Collaborative dynamic improved load balancing method and system based on SAGIN network

Abstract

The invention discloses a collaborative dynamic improved load balancing method and system based on an SAGIN network, and the method comprises the steps: firstly, carrying out the clustering of satellite nodes, determining an initial clustering center, and carrying out the updating, and obtaining a final clustering subdomain; dividing core nodes and non-core nodes according to the minimum distance between the satellite nodes and the clustering sub-domains, taking the clustering sub-domains as controller nodes, and distributing two types of nodes in combination with a load balancing strategy to form a load balancing multi-controller node sub-domain; and finally, an initial population is generated by combining a three-dimensional trigonometric function chaotic system, a candidate scheme is obtained through exploration-launching iterative updating of a snow ablation algorithm, and a final controller deployment scheme is obtained through optimization by using a Levy flight algorithm. The method can dynamically adapt to the characteristics of the SAGIN network, effectively optimizes the load and deployment of the controller, improves the reliability and efficiency of network communication, and is suitable for an optimized deployment scene of the SAGIN network controller.

Description

Technical Field

[0001] This invention relates to the field of network controller optimization deployment technology, specifically to a collaborative dynamic improvement load balancing method and system based on SAGIN networks. Background Technology

[0002] With the increasing complexity of SAGIN (Space-Air-Ground Integrated Network), achieving the fusion and sharing of network data resources across multiple platforms—space-based, low-altitude, and ground-based—has become crucial for meeting the demands of all-time, all-domain, and all-space communication. Against this backdrop, SDN (Software Defined Networking) technology, with its centralized control and programmable characteristics, offers a novel design approach for the global management of network resources. Through centralized management by SDN, the allocation and scheduling of network resources are optimized, significantly improving the efficiency and reliability of resource management and laying the technological foundation for the efficient operation of SAGIN.

[0003] However, current SDN-based SAGIN multi-controller collaborative architectures commonly suffer from controller load imbalance when the satellite network topology changes dynamically and user service demands are uneven. The root cause of this problem lies in the irrationality of controller subdomain partitioning: due to significant dynamic changes in satellite network topology, traditional subdomain partitioning methods based on random selection or fixed threshold strategies struggle to adapt dynamically to network conditions, leading to inaccurate selection of the control center point and consequently, load imbalance. Although existing research focuses on load distribution and latency optimization, the lack of in-depth consideration of the dynamic adaptability of subdomain partitioning often results in methods exhibiting significant randomness and arbitrariness, making it difficult to effectively solve the load imbalance problem. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a collaborative dynamic improved load balancing method and system based on SAGIN network, which solves the problems of controller load imbalance, excessive network latency, and uneven resource allocation in the prior art.

[0005] This invention is achieved through the following technical solution:

[0006] A collaborative dynamic improvement load balancing method based on SAGIN networks includes the following steps:

[0007] Step 1: Cluster the satellite nodes to determine the initial cluster centers, and then update the initial cluster centers to obtain the final cluster subdomains;

[0008] Step 2: Based on the minimum distance between the satellite nodes and the clustered subdomains, the satellite nodes are divided into core nodes and non-core nodes. The clustered subdomains are used as controller nodes, and in combination with the load balancing strategy, the core nodes and non-core nodes are assigned to different controller nodes to obtain a load-balanced multi-controller node subdomain.

[0009] Step 3: Based on the load balancing multi-controller node subdomain and combined with the three-dimensional trigonometric function chaotic system, an initial population is generated. Based on the snow melting algorithm's "exploration-development" method, the initial population is iteratively updated to obtain controller deployment candidate schemes. The Levy flight algorithm's optimization strategy is used to optimize the controller deployment candidate schemes to obtain the final controller deployment scheme.

[0010] Preferably, the step of clustering satellite nodes to determine initial cluster centers, and then updating the initial cluster centers to obtain the final cluster subdomains, includes:

[0011] The DPC algorithm is used to cluster satellite nodes and generate initial cluster centers;

[0012] The k-means algorithm is used to update the initial cluster centers, satellite nodes are assigned to the nearest cluster centers and the calculation is iterated until the preset conditions are met to obtain the final cluster subdomains.

[0013] Preferably, the step of dividing satellite nodes into core nodes and non-core nodes based on the minimum distance between the satellite node and the clustering subdomain includes:

[0014] Within each cluster subdomain, calculate the density center point between each satellite node and each cluster subdomain, and use the minimum distance as the minimum distance between the satellite node and the cluster subdomain.

[0015] The minimum distances of all satellite nodes are sorted, and the satellite nodes are divided into core nodes and non-core nodes according to a set ratio during the sorting.

[0016] Preferably, the method for allocating core nodes using the load balancing strategy is as follows:

[0017] Get the load rate of the core node. When the load rate reaches the set threshold, distribute the load of the core node to the nearest controller node, and the load of the controller node is less than the maximum load.

[0018] Preferably, the method for allocating non-core nodes using the load balancing strategy is as follows:

[0019] When the load rate of non-core nodes reaches the set threshold;

[0020] Determine the load rate of the nearest controller node. If the load of a controller node is greater than the average load, remove that nearest controller node. Repeat this process to assign non-core nodes to the nearest controller node, provided that the load of that nearest controller node is less than the average load.

[0021] Preferably, the step of generating the initial population based on the load-balanced multi-controller node subdomain and combined with the three-dimensional trigonometric function chaotic system includes:

[0022] Based on the load balancing multi-controller node subdomain and combined with the three-dimensional trigonometric function chaotic system to generate a chaotic sequence, the chaotic sequence is mapped to the controller deployment search space to generate the initial population using its ergodicity.

[0023] Preferably, the "exploration-development" method based on the snow ablation algorithm iteratively updates the initial population to obtain candidate controller deployment schemes, including:

[0024] The "exploration-development" method based on the snow melting algorithm calculates the current optimal solution for the initial population. The population is then guided to conduct a local fine-grained search around the current optimal solution using a day-by-day snow melting model to obtain the current deployment scheme. The population position is updated according to the optimal solution, and the "exploration-development" method is repeated until the change in population position is less than a preset threshold or the maximum number of iterations is reached, thus obtaining the controller deployment candidate scheme.

[0025] A collaborative dynamic improvement load balancing system based on SAGIN network, comprising:

[0026] The clustering module is used to cluster satellite nodes to determine initial cluster centers, and then update the initial cluster centers to obtain the final cluster subdomains.

[0027] The load balancing module is used to divide satellite nodes into core nodes and non-core nodes based on the minimum distance between the satellite nodes and the clustered subdomains; the clustered subdomains are used as controller nodes, and in combination with the load balancing strategy, the core nodes and non-core nodes are assigned to different controller nodes to obtain a load-balanced multi-controller node subdomain.

[0028] The deployment module is used to generate an initial population based on the load-balanced multi-controller node subdomain and the three-dimensional trigonometric function chaotic system. Based on the snow melting algorithm's "exploration-development" method, the initial population is iteratively updated to obtain controller deployment candidate schemes. The Levy flight algorithm is used to optimize the controller deployment candidate schemes to obtain the final controller deployment scheme.

[0029] An electronic device, comprising:

[0030] Memory, used to store computer programs;

[0031] A processor, used to implement the steps of the collaborative dynamic improved load balancing method based on the SAGIN network as described, when executing the computer program.

[0032] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the cooperative dynamic improved load balancing method based on the SAGIN network.

[0033] Compared with the prior art, the present invention has the following beneficial technical effects:

[0034] This application presents a collaborative dynamic load balancing method based on SAGIN networks. Addressing the controller load imbalance caused by dynamic changes in SAGIN network topology and uneven user service demands, this method employs a three-step collaborative design: "satellite node clustering - node classification and load allocation - controller deployment optimization," resulting in significant technical advantages. First, initial centers are determined through clustering and then updated to obtain cluster subdomains. This overcomes the limitations of traditional random center selection or fixed threshold partitioning, accurately adapting to the dynamic topology of the satellite network and laying a reasonable subdomain foundation for subsequent load balancing. Second, core and non-core nodes are classified based on the minimum distance between nodes and subdomains, and distributed to controllers using differentiated load balancing strategies. This avoids local overload or inefficient allocation caused by a single allocation rule, effectively balancing the load of each controller. Finally, a uniform initial population is generated using a three-dimensional trigonometric function chaotic system. Iterative optimization using the snowmelt algorithm ("exploration-development") followed by the Levy fly-through algorithm ensures both global search breadth and local search accuracy, breaking through local optima and ultimately achieving efficient controller deployment. The overall solution forms a closed loop from subdomain partitioning and load distribution to deployment optimization, effectively solving the problems of controller load imbalance, high network latency and uneven resource allocation in existing technologies, and adapting to the all-time and all-domain communication requirements of SAGIN network.

[0035] This application also proposes a collaborative dynamic improvement load balancing system based on SAGIN networks, an electronic device, and a computer storage medium, which possess all the advantages of the aforementioned collaborative dynamic improvement load balancing method based on SAGIN networks. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1This is a flowchart of the dynamic deployment process of an air-space-ground network controller based on the DPC-K-means and ISAO algorithms of the present invention.

[0038] Figure 2 This is a load index diagram for a SAGIN network-based collaborative dynamic improved load balancing method with 431 nodes.

[0039] Figure 3 This is a load index diagram for the number of nodes of the collaborative dynamic improved load balancing method based on SAGIN network of this invention, with 811 nodes.

[0040] Figure 4 This is a load index diagram for the SAGIN network-based collaborative dynamic improved load balancing method of the present invention with 1613 nodes.

[0041] Figure 5 This is a comparison chart of the average time delay under different controllers for the collaborative dynamic improved load balancing method based on SAGIN network of this invention;

[0042] Figure 6 This is a comparison chart of the average disconnection rate under different controllers for the collaborative dynamic improved load balancing method based on SAGIN network of this invention;

[0043] Figure 7 This is a comparison of the objective function values ​​of different algorithms in the multi-time slice of the collaborative dynamic improved load balancing method based on SAGIN network of this invention;

[0044] Figure 8 This is a comparison chart of the average latency of different algorithms in the multi-time slice of the collaborative dynamic improved load balancing method based on SAGIN network of this invention;

[0045] Figure 9 This is a graph showing the relationship between the computation time and the number of controllers for the collaborative dynamic improved load balancing method based on SAGIN network according to the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, 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 some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0047] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0048] A collaborative dynamic improvement load balancing method based on SAGIN networks includes the following steps:

[0049] Step 1: Cluster the satellite nodes to determine the initial cluster centers, and then update the initial cluster centers to obtain the final cluster subdomains;

[0050] First, the DPC (Density Peak Clustering) algorithm is used to calculate the local density (the density of surrounding nodes) and relative distance (the nearest distance to a node with a higher density) of each satellite node. High-density, relatively isolated points are then selected as initial cluster centers. However, the initial centers selected by DPC may not be accurate enough, so the K-means algorithm is used for further optimization: each satellite node is assigned to its nearest initial cluster center, and then the new centers for each group are recalculated. This process is repeated until the center positions no longer change significantly or a preset number of iterations is reached, ultimately determining stable "groups" (cluster subdomains).

[0051] On the one hand, it solves the problem of "unstable grouping" caused by the random selection of initial centers in traditional clustering methods (such as directly using K-means). DPC first selects promising initial centers and then uses K-means for optimization, avoiding large differences in clustering results each time, which is particularly suitable for the dynamic changes in satellite network topology. On the other hand, through precise clustering, it divides the vast number of satellite nodes into multiple subdomains.

[0052] Step 2: Based on the minimum distance between the satellite nodes and the clustered subdomains, the satellite nodes are divided into core nodes and non-core nodes. The clustered subdomains are used as controller nodes, and in combination with the load balancing strategy, the core nodes and non-core nodes are assigned to different controller nodes to obtain a load-balanced multi-controller node subdomain.

[0053] This step calculates the minimum distance from each satellite node to the density center of its respective cluster subdomain. The closer the distance, the stronger the connection between the node and the subdomain core. Then, nodes are sorted by distance, and the closest nodes are selected as "core nodes" (these nodes tend to concentrate the load), while the rest are "non-core nodes" (the load distribution is more flexible). When assigning controllers, each cluster subdomain is treated as a controller node, and different strategies are formulated for the two types of nodes: For core nodes, their load rate is monitored in real time. Once a set threshold is reached, their load is assigned to the nearest controller that has not exceeded its maximum load, ensuring that the load on core nodes does not accumulate. For non-core nodes, the load rate is monitored in the same way. If the load exceeds the threshold, controllers with loads exceeding the average level are first excluded, and then the nearest controller is selected from the remaining controllers for assignment, avoiding adding load to already busy controllers.

[0054] By categorizing nodes as "core" or "non-core," the system precisely addresses the load characteristics of different nodes, avoiding localized overload caused by a "one-size-fits-all" approach. Core nodes are allocated based on proximity, ensuring efficient data transmission; non-core nodes are prioritized for low-load controllers, guaranteeing overall balance. In practice, this prevents situations where one controller is overloaded and another is idle, improving overall network response speed and reducing communication delays or interruptions caused by uneven load distribution.

[0055] Step 3: Based on the load balancing multi-controller node subdomain and combined with the three-dimensional trigonometric function chaotic system, an initial population is generated. Based on the snow melting algorithm's "exploration-development" method, the initial population is iteratively updated to obtain controller deployment candidate schemes. The Levy flight algorithm's optimization strategy is used to optimize the controller deployment candidate schemes to obtain the final controller deployment scheme.

[0056] This step combines a three-dimensional trigonometric function chaotic system to generate an initial "deployment scheme set" (initial population). The characteristic of chaotic systems is that they allow these initial schemes to be distributed more evenly across the entire possible deployment space, preventing them from clustering in a small area initially, thus broadening the scope for finding the optimal solution later. Next, the "exploration-development" approach of the snow melting algorithm is used to optimize these initial schemes: the "exploration" phase is like snow turning into steam and randomly spreading, allowing schemes to find potentially good locations over a large area; the "development" phase is like the snow melting process, refining the search within a small area around the better-found schemes, making the schemes more precise. This process is repeated until the changes in the schemes are minimal or the maximum number of iterations is reached, resulting in a batch of candidate deployment schemes. Finally, the Levy fly-through algorithm's optimization strategy is used for a final screening—Levy fly-through occasionally involves long-distance "jumps," helping to escape previously trapped local optima (e.g., a solution that appears best in a certain region, but there are better solutions in the entire space), identifying the truly optimal final controller deployment scheme from the candidate schemes.

[0057] The uniform distribution of the initial population prevents the algorithm from being limited to finding solutions in a small area from the outset, increasing the probability of finding the global optimum. Then, the "exploration-development" combination of the snow melting algorithm ensures both the breadth and accuracy of the search, resulting in higher quality candidate solutions. Finally, the "jump" optimization of Levy flight solves the problem of traditional algorithms easily getting trapped in local optima. The resulting controller deployment scheme can further reduce network latency while ensuring load balancing, and the overall efficiency of the algorithm is also high. Even with a large number of satellite nodes, it can quickly calculate the optimal deployment scheme, adapting to the dynamic changes in the SAGIN network.

[0058] Example 1

[0059] like Figure 1 As shown, a collaborative dynamic improvement load balancing method based on SAGIN networks includes the following steps:

[0060] Step 1: Determine the initial cluster centers for the satellite network nodes using the DPC algorithm, and then optimize the initial cluster centers using the K-means algorithm to obtain the final clustering results and the center nodes of each cluster.

[0061] This step first uses the DPC algorithm to calculate the local density (ρ) and relative distance (δ) of each satellite node, and then combines it with the K-nearest neighbor algorithm to help determine the cluster centers and boundaries, thereby optimizing the evaluation of local density and relative distance, improving clustering accuracy, and avoiding clustering oscillations caused by random initialization. The specific steps are as follows:

[0062] S1.1, DPC is a density-based unsupervised clustering method, with its core being the definition of local density (ρ) and relative distance (δ). Local density is calculated by using the cutoff distance (dc) to count the number of data points surrounding a given point, which is crucial for selecting density peaks. An improperly defined local density can lead to incorrect density peaks, thus affecting the clustering results. Therefore, local density should focus more on the relative density of a sample with its neighbors to more accurately reflect whether a sample can become a cluster center.

[0063] In this embodiment, the local density is calculated using the cutoff distance (ρ) to count the number of data points around a given point, specifically using the following formula:

[0064]

[0065] in, Let be the Euclidean distance between satellite node i and satellite node j, k be the number of nearest neighbors, knn(i) be the set of the k nearest neighbors of satellite node i, and 2·k be ρ. i The normalized parameters. A metric that measures how isolated point i is in the dataset.

[0066] When the outlier degree of satellite node i is high, it indicates that the satellite node is relatively sparse in its surrounding area, that is, its local density is low. Local density is an indicator that measures the sparseness of the area surrounding satellite node i. The higher the value, the sparser the area.

[0067] The advantage of using the above formula to calculate local density is that it reflects the relative density of a sample within its local range. This method can enhance the local density of samples in sparse clusters while reducing the local density of samples in dense clusters, thereby reducing the impact of differences in data point density between different clusters on the clustering results.

[0068] The relative distance (δ) refers to the distance from satellite node i to its nearest and densest satellite node. Before determining the relative distance of satellite node i, the local density of all data points needs to be sorted. For the sample point with the highest local density, its relative distance is defined as infinity, or a sufficiently large value to indicate that it is the point with the highest density. This definition will help identify the true density peak points in the density peak clustering algorithm. The definition of the relative distance (δ) for the point with the highest sample density is shown in Equation (2):

[0069]

[0070] The definition of the relative distance (δ) between other satellite nodes is shown in formula (3):

[0071]

[0072] In DPC, cluster centers are determined by a decision value γ, which is typically represented by the product of local density (ρ) and relative distance (δ), i.e., γ. i =ρ i ·δ i By calculating the γ value of each satellite node, the DPC algorithm is able to find all points with high γ values ​​in the decision map and identify them as potential density peaks, thereby determining the cluster centers in the data.

[0073] S1.2. The k-means algorithm is used to update the initial cluster centers, and satellite nodes are assigned to the nearest cluster centers and the calculation is iterated until the preset conditions are met to obtain the final cluster subdomains, ensuring that the final subdomain partitioning results are stable and reasonable.

[0074] The clustering steps for this satellite node are as follows:

[0075] Input: Dataset G (containing satellite coordinates X, Y, Z and communication load parameters, status information), number of clusters K, maximum number of iterations;

[0076] Step 1: Initialize using the DPC algorithm:

[0077] An improved DPC algorithm is used to determine the initial cluster centers. The improved DPC algorithm incorporates the k-nearest neighbor algorithm to generate better initial cluster centers. The algorithm outputs a set containing multiple cluster centers as the initial center points for the k-means algorithm.

[0078] Step 2: Fitness Function Evaluation

[0079] A fitness function is defined to evaluate the position of each cluster center in the solution space. This fitness function considers factors such as the distance between cluster centers and the compactness within clusters to ensure that the selection of cluster centers maximizes the clustering effect, as shown in Equation (4):

[0080]

[0081] Where x is a data point, c i It is the i-th cluster center, d is the dimension of the data, and x is the dimensionality of the data. j and c ij They are x and c respectively i The value in the j-th dimension.

[0082] Step 3: k-means clustering update:

[0083] Using the initial cluster centers determined by the DPC algorithm, the k-means algorithm is executed. In each iteration, samples are assigned to the nearest cluster centers, and the cluster centers are recalculated as shown in Equation (5):

[0084]

[0085] Among them, S i It is the set of data points in the i-th cluster, |S i | represents the number of data points in the set.

[0086] Step 4: Iterate until convergence:

[0087] Repeat the above process until the cluster centers no longer change significantly, or until the preset number of iterations is reached, at which point the algorithm converges.

[0088] Step 5: Output the optimal clustering result:

[0089] The final output is the cluster to which each sample belongs and the cluster center of each cluster.

[0090] Step 2: Based on the minimum distance between the satellite nodes and the clustered subdomains, the satellite nodes are divided into core nodes and non-core nodes. The clustered subdomains are used as controller nodes, and in combination with the load balancing strategy, the core nodes and non-core nodes are assigned to different controller nodes to obtain a load-balanced multi-controller node subdomain.

[0091] A flexible buffer technology is employed for satellite controller load balancing. Load allocation is based on the controller's load status, using a load balancing mechanism to distribute the load of satellite nodes to controller nodes with lower loads, ensuring even load distribution. To prevent satellite controller load imbalance at density center nodes, which could lead to increased communication latency or interruptions, a satellite controller load allocation strategy based on flexible buffers is proposed, taking into account the satellite controller load status. This strategy achieves balanced load distribution while ensuring reasonable partitioning. The specific implementation steps are as follows:

[0092] S2.1 Within each cluster subdomain, calculate the minimum distance between each satellite node and the density center of the cluster subdomain. The calculation formula is as follows:

[0093] The minimum distance from the satellite node to the density center point is shown in formula (6):

[0094] d rep (i)=min c∈C d(i,c) (6)

[0095] For each satellite node, calculate its distance d(i,c) to all density center points C, and then take the minimum value as the representative distance d of that satellite node. rep (i):

[0096] Sort all satellite nodes by minimum distance and classify the satellite nodes according to the sorting results;

[0097] The satellite nodes are sorted and classified as follows:

[0098] Core={i|d rep (i)≤percentile(d rep (allsamples),p)} (7)

[0099] Non-Core={i|d rep (i)≥percentile(d rep (allsamples),p)} (8)

[0100] All satellite nodes are arranged according to their representative distance d rep (i) Sort in ascending order. Define a percentage p, and define the top p% of satellite nodes after sorting as core nodes, with the rest as non-core nodes:

[0101] The correlation between satellite nodes and the density center of the subdomain is calculated (minimum distance), and satellite nodes are divided into "core nodes" (closely correlated with the subdomain center, and the load is easily concentrated) and "non-core nodes" (low correlation, and the load distribution is more flexible).

[0102] S2.2. Using the clustered subdomains as controller nodes, and combining them with load balancing strategies, core nodes and non-core nodes are assigned to different controller nodes to obtain load balancing multi-controller node subdomains.

[0103] Obtain the load rate of the core node. When the load rate reaches a set threshold, distribute the load of the core node to the controller node with a lower load and update the load rate of that controller node. The update method is as follows:

[0104] L(c)=L(c)+1 if i∈c and L(c)≤L max (c) (9)

[0105] The load L(c) can be defined as the number of samples allocated to the center point C. For each core node, it is directly allocated to the nearest density center point and cannot exceed the maximum load, and the load status of that density center point is updated.

[0106] S2.3. Obtain the load rate of non-core nodes, and when the load rate reaches the set threshold;

[0107] Determine the load rate of the nearest controller node. If the load of a controller node is greater than the average load, remove that nearest controller node. Repeat this process to assign non-core nodes to the nearest controller node, provided that the load of that nearest controller node is less than the average load.

[0108] The formula for calculating average load is as follows:

[0109]

[0110] For each non-core point, first check the load of all controller nodes. If the load of a controller node is greater than the average load L... avg If the current controller node fails to meet the requirements, it will be removed from the candidate allocation pool. Then, the nearest controller node will be selected from the remaining candidate allocation pool for allocation, and the load information will be updated.

[0111] This step divides satellite nodes into "core nodes" (closely associated with subdomain centers and prone to load concentration) and "non-core nodes" (lowly associated and more flexible load distribution) by calculating the degree of association (minimum distance) between satellite nodes and the density centers of cluster subdomains.

[0112] Different allocation strategies are formulated for the different load aggregation risks of core nodes and non-core nodes. Core nodes are prioritized for allocation based on proximity (to ensure efficiency), while non-core nodes are first selected based on low-load controllers (to ensure balanced load distribution), thus avoiding "local overload" or "inefficient allocation" caused by a single rule.

[0113] Core node load allocation: Implement the acquisition of the load rate of core nodes. When the load rate reaches the threshold, allocate the load of the core node to the "nearest density center point" without exceeding the maximum load, and update the load rate of the controller nodes with lower load in real time.

[0114] Non-core node load distribution: Implement load rate acquisition for non-core nodes, remove the center point whose load exceeds the average load, and distribute the load of non-core nodes whose load rate exceeds the threshold to the "nearest center point".

[0115] Step 3: Based on the load balancing multi-controller node subdomain and combined with the three-dimensional trigonometric function chaotic system to generate chaotic sequences, generate an initial population. Based on the snow melting algorithm's "exploration-development" method, iteratively update the initial population to obtain controller deployment candidate schemes. Use the Levy flight algorithm's optimization strategy to optimize the controller deployment candidate schemes to obtain the final controller deployment scheme.

[0116] The ISAO algorithm initializes the population using a complex chaotic system and optimizes the deployment locations of controller nodes using the Levy flight strategy. The goal of this step is to enhance global search capabilities, avoid getting trapped in local optima, and thus improve the accuracy and efficiency of controller allocation.

[0117] S3.1. Based on the load balancing multi-controller node subdomain and combined with the three-dimensional trigonometric function chaotic system, a chaotic sequence is generated. The sequence is then mapped to the controller deployment search space (including X / Y / Z coordinates, load carrying threshold, etc.) using its ergodicity to generate an initial population. This ensures that the initial solution is evenly distributed within the search domain, thus completing the population initialization.

[0118] A composite chaotic system is used to initialize the population. The chaotic mapping method can better distribute the population evenly in the solution domain, improve diversity, and thus mitigate the problem of premature convergence. Cubic mapping has stronger ergodicity than logistic mapping and produces a more uniform chaotic sequence. Therefore, a chaotic sequence generated by a three-dimensional trigonometric function chaotic system is used to ensure the uniform distribution of the population in the search space and avoid premature convergence.

[0119] S3.2. The "exploration-development" method based on the snow melting algorithm calculates the current optimal solution of the initial population. The population is guided to conduct a local fine-grained search around the current optimal solution through the daily snow melting model to obtain the current deployment scheme. The population position is updated according to the optimal solution and the "exploration-development" method is repeated until the change in population position is less than the preset threshold or the maximum number of iterations is reached to obtain the controller deployment candidate scheme.

[0120] In the exploration phase, the Snow Ablation Algorithm (SAO) utilizes Brownian motion to simulate the irregular motion of snow or liquid water converted from snow as it transforms into vapor, resulting in highly dispersed search instances and thus exploring potential regions in the search space. The mathematical expressions for this phase are shown in (13-15):

[0121]

[0122] Elite(t)∈[G(t),Z second (t),Z third (t),Z c (t)] (15)

[0123] Where G(t) is the current optimal solution, Elite(t) represents an individual randomly selected from several elite groups in the population, and BM i (t) represents Brownian motion, and θ1 is an adjustment factor. It is the center of mass of the group. In the formula, Z second (t), Z third (t) represents the positions of the second and third best particles; Z c (t) represents the centroid positions of the top 50% of particles by fitness value. This formula describes the one-dimensional distribution of Brownian motion and can be extended to two-dimensional and three-dimensional search spaces to simulate the diffusion of vapor in the search space. In satellite controller allocation, this corresponds to exploring different controller allocation schemes to find the optimal load balancing state.

[0124] During the development phase, search entities are encouraged to develop high-quality solutions around the current best solution, rather than expanding highly dispersed features in the solution space. As one of the most classic snowmelt models, the day-to-day method is used to reflect the snowmelt process, as shown in expression (16):

[0125] M=DDF×(T(t)-T1(t)) (16)

[0126] Where M is the snowmelt rate, a key parameter for simulating snowmelt behavior during the development phase. T(t) represents the daily average temperature. T1(t) is the base temperature, which is usually set to 0.

[0127] M=DDF×T(t) (17)

[0128] The definition of DDF is shown in formula (18):

[0129]

[0130] The position update formula for this stage is shown in formula (19):

[0131]

[0132] Here, θ2 is a random number with a control factor of [-1, 1], used to balance the global and local searches. The SAO algorithm employs a global position update equation, combining exploration and exploitation strategies with population dynamics to optimize the position of each individual. In satellite controller allocation, this means that the controller's position and task allocation can be dynamically adjusted based on the current load and mission requirements to achieve load balancing.

[0133] S3.3. The Levy flight algorithm is used to optimize the candidate controller deployment schemes to obtain the final controller deployment scheme.

[0134] To improve the overall performance of SAO in solving the master-slave allocation problem of satellite controller, the snow melting algorithm (SAO) is optimized and incorporated into the levy-based optimization strategy, as shown in formula (20).

[0135]

[0136] The parameters are represented as shown in formulas (21-23):

[0137]

[0138] G1 = 2 × rand-1 (22)

[0139]

[0140] Among them, X i (t+1) represents the population position after the shrinking development phase update, Q F This is the quality function value used to balance the search strategy, T is the maximum number of iterations, G1 and G2 are parameters that control the search behavior, rand() is a random number in the range [0,1], t is the current iteration number, is the average position of the current population, and Levy(D) represents the Levy flight distribution function, used to simulate the irregular motion of snow or liquid water when it is converted into steam.

[0141] This optimization enhances SAO's local search capabilities, improving the efficiency and accuracy of finding better solutions, helping the algorithm avoid getting trapped in local optima, balancing the needs of exploration and development, enhancing the algorithm's robustness, and improving the efficiency of the optimization process. In the context of satellite controller master-slave allocation, optimizing the controller allocation strategy can improve resource utilization, reduce latency, and enhance the stability and communication efficiency of the entire satellite network.

[0142] This step generates a chaotic sequence using a three-dimensional trigonometric function chaotic system based on the subdomain of the load-balanced multi-controller node, and maps the chaotic sequence to the controller deployment search space to generate an initial population.

[0143] In the exploration phase, the population position is updated based on the Brownian motion model, and multiple sets of candidate controller deployment schemes are generated. In the development phase, the snow melting rate M is calculated based on the daily snow melting model. The population is made to carry out local fine-grained search around the current optimal solution through position updates to optimize the accuracy of the deployment scheme. Then the exploration and development phases are repeated until the population position change is less than the preset threshold or the maximum number of iterations is reached, and a set of deployment schemes is obtained.

[0144] Finally, based on the iterative population, the improved position update formula is substituted into the formula, and the Levy flight algorithm is used to realize random jump updates of the population position. At the same time, the fitness is evaluated based on "controller load balancing degree + deployment latency", and the population with the best fitness after the update is selected. In this way, the local optimum is broken and the final controller deployment scheme is determined.

[0145] The collaborative dynamic improvement load balancing method based on SAGIN networks proposed in this application involves step 1, which calculates the local density and relative distance of nodes using the DPC algorithm, and identifies cluster centers based on density peaks, thereby achieving efficient subdomain partitioning. The principle is to select nodes with higher density as cluster centers by calculating the density and relative distance between nodes, ensuring cluster stability. Combining the DPC algorithm with the K-nearest neighbor algorithm effectively reduces cluster oscillations caused by random initialization. Its advantage lies in its ability to adapt to complex satellite network topologies and improve network load balancing performance.

[0146] Step 2 involves load balancing of satellite controllers. This is achieved by calculating the load index of each controller and allocating the load to controller nodes with lower loads according to load balancing principles. The advantage of this method is that it effectively avoids excessive controller load, reduces network communication latency and disconnection rates, thereby improving network response speed and reliability.

[0147] Step 3 optimizes controller allocation using an improved ISAO algorithm. The principle is to initialize the population using a complex chaotic system and combine it with the Levy flight strategy for global search and local optimization, preventing premature convergence and improving the efficiency and accuracy of controller deployment. The advantage of this method is its ability to find the global optimum in a complex search space, enhancing the algorithm's adaptability and stability.

[0148] The following simulation examples further illustrate this embodiment.

[0149] The experiment conducted simulation tests on low-Earth orbit (LEO) satellite constellations of different sizes and comprehensively evaluated the performance of the deployment strategy by comparing and analyzing load balancing indicators under different scales. To ensure the accuracy and reliability of the results, the experiment adopted the method of averaging multiple simulations and compared with other algorithms, proving the effectiveness of the proposed strategy under different constellation sizes and providing strong theoretical support for the optimized deployment of LEO satellite constellations.

[0150] from Figure 2 , Figure 3 and Figure 4 The subdomain partitioning results shown for different numbers of nodes indicate that the load balancing index of the proposed algorithm steadily decreases with the increase in the number of controllers, reaching the minimum. In contrast, the load balancing indices calculated by FCM and K-means fluctuate significantly and are generally higher than those of the proposed algorithm. Under three different scenarios, the standard deviation of the average load index of the proposed algorithm is 0.0201, while the standard deviations of the average load indices of FCM and K-means algorithms are 0.0273 and 0.03185, respectively. The proposed algorithm reduces the load by an average of 26.8% and 36.9% compared to FCM and K-means algorithms, respectively. Among the three subdomain partitioning strategies without overload, the OCLDS, FCM, and K-means algorithms are highly sensitive to the selection of the initial centroid, meaning different initial values ​​can significantly impact the final result, reducing the algorithm's stability. The algorithm presented in this paper uses the cluster centers determined by the DPC algorithm as the initial center points for the K-Means algorithm, effectively reducing the randomness of the K-Means algorithm. Based on the above results, a correction is made and combined with a load-balancing-based buffer allocation strategy for the final subdomain partitioning, enhancing the stability and accuracy of the clustering algorithm. Therefore, the algorithm proposed in this invention can reasonably partition subdomains according to network node information, thereby significantly reducing the load index and improving the load balancing performance of the controller.

[0151] The performance of the controller allocation strategy was validated by comparing the FCM, OCLDS, and K-means algorithms. Simulation experiments were conducted on over 800 satellite network nodes, and the average propagation delay T was evaluated under different numbers of controllers.avg and average disconnection rate σ avg .

[0152] from Figure 5 and Figure 6 As can be seen, the controller allocation strategy proposed in this invention exhibits superior performance compared to FCM, OCLDS, and K-means algorithms in 100 experiments, with an overall average improvement in latency and disconnection rate of 16.3% and 7%, respectively. This improvement is attributed to the enhanced stability and accuracy of the algorithm, particularly the reduction of randomness in the K-Means algorithm by using cluster centers determined through the DPC algorithm, thus optimizing the subdomain partitioning of network nodes. Furthermore, this invention incorporates a complex chaotic system as the initialization method and combines it with an optimization strategy based on the Levy flight strategy, enhancing the optimization capability and resulting in better flexibility and adaptability of the proposed algorithm.

[0153] To evaluate the performance improvement capabilities of the proposed algorithm in satellite communication, we conducted experiments with different scenarios and algorithm parameters based on subdomain partitioning. We compared and analyzed the performance of OCLDS, FCM, K-means algorithms, and the proposed algorithm in subdomain partitioning and controller deployment optimization. The experiments were conducted in a hybrid constellation 2 satellite network environment containing 811 nodes, covering 200 time slices within one cycle. Evaluation metrics included the objective function value, average latency, and algorithm execution efficiency.

[0154] from Figure 7 and Figure 8 As can be seen, under different time slices, the objective function value and average propagation delay fluctuate within a certain range as the number of satellite nodes changes. Compared with OCLDS, FCM, and K-means algorithms, the algorithm proposed in this invention achieves better results in terms of objective function and average propagation delay, with average performance improvements of 47% and 28.5%, respectively. This performance advantage is attributed to the fact that the algorithm comprehensively considers multiple parameters of nodes when partitioning subdomains and selects more suitable subdomain interaction center nodes through optimization algorithms, enhancing the adaptability of the algorithm. Furthermore, the standard deviation of the objective function of the algorithm proposed in this invention is small, demonstrating good stability. Therefore, the bottom-up, hierarchical optimization strategy proposed in this invention achieves better optimization of network performance.

[0155] from Figure 9It can be seen that the computation time of the four algorithms is directly proportional to the increase in the number of controllers, and the algorithm proposed in this paper has the shortest computation time overall. This is because the algorithm proposed in this invention can efficiently complete the deployment planning of satellite network nodes with an increase in the number of controllers. It can be seen that the number of controller nodes in the four algorithms is directly proportional to the computation time. The algorithm used in this method improves the efficiency by an average of 15%, 42%, and 22% compared to OCLDS, FCM, and K-means algorithms, respectively. The FCM algorithm, due to its fuzzy theory, requires iterative calculations, resulting in high time complexity. Although the K-means algorithm is simple and easy to implement, its convergence speed is slow for large-scale datasets, especially when the initial cluster center selection is poor, which may require more iterations to reach a stable state. The algorithm proposed in this invention is more time efficient than other comparative algorithms, possibly because it effectively improves the computation efficiency of multi-controller allocation for satellite nodes by optimizing the initial cluster center selection, reducing algorithm complexity, enhancing adaptability and flexibility, adopting an improved SAO algorithm optimization strategy, reducing sensitivity to initial conditions, and accelerating global optimization capabilities.

[0156] In this embodiment, a low-Earth orbit satellite constellation with 431 nodes was set up for the experiment. The DPC-K-means algorithm was used for subdomain partitioning, and the results were compared with the FCM and K-means algorithms. Simulation results show that as the number of controllers increases, the load balancing index of the proposed algorithm shows a stable decreasing trend, and the load balancing index is the smallest.

[0157] In this embodiment, a low-Earth orbit satellite constellation of 811 nodes was used for simulation. The DPC-K-means algorithm was employed for subdomain partitioning, and the results were compared with those of the FCM and K-means algorithms. Simulation results show that the load balancing index continuously decreases as the number of controllers increases, and the minimum value is better than the other two algorithms.

[0158] In this embodiment, a low-Earth orbit satellite constellation with 1613 nodes was simulated. The DPC-K-means algorithm was used for subdomain partitioning, and the results were compared with the FCM and K-means algorithms. Experimental results show that the proposed algorithm effectively reduces the imbalance of controller load in subdomain partitioning, and the load balancing index and standard deviation are significantly optimized compared to the comparison algorithms. The standard deviation of the proposed algorithm is reduced by 26.8% and 36.9% compared to the FCM and K-means algorithms, respectively, further demonstrating that the DPC-K-means algorithm can effectively guarantee load balancing in larger-scale satellite networks.

[0159] In this embodiment, the experiment was conducted in an integrated air-space-ground network containing over 800 satellite network nodes, and the DPC-K-means algorithm was used to optimize the controller allocation strategy. Compared with FCM, OCLDS, and K-means algorithms, simulation results show that the proposed algorithm improved the overall average latency and disconnection rate by 16.3% and 7%, respectively, in 100 experiments. This is attributed to the fact that the cluster centers determined by the DPC algorithm reduce the randomness of the K-means algorithm and optimize the subdomain partitioning of network nodes. This embodiment demonstrates that the proposed algorithm exhibits high performance in controller allocation strategies, improving the stability and accuracy of the network.

[0160] In this embodiment, the experiment was conducted in a hybrid constellation satellite network environment containing 811 nodes. The DPC-K-means and ISAO algorithms were used for subdomain partitioning and controller deployment optimization. The experiment verified the proposed algorithm's ability to improve satellite communication performance by setting different scenarios and algorithm parameters. Experimental results show that compared with OCLDS, FCM, and K-means algorithms, the proposed algorithm outperforms the OCLDS, FCM, and K-means algorithms in different time slices, improving the objective function value and average propagation delay by 47% and 28.5%, respectively. By optimizing node information and selecting appropriate subdomain interaction center nodes, the proposed algorithm demonstrates good adaptability and stability in controller allocation.

[0161] Example 2

[0162] A collaborative dynamic improvement load balancing system based on SAGIN network, comprising:

[0163] The clustering module is used to cluster satellite nodes to determine initial cluster centers, and then update the initial cluster centers to obtain the final cluster subdomains.

[0164] The load balancing module is used to divide satellite nodes into core nodes and non-core nodes based on the minimum distance between the satellite nodes and the clustered subdomains; the clustered subdomains are used as controller nodes, and in combination with the load balancing strategy, the core nodes and non-core nodes are assigned to different controller nodes to obtain a load-balanced multi-controller node subdomain.

[0165] The deployment module is used to generate an initial population based on the load-balanced multi-controller node subdomain and the three-dimensional trigonometric function chaotic system. Based on the snow melting algorithm's "exploration-development" method, the initial population is iteratively updated to obtain controller deployment candidate schemes. The Levy flight algorithm is used to optimize the controller deployment candidate schemes to obtain the final controller deployment scheme.

[0166] It should be noted that, in the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another device, or some features may be ignored or not executed. The modules described as separate components may or may not be physically separated. The components shown as modules may be one or more physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs.

[0167] Furthermore, in the various embodiments of the present invention, the modules can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The integrated unit described above can be implemented in hardware or as a software functional unit.

[0168] An electronic device provided in this application includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the collaborative dynamic improvement load balancing method based on the SAGIN network as described in any of the above embodiments.

[0169] Another electronic device provided in this application embodiment may further include: an input port connected to a processor for transmitting multimodal data collected by an external acquisition device to the processor; a display unit connected to the processor for displaying the processor's processing results to the outside world; and a communication module connected to the processor for enabling communication between the electronic device and the outside world. The display unit may be a display panel, a laser scanning display, etc.; the communication method adopted by the communication module includes, but is not limited to, Mobile High Definition Link (HML), Universal Serial Bus (USB), High Definition Multimedia Interface (HDMI), and wireless connection (including Wi-Fi, Bluetooth, Bluetooth Low Energy, and IEEE 802.11s-based communication technology).

[0170] This application provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the steps of the collaborative dynamic improvement load balancing method based on the SAGIN network as described in any of the above embodiments.

[0171] For descriptions of relevant parts of the SAGIN-based collaborative dynamic improved load balancing system, electronic device, and computer-readable storage medium provided in this application's embodiments, please refer to the detailed descriptions of the corresponding parts in the SAGIN-based collaborative dynamic improved load balancing method provided in this application's embodiments; they will not be repeated here. Furthermore, parts of the technical solutions provided in this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0172] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A collaborative dynamic improvement load balancing method based on SAGIN networks, characterized in that, Includes the following steps: Step 1: Cluster the satellite nodes to determine the initial cluster centers, and then update the initial cluster centers to obtain the final cluster subdomains; Step 2: Based on the minimum distance between the satellite nodes and the clustered subdomains, the satellite nodes are divided into core nodes and non-core nodes. The clustered subdomains are used as controller nodes, and in combination with the load balancing strategy, the core nodes and non-core nodes are assigned to different controller nodes to obtain a load-balanced multi-controller node subdomain. Step 3: Based on the load balancing multi-controller node subdomain and combined with the three-dimensional trigonometric function chaotic system, an initial population is generated. Based on the snow melting algorithm's "exploration-development" method, the initial population is iteratively updated to obtain controller deployment candidate schemes. The Levy flight algorithm's optimization strategy is used to optimize the controller deployment candidate schemes to obtain the final controller deployment scheme.

2. The collaborative dynamic improved load balancing method based on SAGIN network according to claim 1, characterized in that, The process of clustering satellite nodes to determine initial cluster centers, and then updating these initial cluster centers to obtain the final cluster subdomains, includes: The DPC algorithm is used to cluster satellite nodes and generate initial cluster centers; The k-means algorithm is used to update the initial cluster centers, satellite nodes are assigned to the nearest cluster centers and the calculation is iterated until the preset conditions are met to obtain the final cluster subdomains.

3. The collaborative dynamic improvement load balancing method based on SAGIN network according to claim 1, characterized in that, The process of dividing satellite nodes into core nodes and non-core nodes based on the minimum distance between the satellite node and the clustering subdomain includes: Within each cluster subdomain, calculate the density center point between each satellite node and each cluster subdomain, and use the minimum distance as the minimum distance between the satellite node and the cluster subdomain. The minimum distances of all satellite nodes are sorted, and the satellite nodes are divided into core nodes and non-core nodes according to a set ratio during the sorting.

4. The collaborative dynamic improvement load balancing method based on SAGIN network according to claim 1, characterized in that, The method for allocating core nodes using a load balancing strategy is as follows: Get the load rate of the core node. When the load rate reaches the set threshold, distribute the load of the core node to the nearest controller node, and the load of the controller node is less than the maximum load.

5. The collaborative dynamic improvement load balancing method based on SAGIN network according to claim 1, characterized in that, The method for allocating non-core nodes using a load balancing strategy is as follows: When the load rate of non-core nodes reaches the set threshold; Determine the load rate of the nearest controller node. If the load of a controller node is greater than the average load, remove that nearest controller node. Repeat this process to assign non-core nodes to the nearest controller node, provided that the load of that nearest controller node is less than the average load.

6. The collaborative dynamic improvement load balancing method based on SAGIN network according to claim 1, characterized in that, The process of generating an initial population based on the load-balanced multi-controller node subdomain and combined with a three-dimensional trigonometric function chaotic system includes: Based on the load balancing multi-controller node subdomain and combined with the three-dimensional trigonometric function chaotic system to generate a chaotic sequence, the chaotic sequence is mapped to the controller deployment search space to generate the initial population using its ergodicity.

7. The collaborative dynamic improved load balancing method based on SAGIN network according to claim 1, characterized in that, The "exploration-development" method based on the snow ablation algorithm iteratively updates the initial population to obtain candidate controller deployment schemes, including: The "exploration-development" method based on the snow melting algorithm calculates the current optimal solution for the initial population. The population is then guided to conduct a local fine-grained search around the current optimal solution using a day-by-day snow melting model to obtain the current deployment scheme. The population position is updated according to the optimal solution, and the "exploration-development" method is repeated until the change in population position is less than a preset threshold or the maximum number of iterations is reached, thus obtaining the controller deployment candidate scheme.

8. A collaborative dynamic improved load balancing system based on SAGIN network, characterized in that, include: The clustering module is used to cluster satellite nodes to determine initial cluster centers, and then update the initial cluster centers to obtain the final cluster subdomains. The load balancing module is used to divide satellite nodes into core nodes and non-core nodes based on the minimum distance between the satellite nodes and the clustered subdomains; the clustered subdomains are used as controller nodes, and in combination with the load balancing strategy, the core nodes and non-core nodes are assigned to different controller nodes to obtain a load-balanced multi-controller node subdomain. The deployment module is used to generate an initial population based on the load-balanced multi-controller node subdomain and the three-dimensional trigonometric function chaotic system. Based on the snow melting algorithm's "exploration-development" method, the initial population is iteratively updated to obtain controller deployment candidate schemes. The Levy flight algorithm is used to optimize the controller deployment candidate schemes to obtain the final controller deployment scheme.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the collaborative dynamic improved load balancing method based on the SAGIN network as described in any one of claims 1-8 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the collaborative dynamic improved load balancing method based on the SAGIN network as described in any one of claims 1-8.

Images