, Figure 3 This is a structural diagram of the integrated air-space-ground network system provided in this application.

[0114] This application establishes an integrated air-space-ground network model based on cellular automata, such as... Figures 4-5 As shown, Figure 4 This is a dynamic network topology diagram based on cellular automata provided in this application. Figure 5 This is the space-air-ground integrated cellular automaton network model provided in this application.

[0115] Cellular automata are efficient spatial and temporal discretization mathematical modeling tools. Two-dimensional cellular automata are used to model space-air-ground network models.

[0116] HAPs-MEC networks refer to the deployment of mobile edge computing (MEC) technology on high-altitude communication platforms (HAPs), which include several high-altitude communication nodes.

[0117] The LEO-MEC network deploys mobile edge computing (MEC) servers on (LEO) satellites, including satellite nodes.

[0118] A UAV-MEC network deploys mobile edge computing (MEC) servers on unmanned aerial vehicles (UAVs), and includes several high-altitude UAV nodes and several low-altitude UAV nodes.

[0119] The G-MEC network deploys mobile edge computing (MEC) servers on terrestrial network infrastructure and includes several terrestrial nodes.

[0120] Initialize the LEO-MEC network. express At time n, the LEO of the satellite node This indicates the MEC server that the LEO satellite node is connected to. Let m and n be the numbers, if =0 means Time points m and n are not connected; if =1 indicates Connect time points m and n.

[0121] Initialize the HAPs-MEC network. express At time n, the high-altitude platform node HAP, This indicates the MEC server to which the high-altitude communication node (HAP) is connected. Represent the topological relationship between the m-th and n-th network nodes, if =0 means Time points m and n are not connected; if =1 indicates that m and n are connected at time t.

[0122] Initialize UAV-MEC network ,express The nth high-altitude unmanned aerial vehicle (UAV) node at time n, This indicates the MEC server to which the high-altitude unmanned aerial vehicle (UAV) node is connected. Represent the topological relationship between the m-th and n-th network nodes, if =0 means Time points m and n are not connected; if =1 indicates that m and n are connected at time t.

[0123] Initialize the G-MEC network. express At time n, the ground node G, This indicates the MEC server that the ground node G is connected to. Represent the topological relationship between the m-th and n-th network nodes, if =0 means Time points m and n are not connected; if =1 indicates that m and n are connected at time t.

[0124] Will , , , The value is used as the current node to calculate the unload request. , , This represents the current computing power of the MEC server.

[0125] Initialize and establish the system network topology according to the above steps, and update the status of each node and the data traffic received and sent on each node.

[0126] Specifically, each node in the network is treated as a current node for processing. For each current node, the total amount of task data it receives from all other connected nodes in the network at the current moment is first calculated, and this sum is defined as the current node's received traffic. Subsequently, when determining the current node's sending strategy for the next moment, the sending traffic that it needs to forward all task data to other nodes in the next moment is directly configured to be equal to the aforementioned received traffic.

[0127] This process demonstrates the application of cellular automata in network traffic evolution, namely, driving global state updates through local rules.

[0128] In one embodiment of this application, the ground node is used to process a preset computing task; the high-altitude communication node is configured with a lightweight mobile edge computing server to process the unloading task of the mobile terminal, and the high-altitude communication node forwards the unloading task of the mobile terminal to the cloud data center for processing; the high-altitude UAV node is configured with a mobile edge computing server to process the unloading task of the low-altitude UAV node.

[0129] Among them, the communication delay in the load impact factor is calculated in real time by the cellular automata model, and the Harris Eagle algorithm dynamically adjusts the step size during the iteration process to cope with the load fluctuations caused by changes in network topology.

[0130] In a specific embodiment of this application, for a low-orbit satellite node, the received traffic includes user computing request data from a ground base station and relay data from an upper-level platform. The transmission traffic at the next moment is configured to forward these received data to adjacent satellites or the ground core network as is, to ensure that the data is not lost and the transmission link load is balanced.

[0131] In specific embodiments of this application, for edge computing server nodes, if they have local processing capabilities, the received traffic may contain some local computing tasks that do not need to be forwarded. However, according to the limitation of this claim, the sent traffic is only for the data portion that needs to be forwarded to other nodes, that is, the amount of data received that needs to be offloaded to the cloud or other edge nodes, thereby precisely controlling the transmission load on the network side.

[0132] In a specific embodiment of this application, for high-altitude platform nodes, the received traffic includes task data from a drone swarm composed of high-altitude drone nodes. The transmission traffic at the next moment is configured to be equal to the received traffic, and the data is forwarded to the ground data center via a satellite-to-ground link. This mechanism avoids data accumulation or loss at network nodes, ensuring transmission synchronization. By forcing the transmission volume at the next moment to be equal to the received volume at the current moment, data transmission in the network follows a strict conservation law, which helps eliminate data backlog between nodes and improves the determinism and stability of network transmission.

[0133] This application comprehensively reflects the real-time status of network nodes by acquiring load influencing factors including current task load, computing power, and communication latency, making the task allocation strategy more closely aligned with the actual network load. By utilizing the Harris Eagle algorithm combined with an adjustment factor related to the iteration count to control the Levy flight step size, the algorithm possesses strong global exploration capabilities in the early stages of iteration, enabling a broad search of the solution space. Furthermore, in the later stages of iteration, adjusting the step size amplitude helps the algorithm escape local optima traps, further improving convergence speed and solution accuracy. This adaptive search strategy allows the integrated air-space-ground network to achieve better task allocation when facing dynamically changing traffic and complex node relationships, resulting in a more balanced data traffic distribution and improved overall network processing capacity and response speed.

[0134] Please refer to details. Figure 6 , Figure 6 This is a schematic diagram of an embodiment of the load balancing device for the integrated air-space-ground network provided in this application.

[0135] The load balancing device 700 in this embodiment includes a processor 71, a memory 72, an input / output device 73, and a bus 74.

[0136] The processor 71, memory 72, and input / output device 73 are respectively connected to the bus 74. The memory 72 stores program data, and the processor 71 is used to execute the program data to implement the load balancing method described in the above embodiments.

[0137] In this embodiment, processor 71 can also be referred to as a CPU (Central Processing Unit). Processor 71 may be an integrated circuit chip with signal processing capabilities. Processor 71 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor, or processor 71 can be any conventional processor.

[0138] This application also provides a computer storage medium; please refer to the following: Figure 7 , Figure 7 This is a schematic diagram of a computer storage medium according to an embodiment of the present application. The computer storage medium 600 stores a computer program 61, which, when executed by a processor, is used to implement the load balancing method of the above embodiment.

[0139] When the embodiments of this application are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0140] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A load balancing method for an integrated air-space-ground network, characterized in that, The method includes: Obtain the load impact factor of each node in the integrated air-space-ground network, wherein the load impact factor includes the current workload, computing power and / or communication latency between nodes; Using the Harris Eagle algorithm, a task allocation strategy is determined based on the load influence factor of each node. The step size of the Levy flight used in the Harris Eagle algorithm is controlled by an adjustment factor, which is related to the iteration number. Based on the task allocation decision, update the data traffic of each node.

2. The method according to claim 1, characterized in that, The adjustment factor includes an adjustment function with the current iteration number as the independent variable, and the adjustment function is any one of a sine function, a cosine function, or a linearly decreasing function; And / or, the adjustment factor first gradually increases and then gradually decreases with the increase of the iteration number, or decreases monotonically with the increase of the iteration number.

3. The method according to claim 1 or 2, characterized in that, The distribution correction coefficient for the Levy flight step size is obtained by adjusting the original correction coefficient using the adjustment factor.

4. The method according to claim 3, characterized in that, The Lévy step size is determined by the product of the Lévy step size numerator random number and the distribution correction coefficient, and the Lévy step size denominator random number. The Lévy step size numerator random number follows a normal distribution. The distribution correction coefficient is obtained by multiplying the adjustment factor and the original correction coefficient. The original correction coefficient is the standard deviation of the Lévy step size numerator random number. The Lévy step size denominator random number follows a standard normal distribution.

5. The method according to claim 1, characterized in that, The method of using the Harris Eagle algorithm to determine the task allocation strategy based on the load impact factor of each node includes: Based on the load impact factor of each node, multiple different task allocation schemes are generated as the initial population. Repeat the following iterative process until the preset maximum number of iterations is reached: The optimal task allocation scheme is selected from the current population using the comprehensive performance index of the task allocation scheme. The escape energy value is calculated based on the current iteration number, and different encirclement strategies are determined based at least on the escape energy value, so as to update each of the task allocation schemes based on the current optimal allocation scheme. The different encirclement strategies include a dive encirclement strategy, which includes executing the Levi flight. The optimal allocation scheme obtained at the end of the iteration is used as the task allocation decision.

6. The method according to claim 5, characterized in that, The method of using the Harris Eagle algorithm to determine the task allocation strategy based on the load impact factor of each node also includes: When the current optimal allocation scheme fails to meet the improvement requirements in multiple consecutive iterations, the adjustment factor is increased.

7. The method according to claim 5, characterized in that, The overall performance metrics include at least the total time cost and the degree of load balancing among nodes.

8. The method according to claim 5, characterized in that, The method further includes: Multiple allocation schemes are generated using chaotic mapping as the initial population.

9. The method according to claim 1, characterized in that, The data traffic for updating each node includes: Each of the aforementioned nodes is designated as the current node; The total amount of task data received by the current node from other nodes at the current moment is taken as the received traffic of the current node; Configure the current node's transmission traffic in the next moment to be equal to the received traffic, wherein the current node's transmission traffic in the next moment is the total amount of task data that the current node needs to forward to other nodes in the next moment.

10. The load balancing method according to any one of claims 1-9, characterized in that, The nodes include ground nodes, high-altitude communication nodes, UAV nodes, and satellite nodes; The ground node is used to process preset computing tasks; The high-altitude communication node is configured with a lightweight mobile edge computing server to handle the unloading task of the mobile terminal, and / or the high-altitude communication node forwards the unloading task of the mobile terminal to the cloud data center for processing. The satellite node is configured with a lightweight mobile edge computing server to handle the offloading tasks of mobile terminals; The mobile edge computing server configured in the high-altitude drone node of the drone node is used to handle the unloading tasks of the low-altitude drone node of the drone node.

11. A load balancing device for an integrated air-space-ground network, characterized in that, Includes a processor and a memory connected to the processor, wherein, The memory stores program instructions; The processor is used to execute program instructions stored in the memory to implement the method as described in any one of claims 1 to 10.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions that, when executed, implement the method as described in any one of claims 1 to 10.

Images