Space-ground integrated load balancing method based on improved particle swarm and Hereum algorithm

Abstract

The invention discloses a space-ground integrated load balancing method based on an improved particle swarm and a Hereum algorithm, and the method comprises the steps: building a target optimization model and constraint conditions according to a space-ground integrated intelligent network structure; the method comprises the following steps of: representing an improved scheme of a space-ground integrated intelligent network structure by particles, forming a population by a particle set, and introducing adaptive meta-heuristic mapping; the target optimization model is solved, a selective reset mechanism is introduced, the defense behavior and the escape behavior of particles are simulated by adopting a He horse optimization algorithm, and better particles are selected as the optimal space-ground integrated intelligent network structure output; the method has better adaptability, robustness and expansibility, improves the efficiency, can respond to the change of the network state, and provides an efficient flow management scheme for the link load balancing problem in the space-ground integrated intelligent network.

Description

Technical Field

[0001] The present invention relates to the field of information engineering, and in particular to a ground-ground integrated load balancing method based on an improved particle swarm and hippopotamus algorithm. Background Art

[0002] As a comprehensive communication network system, the space-ground integrated network aims to achieve seamless coverage and information transmission on a global scale by integrating space-based satellite networks and ground network resources, providing users with the ability to access and provide services on demand, while ensuring the security and reliability of the network. However, with the surge in the number of Internet users and the explosive growth of data traffic, the space-ground integrated network faces unprecedented challenges, such as network congestion and data security issues, which seriously threaten the efficient operation of the network and the normal flow of data. In order to solve these problems and ensure the reasonable allocation and optimization of network resources, network load balancing algorithms have emerged. Load balancing algorithms are key technologies for improving network performance, reducing network load, and enhancing network service quality. In the current research on load balancing algorithms, heuristic algorithms have received widespread attention internationally due to their effectiveness in solving complex optimization problems.

[0003] Heuristic algorithms, especially genetic algorithms and ant colony algorithms, researchers have done a lot of work on these basic algorithms and developed a variety of variants and improved algorithms to improve the convergence speed of the algorithm and the quality of the solution. The existing technology integrates the heuristic algorithm into a memetic algorithm and uses it as an initialization method and mutation operator; some solutions improve and optimize the existing PSO load balancing algorithm by dynamically adjusting the parameters or mechanisms of the PSO algorithm, such as adaptive adjustment of the inertia mechanism, improved compression mechanism, dynamic neighborhood scheduling strategy, dual population cross learning, combined with the divide-and-conquer algorithm and traffic scheduling.

[0004] The above heuristic algorithms are suitable for solving complex optimization problems and are widely used. However, when solving problems, heuristic algorithms are prone to fall into local optimal solutions and premature convergence. The local optimal solution is the optimal solution in a certain area of ​​the search space, but it is not necessarily the global optimal solution. Heuristic algorithms rely on local information and are easily attracted to local optimal solutions. For example, poor initial population quality or conservative selection operations in genetic algorithms will lead to the loss of good genes and cause the algorithm to converge prematurely. Premature convergence means that the algorithm stops searching before finding the global optimal solution, usually due to improper parameter settings or invalid search strategies. For example, in particle swarm optimization, the inertia weight is too large, causing particles to move quickly and unable to search in detail, resulting in premature convergence. Insufficient population diversity will also cause the algorithm to lose the ability to explore new solution spaces, resulting in premature convergence. Summary of the invention

[0005] Purpose of the present invention: In order to address the problems of the existing particle swarm optimization algorithm, such as the tendency of particles to converge prematurely, and the lack of global exploration ability and the tendency to fall into local optimum in the standard Hippopotamus optimization algorithm, the present invention provides a ground-to-earth integrated load balancing method (PH-PSO) based on the improved particle swarm and Hippopotamus algorithm. In a ground-to-earth integrated network environment, this method realizes an innovative hybrid optimization method by integrating the advantages of the improved particle swarm optimization and the Hippopotamus optimization algorithm (HO). This hybrid optimization method not only improves the adaptability and robustness of the method, but also improves its efficiency and scalability, so that the PH-PSO method can intelligently respond to changes in network status, and provides an efficient traffic management solution for the link load balancing problem in the ground-to-earth integrated network.

[0006] To achieve the above functions, the present invention designs a space-ground integrated load balancing method based on an improved particle swarm and Hippo algorithm. For the space-ground integrated network, the following steps S1 to S3 are performed to complete the optimization of the space-ground integrated network structure: Step S1: Establish a network topology based on the integrated network structure of the ground and the sky, establish a target optimization model based on the multi-commodity flow characteristics of network traffic, link bandwidth constraints, and optimization requirements for traffic distribution, and define corresponding constraints for the target optimization model; Step S2: An improved scheme of using particles to represent the space-ground integrated network structure. According to the characteristics of the space-ground integrated network structure, the position and velocity of the particles are encoded. The particles are taken as individuals, and the collection of all particles constitutes a population. In the population initialization stage, an adaptive meta-heuristic mapping is introduced to increase the diversity of the population. Step S3: Solve the target optimization model, introduce a selective reset mechanism to reset the position of the particles, use the Hippo optimization algorithm to simulate the defense and escape behaviors of the particles, compare the new positions and fitness values ​​of the particles after the defense and escape behaviors, select better particles as the optimal ground-ground integrated network structure output, and complete the optimization of the ground-ground integrated network structure.

[0007] Beneficial effects: Compared with the prior art, the advantages of the present invention include: 1. The present invention adopts adaptive meta-heuristic mapping, selective reset mechanism and Hippo optimization algorithm; adaptive meta-heuristic mapping generates high-quality and diverse initial populations by selecting or combining different initialization methods; the selective reset mechanism enhances global search, allowing the algorithm to jump out of the local optimum and explore a wider solution space; the Hippo optimization algorithm increases diversity and local optimum sensitivity and refines the search process. Multiple mechanisms complement each other and jointly promote search and optimization effects; global search quickly locates high-quality solution areas, local search deeply explores solution details, achieves global and local balance, efficiently optimizes network traffic distribution and load balancing strategies, significantly improves network performance, and guarantees service QoS requirements; 2. The present invention makes the links in the integrated space-ground network tend to be balanced, avoiding link congestion. At the same time, the method can balance the traffic in the network and ensure the utilization rate of the links. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 is a flow chart of a ground-ground integrated load balancing method based on an improved particle swarm and hippopotamus algorithm provided according to an embodiment of the present invention; Figure 2 is a schematic diagram of a particle resetting mechanism provided according to an embodiment of the present invention; Figure 3 is a fitness convergence comparison diagram provided according to an embodiment of the present invention; Figure 4 is a comparison diagram of jitter conditions provided according to an embodiment of the present invention; Figure 5 It is a link load display diagram of a ground-ground integrated load balancing method based on an improved particle swarm and hippopotamus algorithm provided in an embodiment of the present invention; Figure 6 is a link load display diagram of the improved particle swarm optimization algorithm provided by an embodiment of the present invention; Figure 7 is a link load display diagram of a particle reset optimization algorithm provided in an embodiment of the present invention; Figure 8 It is a link load display diagram of the basic particle swarm algorithm provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0010] The load balancing method for integrated space and earth network based on improved particle swarm and hippo algorithm provided in the embodiment of the present invention is directed to the integrated space and earth network. Figure 1 , perform the following steps S1 to S3 to complete the optimization of the space-ground integrated network structure: Step S1: Establish a network topology based on the integrated network structure of the ground and the sky, establish a target optimization model based on the multi-commodity flow characteristics of network traffic, link bandwidth constraints, and optimization requirements for traffic distribution, and define corresponding constraints for the target optimization model; The specific steps for establishing the target optimization model in step S1 are as follows: Step S1.1: The network topology is a directed graph ,in V is a collection of nodes, E is a set of edges, for nodes in the network i To Node jThe link is edge To express, and use Represents edge The bandwidth available, Represents edge Reliability weight; Assuming that in the network G There are K Data streams are transmitted simultaneously. k ,use Indicates the bandwidth required; On, data flow k The flow rate To represent the entire data flow k The flow rate is expressed as vector To represent; data flow k The transmission cost per unit flow is represented by the vector To represent; data flow k Path The length is ; Step S1.2: Each path is composed of many links, and a linear function is used to calculate the transmission cost. Define as follows: ; in, , is a constant, for Length, for Utilization rate; In a given network In the above formula, the objective function of the flow distribution algorithm based on the path length and link utilization is expressed as follows: ; In the framework of the algorithm, the position of each particle corresponds to a fitness value calculated by the objective function, which is used to measure the quality of the particle solution. When the algorithm is applied to constrained optimization problems, the key lies in two main steps: first, the specific solution of the problem needs to be converted into the encoding form of particles; second, a fitness function that can accurately reflect the performance of the particles must be defined; this fitness function will guide the evolution of the particles in the search space in order to find the optimal solution; so the final objective function model is expressed as: ; In the formula, represents the target optimization model, , is a constant, For data flow k Path Length, for Utilization rate; Step S1.3: For the target optimization model, define link bandwidth constraints, traffic request bandwidth constraints, non-negativity constraints, and traffic conservation constraints. That is, under these constraints, the transmission cost of the entire network is minimized. The details are as follows: Link bandwidth constraints: ; In the formula, Represents a slave node in the network i To Node j Traffic of all links; Link bandwidth constraints ensure that traffic allocation does not exceed the actual carrying capacity of the link, thereby avoiding link congestion, optimizing network resource utilization, and providing necessary constraints for the optimization algorithm to achieve effective load balancing and network performance improvement.

[0011] Traffic request bandwidth constraints: ; Traffic request bandwidth constraints ensure that the traffic demand of each data flow is fully met in path allocation, providing a clear allocation target for the optimization algorithm, thereby supporting the implementation of load balancing and ensuring efficient operation of the network.

[0012] Non-negativity constraints: ; The role of the non-negativity constraint is to ensure the rationality of traffic distribution. It stipulates that the traffic on each link must be greater than or equal to zero, thereby avoiding the occurrence of unrealistic negative traffic. This constraint, combined with the link bandwidth constraint and the traffic request constraint, provides clear boundary conditions for the optimization algorithm, ensuring that the traffic distribution solution solved by the algorithm is feasible in the actual network environment, while ensuring the stability of network operation and the realization of load balancing.

[0013] Conservation constraints for flow: ; in, and They represent the source node and destination node of data flow k respectively; The flow conservation constraint ensures that the flow input and output of each node are balanced in the flow distribution problem, that is, the sum of the flow entering a node is equal to the sum of the flow leaving the node. This is necessary in the actual network environment, because data will not be created or disappeared in the transmission process, and the conservation principle must be followed. The flow conservation constraint provides a basic constraint condition for the optimization algorithm, ensuring that the flow distribution solution solved by the algorithm is feasible in the actual network.

[0014] Step S2: An improved scheme of using particles to represent the space-ground integrated network structure. According to the characteristics of the space-ground integrated network structure, the position and velocity of the particles are encoded. The particles are taken as individuals, and the collection of all particles constitutes a population. In the population initialization stage, an adaptive meta-heuristic mapping is introduced to increase the diversity of the population. The specific steps of step S2 are as follows: Step S2.1: Each particle represents a potential improvement plan for the space-ground integrated network structure. Each particle not only explores the best position it has found in the past (individual extreme value, denoted as ), and also finds the best known position of all particles (global extremum, denoted by ); During the iteration process, the particle adjusts its speed and position according to these two extreme values ​​to get closer to these extreme values; the update rules for the speed and position of the particle are defined as follows: ; ; In the formula, represents the number of iterations, , Represents the particle The speed and position of the round iteration, and is the learning factor, ranging from 0 to 2. The learning factor helps particles explore and utilize information in the search space. is the inertia weight, between 0.1 and 0.9, which controls the tendency of the particle to maintain its previous velocity. is a random number between 0 and 1, which introduces randomness to the particle update and increases the diversity of the search. represents the individual optimal position of the particle, Represents the global optimal position of the particle; the particle swarm continuously adjusts and optimizes the solution during the iteration process to seek the optimal solution to the problem; Step S2.2: In the population initialization phase, an adaptive meta-heuristic mapping is introduced, including global optimization and local optimization; global optimization allocates initial flow values ​​for each data flow on all possible links, and local optimization allocates initial flow values ​​based on the sharing degree of the links. Perform dynamic traffic allocation; Adaptive metaheuristic mapping is a technique for dynamically adjusting the initialization strategy. It adaptively selects or combines different initialization methods according to the characteristics of the optimization problem (such as dimension, constraints, etc.) to generate a high-quality and diverse initial population. This mapping method can further improve the global search capability, adaptability and robustness of the method of the present invention, reduce the risk of premature convergence, and reduce the computational cost. Adaptive metaheuristic mapping is divided into two parts: global optimization (fixed dimension) and local optimization (dynamic dimension).

[0015] The goal of the global optimization is to assign initial flow values ​​to each data flow on all possible links to optimize the performance of the entire network. The specific method is as follows: Due to the dimension of the global optimization of the present invention Very huge. is the total number of links, and the initialization strategy directly adopts a hybrid method of uniform distribution and normal distribution. k and each link , the globally optimized value Initialized to: ; in, For uniformly distributed sampling, is a normal distribution sample, Indicates that x The value of is restricted to the range [0,1]: ; Globally optimized vector By all constitute: ; in, Length is , is the total number of links, is the total number of data flows; This method combines the randomness of uniform distribution and the concentration of normal distribution, and can generate initial solutions with good diversity and stability in high-dimensional space.

[0016] The dimension of the local optimization is dynamic and depends on the actual path length of each flow. ; For each data stream k , whose path Contains several links, Represents the local optimization value, for each data flow k and its path Each link in , according to the path length Choose a different sampling strategy: when When , uniform distribution sampling is used: ; Represents a uniform distribution The random number obtained by sampling is used to introduce randomness; when When , normal distribution sampling is adopted: ;

[0017] Represents a normal distribution The random numbers sampled from are used to introduce centralization.

[0018] when When , a mixed method of uniform distribution and normal distribution is used: ; By mixing the sampling results of these two distributions, an initial flow distribution value that is both random and has a certain central tendency can be generated, thus providing a better starting point for the optimization process.

[0019] In order to consider the sharing degree of the link, it is assumed that the link Shared by multiple data streams, the degree of sharing, i.e. the number of times used, is , the adjusted flow rate is:

[0020] in, Indicates link the degree of sharing; Finally, the locally optimized vector By all Its length is the sum of all path lengths: ; in, Length is , is the total number of links, The total number of data flows.

[0021] The dimension and sampling strategy of local optimization depend on the actual path length of the data flow. . When the path length is short (e.g. ), use uniform distribution sampling; when the path length is moderate (such as ), use normal distribution sampling; when the path length is long (such as ), using a mixture of uniform distribution and normal distribution. This strategy can dynamically adjust the initialization method according to the complexity of the path, thereby better adapting to the characteristics of different paths.

[0022] In the local optimization part, traffic distribution will be based on the sharing degree of the link. Make adjustments to ensure that traffic is distributed more reasonably and avoid performance issues caused by link overload.

[0023] Step S2.3: construct particles as follows: Each particle is composed of a global optimization part and a local optimization part: ; In the formula, represents particles, represents the globally optimized vector, represents the local optimization vector, represents a vector concatenation operation; the collection of all particles constitutes a population: ; In the formula, Represents the population, represents the particles in the population, Represents the total number of all particles.

[0024] Step S3: Solve the target optimization model, introduce a selective reset mechanism to reset the position of the particles, use the Hippo optimization algorithm to simulate the defense and escape behaviors of the particles, compare the new positions and fitness values ​​of the particles after the defense and escape behaviors, select better particles as the optimal ground-ground integrated network structure output, and complete the optimization of the ground-ground integrated network structure.

[0025] The specific steps of step S3 are as follows: Step S3.1: Start iteration, introduce a selective reset mechanism, calculate the fitness value of each particle in the population, so as to quickly locate the high-quality solution area, thereby improving the global search ability of the algorithm; when the convergence speed of the particle exceeds the set threshold, that is, the particle moves quickly to the local optimal solution, reset the position of the particle and place it at a random position in the search space; In step S3.1, for each particle in the population, if the difference between the current particle's fitness and the individual best fitness is greater than 1% of the individual best fitness, it is considered that the particle converges too fast, and the particle position is reset to a random position: ; In the formula, is the individual optimal fitness of the particle, is the current fitness of the particle; If the current fitness of the particle is greater than the individual optimal fitness, the individual optimal fitness is updated; if the current fitness of the particle is greater than the global optimal fitness, the global optimal fitness is updated, and the particle's speed and position are updated as the optimal solution output of the target optimization model.

[0026] The pseudo code of the above steps is shown in Table 1: Table 1: Particle reset mechanism pseudo code

[0027] The schematic diagram of the particle reset mechanism is as follows Figure 2 As shown, Figure 2 In the figure, the red dot indicates the initial position of the particle, the blue dot indicates the position after reset, the green dot indicates the global optimal solution, and the orange dot indicates the individual optimal solution. It can be seen that when the particle moves quickly to the local optimal solution, the particle is reset to a random position in the search space through the reset mechanism, thereby improving the global search ability of the algorithm.

[0028] Step S3.2: During the iteration process, for particles whose positions are reset, the defense behavior and escape behavior of the Hippo optimization algorithm are simulated to calculate the new positions of the particles after the defense behavior and escape behavior. According to the fitness value of the new positions of the particles after the defense behavior and escape behavior, the algorithm can deeply explore the details of the solution area, thereby improving the local search ability of the algorithm; and select the better particle position as the final result; The defense and escape behaviors of the Hippo optimization algorithm are simulated to improve local search capabilities. This simulation enables particles to better avoid falling into local optimal solutions during the search process, while enhancing the detailed exploration of the solution space. When particles approach the local optimal area, the defense behavior prompts them to disperse, while the escape behavior helps the particles break free from the constraints of the local optimal solution and continue to move towards a better solution.

[0029] Hippopotamus's defense against predators: In nature, hippos protect their young from predators by forming a defensive formation. In the optimization algorithm, this behavior is simulated as a repulsion mechanism between particles: when particles are too close to each other, they repel each other to avoid gathering near the local optimal solution. This mechanism is achieved by randomly selecting a "predator" particle and moving the current particle away from the "predator".

[0030] Hippopotamus fleeing from predators: When encountering a predator, hippopotamus will flee quickly to ensure their own safety. In the optimization algorithm, this behavior is simulated as the sensitivity of particles to the global optimal solution (gBest): if the particle finds itself too close to gBest, it will be pushed away from gBest.

[0031] By simulating two key behaviors of the Hippo optimization algorithm, the diversity among particles and the sensitivity to local optimality are increased, thereby improving the local search ability of the algorithm.

[0032] The specific steps of step S3.2 are as follows: Step S3.2.1: For the particle whose position is reset in step S3.1, if its random search probability is less than 0.5, that is, it satisfies the following formula, the defense behavior and escape behavior of the Hippopotamus optimization algorithm are simulated for the particle: ; In the formula, Indicates that the probability of random search is small. In this way, a 50% probability is used to determine whether the current particle needs to perform a local search. The role of this probability mechanism is to balance the global search and the local search: to avoid all particles from entering the local search at the same time, thereby retaining a certain global search capability.

[0033] Perform steps S3.2.2-S3.2.4 to simulate the behavior of a hippopotamus defending itself against a predator: Step S3.2.2: Generate predator positions As follows: ; in, is the lower bound of the global search space, It is the upper bound of the global search space, ensuring that the initial position of the particle is within a reasonable range and avoiding the particle starting the search from an unreasonable position. is a uniformly distributed random number in the range ; Its function is to generate a random position in the current search space as the position of the predator. By randomly generating the position of the predator, it simulates the scene of the hippopotamus facing the predator, thereby introducing randomness and enhancing the diversity of the search.

[0034] Step S3.2.3: Calculate the distance between the current solution (hippopotamus) and the predator As follows: ; in, is the current particle position; Step S3.2.4: Generate a defense strategy and calculate the new position of the particle after the defense behavior: If the predator's position has lower fitness than the current solution, that is, it is better than the current solution, then adjust the position of the current solution according to the defense strategy: ; Otherwise, use a more conservative defense strategy: ; in, is the new position of the particle after the defensive action, , that is, randomly generate a value from the interval [2,4); , that is, randomly generate a value from the interval [1,1.5); , that is, randomly generate a value from the interval [2,3); is the sampling angle parameter in the uniform distribution, , that is, from the interval [−2 π ,2 π ) generates a random value; is a uniformly distributed random number in the range , for A random number; Perform steps S3.2.5-S3.2.8 to simulate the behavior of a hippopotamus fleeing from a predator: Step S3.2.5: Define the local search range: According to the current number of iterations and the maximum number of iterations T , dynamically adjust the upper and lower bounds of local search: ; ; in, is the upper bound of the local search, is the lower bound of the local search; Step S3.2.6: Randomly select an escape strategy: Randomly select an escape strategy , the escape strategy is one of the following three: ; In the formula, is a uniformly distributed random number in the range of [-1,1); ; In the formula, is a uniformly distributed random number in the range [0,1); ; In the formula, represents normally distributed random numbers; Step S3.2.7: Calculate the new position of the particle after the escape behavior according to the selected escape strategy: ; in, is the new position of the particle after the escape action, for A random number within the range; by multiplying the random number, the step size can be adjusted so that the step size of each escape behavior is slightly different, increasing the diversity of the search.

[0035] Step S3.2.8: Compare the fitness of the new positions of the particles after the defense behavior and the escape behavior, and select the better particle position as the final result: ; In the formula, is the fitness of the new position of the particle after the escape behavior, is the fitness of the new position of the particle after the defensive behavior; Constrain the generated new positions to be within the bounds of the search space: ; in, Represents a crop operation.

[0036] This step is used to limit the particle position within the boundary of the solution space (0 to 1), ensuring that the particle is always within the valid solution range, preventing calculation errors caused by exceeding the boundary or violating the problem constraints, thereby ensuring the stability and convergence of the algorithm.

[0037] Step S3.3: Determine whether the maximum number of iterations is met. If so, the space-ground integrated network structure corresponding to the individuals in the iterated population is output as the optimal space-ground integrated network structure. Otherwise, execute step S3.1.

[0038] The fitness convergence comparison diagram of the method of the present invention (PH-PSO) and the basic particle swarm optimization algorithm (PSO), particle reset optimization algorithm (PR-PSO), and improved particle swarm optimization algorithm (IPSO) is shown in the figure. Figure 3 , jitter comparison chart reference Figure 4 ,Comparison shows that the jitter index of the proposed method is improved by 28.018%, 17.996%, and 11.111% respectively compared with the basic particle swarm optimization algorithm, the particle reset optimization algorithm, and the improved particle swarm optimization algorithm; The link load display diagrams of the method of the present invention (PH-PSO), basic particle swarm optimization algorithm (PSO), particle reset optimization algorithm (PR-PSO), and improved particle swarm optimization algorithm (IPSO) are respectively referred to Figure 5 , Figure 6 , Figure 7 , Figure 8, where different link lines represent different link utilizations, the thin solid line represents a link utilization of 0-0.3, the dashed line represents a link utilization of 0.3-0.5, the dotted line represents a link utilization of 0.5-0.7, and the dotted-dashed line represents a link utilization of 0.7-1; the other three algorithms all generally experience link congestion (utilization greater than 0.3), while the link load display diagram corresponding to the method of the present invention is basically a thin solid line, indicating that after adopting this method, the links tend to be balanced and there is no link congestion. At the same time, it also verifies that this method can balance the traffic in the network and well ensure the link utilization.

[0039] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A space-ground integrated load balancing method based on improved particle swarm and Hippo algorithm, characterized in that: For the integrated space-ground intelligent network, the following steps S1 to S3 are performed to complete the optimization of the integrated space-ground intelligent network structure: Step S1: Establish a network topology based on the integrated network structure of the ground and the sky, establish a target optimization model based on the multi-commodity flow characteristics of network traffic, link bandwidth constraints, and optimization requirements for traffic distribution, and define corresponding constraints for the target optimization model; Step S2: An improved scheme of using particles to represent the space-ground integrated network structure. According to the characteristics of the space-ground integrated network structure, the position and velocity of the particles are encoded. The particles are taken as individuals, and the collection of all particles constitutes a population. In the population initialization stage, an adaptive meta-heuristic mapping is introduced to increase the diversity of the population. Step S3: Solve the target optimization model, introduce a selective reset mechanism to reset the position of the particles, use the Hippo optimization algorithm to simulate the defense and escape behaviors of the particles, compare the new positions and fitness values ​​of the particles after the defense and escape behaviors, select better particles as the optimal ground-ground integrated network structure output, and complete the optimization of the ground-ground integrated network structure.

2. The space-ground integrated load balancing method based on improved particle swarm and Hippo algorithm according to claim 1 is characterized in that: The specific steps for establishing the target optimization model in step S1 are as follows: Step S1.1: The network topology is a directed graph ,in V is a collection of nodes, E is a set of edges. For nodes in the network i To Node j Link, with edge To express, and use Represents edge The bandwidth available, Represents edge Reliability weight; Assuming that in the network G There are K Data streams are transmitted simultaneously. k ,use Indicates the bandwidth required; On, data flow k The flow rate To represent the entire data flow k The flow rate is expressed as vector To express; Data Flow k The transmission cost per unit flow is represented by the vector To express; Data Flow k Path The length is ; Step S1.2: In a given network In the above example, the target optimization model is established as follows: ； In the formula, represents the target optimization model, , is a constant, For data flow k Path Length, for Utilization rate; Step S1.3: For the target optimization model, define the link bandwidth constraint, traffic request bandwidth constraint, non-negativity constraint, and traffic conservation constraint as follows: Link bandwidth constraints: ； In the formula, Represents a slave node in the network i To Node j Traffic of all links; Traffic request bandwidth constraints: ； Non-negativity constraints: ； Conservation constraints for flow: ； in, and Respectively represent data flow k The source and destination nodes.

3. The space-ground integrated load balancing method based on improved particle swarm and hippo algorithm according to claim 1, characterized in that: The specific steps of step S2 are as follows: Step S2.1: Define the update rules for the particle's velocity and position as follows: ； ； In the formula, represents the number of iterations, , Represents the particle The speed and position of the round iteration, and is the learning factor, ranging from 0 to 2. is the inertia weight, between 0.1 and 0.9, is a random number between 0 and 1, represents the individual optimal position of the particle, represents the global optimal position of the particle; Step S2.2: In the population initialization phase, an adaptive meta-heuristic mapping is introduced, including global optimization and local optimization; global optimization allocates initial flow values ​​for each data flow on all possible links, and local optimization allocates initial flow values ​​based on the sharing degree of the links. Perform dynamic traffic allocation; Step S2.3: construct particles as follows: Each particle is composed of a global optimization part and a local optimization part: ； In the formula, represents particles, represents the globally optimized vector, represents the local optimization vector, represents a vector concatenation operation; the collection of all particles constitutes the population: ； In the formula, Represents the population, represents the particles in the population, Represents the total number of all particles.

4. The space-ground integrated load balancing method based on improved particle swarm and Hippo algorithm according to claim 3 is characterized in that: The specific method of the global optimization described in step S2.2 is as follows: For each data stream k and each link , is the total number of links, the globally optimized value Initialized to: ； in, For uniformly distributed sampling, is a normal distribution sample, Indicates that x The value of is restricted to the range [0,1]: ； Globally optimized vector By all constitute: ； in, Length is , is the total number of links, The total number of data flows.

5. The space-ground integrated load balancing method based on improved particle swarm and hippo algorithm according to claim 3 is characterized in that: The method of local optimization described in step S2.2 is as follows: Represents the local optimization value, for each data flow k and its path Each link in , according to the path length Choose a different sampling strategy: when When , uniform distribution sampling is used: ； Represents a uniform distribution The random number sampled from ; when When , normal distribution sampling is adopted: ； Represents a normal distribution The random number sampled from ; when When , a mixed method of uniform distribution and normal distribution is used: ； Assume Link If it is shared by multiple data flows, the adjusted traffic is: ； in, Indicates link the degree of sharing; The final locally optimized vector By all Its length is the sum of all path lengths: ； in, Length is , is the total number of links, The total number of data flows.

6. The space-ground integrated load balancing method based on improved particle swarm and Hippo algorithm according to claim 1 is characterized in that: The specific steps of step S3 are as follows: Step S3.1: Start iteration, introduce a selective reset mechanism, calculate the fitness value of each particle in the population, and reset the position of the particle when the convergence speed of the particle exceeds the set threshold; Step S3.2: During the iteration process, for particles whose positions are reset, the defense behavior and escape behavior of the Hippo optimization algorithm are simulated to calculate the new positions of the particles after the defense behavior and escape behavior. According to the fitness values ​​of the new positions of the particles after the defense behavior and escape behavior, a better particle position is selected as the final result; Step S3.3: Determine whether the maximum number of iterations is met. If so, the space-ground integrated network structure corresponding to the individuals in the iterated population is output as the optimal space-ground integrated network structure. Otherwise, execute step S3.

1.

7. The space-ground integrated load balancing method based on improved particle swarm and Hippo algorithm according to claim 6 is characterized in that: In step S3.1, for each particle in the population, if the particle satisfies the following formula, the position of the particle is reset to a random position: ； In the formula, is the individual optimal fitness of the particle, is the current fitness of the particle; If the current fitness of the particle is greater than the individual optimal fitness, the individual optimal fitness is updated; if the current fitness of the particle is greater than the global optimal fitness, the global optimal fitness is updated, and the particle's speed and position are updated as the optimal solution output of the target optimization model.

8. The space-ground integrated load balancing method based on improved particle swarm and hippo algorithm according to claim 6, characterized in that: The specific steps of step S3.2 are as follows: Step S3.2.1: For the particle whose position is reset in step S3.1, if its random search probability is less than 0.5, the defense behavior and escape behavior of the Hippopotamus optimization algorithm are simulated for the particle; Perform steps S3.2.2-S3.2.4 to simulate the behavior of a hippopotamus defending itself against a predator: Step S3.2.2: Generate predator positions As follows: ； in, is the lower bound of the global search space, is the upper bound of the global search space, is a uniformly distributed random number in the range ; Step S3.2.3: Calculate the distance between the current solution and the predator As follows: ； in, is the current particle position; Step S3.2.4: Generate a defense strategy and calculate the new position of the particle after the defense behavior: If the predator's position has lower fitness than the current solution, the position of the current solution is adjusted according to the defense strategy: ； Otherwise, use a more conservative defense strategy: ； in, is the new position of the particle after the defensive action, ; ; ; is the sampling angle parameter in the uniform distribution, , is a uniformly distributed random number in the range , for A random number; Perform steps S3.2.5-S3.2.8 to simulate the behavior of a hippopotamus fleeing from a predator: Step S3.2.5: Define the local search range: According to the current number of iterations and the maximum number of iterations T , dynamically adjust the upper and lower bounds of local search: ； ； in, is the upper bound of the local search, is the lower bound of the local search; Step S3.2.6: Randomly select an escape strategy: Randomly select an escape strategy , the escape strategy is one of the following three: ； In the formula, is a uniformly distributed random number in the range of [-1,1); ； In the formula, is a uniformly distributed random number in the range [0,1); ； In the formula, represents normally distributed random numbers; Step S3.2.7: Calculate the new position of the particle after the escape behavior according to the selected escape strategy: ； in, is the new position of the particle after the escape action, for A random number in a range; Step S3.2.8: Compare the fitness of the new positions of the particles after the defense behavior and the escape behavior, and select the better particle position as the final result: ； In the formula, is the fitness of the new position of the particle after the escape behavior, is the fitness of the new position of the particle after the defensive behavior; Constrain the generated new positions to be within the bounds of the search space: ； in, Represents a crop operation.

Images