Mobile edge server deployment method based on star-sparrow optimization strategy

Abstract

The invention discloses a mobile edge server deployment method based on a star-sparrow optimization strategy, and provides an improved star-sparrow optimization algorithm for the deployment problem of an edge server in a smart city scene. In a smart city base station scene, in order to minimize time delay and load balance, multi-objective optimization is converted into single-objective optimization by using a fitness function. And meanwhile, a DPC-NOA algorithm is used for automatically determining a clustering center of the data and the deployment number of the servers. According to the DPC-NOA algorithm, the generalized neighborhood similarity is introduced to improve the density peak value clustering process, so that the clustering accuracy and robustness are enhanced. Five algorithms are used for a contrast experiment with the algorithm, and the experiment result shows that the problem of deployment of the edge server can be effectively solved by the DPC-NOA algorithm. In a simulation experiment, the DPC-NOA algorithm can obtain the optimal fitness through a small number of iterations, meanwhile, the DPC-NOA algorithm shows high stability, and compared with other algorithms, the DPC-NOA algorithm shows high performance advantages and has certain value.

Description

Technical Field

[0001] The present invention belongs to the fields of cloud computing and edge computing, and particularly relates to a method for deploying mobile edge servers based on a starling optimization strategy Background Art

[0002] In recent years, due to the rapid development of information technology and the integration of technologies such as the Internet of Things, big data, and artificial intelligence, urban resource management has become more intelligent. However, the widespread application of these technologies has also brought about an explosive growth in the amount of data, posing a severe challenge to the traditional centralized data processing mode. Users have an urgent need for an efficient and smooth user experience, especially in application scenarios with extremely high real-time requirements such as virtual reality, augmented reality, and unmanned driving. If relying on the traditional cloud computing mode, data needs to be transmitted over a long distance to a remote cloud data center for processing, which will result in the inability to meet the real-time requirements of these applications and thus affect the user experience

[0003] Mobile Edge Computing (MEC) reduces the transmission distance of data in the network, reduces network latency, and alleviates network congestion by deploying computing and storage resources at network edge nodes close to users. This enables data to be processed close to the source, thereby improving the response speed of application programs and the user experience. The efficiency of mobile edge computing depends to a large extent on the reasonable deployment of edge servers, so determining the optimal location and quantity of servers is a key issue

[0004] Based on the analysis of the above background and problems, the present invention solves the problem of how to efficiently deploy edge servers in a smart city environment. By reasonably deploying edge servers at some base stations, intelligent devices can wirelessly connect to the base stations at a short distance and submit task loads to the edge servers for processing. This deployment method not only improves the utilization efficiency of edge servers but also greatly optimizes the service experience of users Summary of the Invention

[0005] The objective of the present invention is to solve the deployment problem of edge servers in the smart city scenario, and an improved starling optimization algorithm is proposed. First, in the scenario of large-scale base stations in the smart city, in order to minimize the access delay and load balancing degree of the servers, a fitness function is used to transform the multi-objective optimization problem into a single-objective optimization problem. At the same time, a weight parameter w is used to represent the emphasis on the two objectives; and the DPC-NOA algorithm is used to automatically determine the clustering center of the data and the number of deployed servers. The DPC-NOA algorithm improves the density peak clustering process by introducing the generalized neighborhood similarity, thereby enhancing the accuracy and robustness of clustering. Based on the clustering results, the initialization process of the algorithm is further optimized to improve the search efficiency and performance. In order to verify the feasibility of the proposed algorithm, five other algorithms are used for comparative experiments with this algorithm. The experimental results show that the DPC-NOA algorithm can effectively solve the deployment problem of edge servers. In the simulation experiment, the DPC-NOA algorithm can obtain the best fitness through fewer iteration times, and at the same time shows high stability. Even when the deployment environment fluctuates greatly, it can still maintain excellent performance. In the test of the actual scenario, although the number of iterations of the algorithm increases relatively, compared with other algorithms, it still shows strong performance advantages and has certain value.

[0006] The method for deploying mobile edge servers based on the starling optimization strategy of the present invention mainly includes the following key steps:

[0007] Step 1: Construction of the system model:

[0008] Step 1.1: Establishment of the network model;

[0009] Step 1.2: Establishment of the communication model;

[0010] Step 1.3: Establishment of the access delay model;

[0011] Step 1.4: Establishment of the load balancing model;

[0012] Step 2: Algorithm design:

[0013] Step 2.1: DPC clustering algorithm based on generalized neighborhood similarity;

[0014] Step 2.2: Starling optimization algorithm;

[0015] Step 2.3: Method for deploying mobile edge servers based on the starling optimization strategy;

[0016] Step 3: Experiment:

[0017] Step 3.1: Experimental parameters;

[0018] Step 3.2: Actual simulation settings.

[0019] Further, in step 1.1, a network model is established. The server deployment problem is abstracted as a network topology structure and modeled using an undirected graph G=(V, E). The network consists of nodes and edges. Among them, the nodes V represent network entities such as base stations and edge servers, and the edges E represent the connection relationships between these entities.

[0020] Further, the method for establishing the communication model in step 1.2 is as follows. The communication between the edge server and the base station is transmitted through a channel. Using Shannon's theorem, the maximum transmission rate between the base station and the edge server is calculated according to the channel bandwidth and signal-to-noise ratio.

[0021] Denote D as the channel bandwidth, H / N as the signal-to-noise ratio, where H represents the signal power received during communication and N represents the noise power received by the receiver during communication. Then the maximum channel data transmission rate C in an additive white Gaussian noise channel satisfies:

[0022]

[0023] The communication rate between the base station and its affiliated edge server is obtained according to Shannon's theorem as:

[0024]

[0025] where C ij is the maximum data transmission rate between the base station and the edge server, D is the communication channel bandwidth between the base station and the edge server, W is the channel gain for the task to be transmitted to the edge server, Q j is the transmission power of the local device of base station b j , and N 0 is the white noise power.

[0026] Further, the method for establishing the access delay model in step 1.3 is as follows: In the delay model of server deployment, the access delay includes propagation delay, transmission delay, and queuing delay. The propagation delay refers to the delay generated when the base station sends the load to the edge server through the fiber optic network. In the actual scenario, assume that an edge server s i is responsible for a base station b j and the communication distance is d i,j . The edge server provides computing services for the base station. Then the calculation formula for d i,j is:

[0027]

[0028] where R is the radius of the earth, and represent the longitude and latitude of base station b j , and Denote the longitude and latitude of the edge server s i , in radians;

[0029] Therefore, the formula for calculating the average propagation delay of all base stations that the edge server s i needs to be responsible for is:

[0030]

[0031] where v is the data propagation speed in the network, and LG i,j is a binary variable;

[0032] The access delay T of the edge server s i is the sum of the propagation delay, transmission delay, and queuing delay, and its calculation formula is: j

[0033]

[0034] The average access delay T of the edge server in the edge server deployment model is:

[0035]

[0036] Furthermore, the method for establishing the load balancing model in step 1.4 is as follows: During the edge server deployment process, the goal of load balancing is to reasonably allocate the load of each server. By modeling and analyzing the load balancing, the reasonable allocation of resources can be achieved, and the server deployment effect can be optimized.

[0037] Furthermore, in step 2.1, the DPC clustering algorithm based on generalized neighborhood similarity is used. The DPC algorithm automatically determines the number of clusters by calculating the local density ρ of each data point and the minimum distance δ to the high-density points. This enables the DPC algorithm to adapt to the actual distribution of the data. Regardless of how the data volume and distribution change, it can be dynamically adjusted to accurately identify the dense regions;

[0038] The local density ρ in the DPC algorithm is calculated by setting a distance threshold dc. This distance threshold dc is a user-defined parameter used to determine which points are considered neighbors. By setting an appropriate distance threshold dc, the points that affect the local density value of a certain data point can be effectively controlled, and the size of the density threshold directly determines the range of the neighborhood;

[0039] The steps of the DPC clustering algorithm based on generalized neighborhood similarity are:

[0040] First, calculate the local density of each data point. The improved algorithm no longer relies solely on the direct distance between data points but introduces the generalized neighborhood similarity to redefine the local density. The formula for the local density ρ i is:

[0041]

[0042] Among them, Gns(i, j) is the generalized neighborhood similarity between data points i and j;

[0043] Then, calculate the relative distance δ from the data point to the high-density point i , in order to enhance the recognition of low-density points, the information of adjacent points is considered in the new relative distance definition, so that the points in the low-density environment can obtain compensation in the distance calculation and increase the possibility of becoming the clustering center. The definition is as follows:

[0044]

[0045] Among them, d ij is the distance between points i and j, KNN i and KNN j are the sets of the nearest neighbor points of points i and j respectively, and p and q represent any points in the neighborhood;

[0046] After determining the local density and relative distance, calculate the decision value γ of each point, and select the point with the highest decision value as the clustering center. The calculation formula of the decision value γ is:

[0047] γ = ρ × δ (9)

[0048] Then, the algorithm enters the data point allocation stage. First, allocate the core points based on the generalized neighborhood similarity. For the unallocated data point q, if q is the nearest neighbor of an allocated point p and Gns(p, q) is higher than the average similarity between p and its neighbors, then allocate q to the same cluster as p. For the still unallocated points, auxiliary allocation is performed through the support coefficient, which is used to measure the support strength of a point for its adjacent points. The support coefficient is defined as:

[0049]

[0050] Furthermore, the algorithm described in step 2.2 is the starling optimization algorithm. The specific steps of the starling optimization algorithm are as follows:

[0051] First, the algorithm randomly generates a population in the solution space. Each individual represents a potential solution, and the relevant parameters of the algorithm, such as the population size and the maximum number of iterations, are initialized. The position X i of each individual represents a point in the solution space. These initial positions are usually randomly generated to ensure coverage of a wide solution space;

[0052] In each iteration of the algorithm, first calculate the fitness f(X i) to evaluate the advantages and disadvantages of its current position. This step determines the survival ability of an individual in the population. During the foraging stage, the starling will continuously adjust its position according to the surrounding environment. The evaluation formula is as follows:

[0053]

[0054] where τ 1 、τ 2 and τ 3 are random numbers within the range of [0, 1], used to determine the rule of position update. represents the new position of the starling in the (t + 1)-th iteration. μ is a random number generated based on the flight distribution. X A and X B are the positions of different individuals randomly selected from the population;

[0055] Then, the algorithm simulates the behavior of the starling storing food, stores the excellent solutions found by each individual. The storage process is achieved through spatial memory and surrounding references. The starling will return to the storage location when necessary. The storage formula is:

[0056]

[0057] where represents the position of the first reference point of the starling in the t-th iteration;

[0058] In the recovery stage, the starling will return to the storage location to find food, thus achieving a balance between wide search and local exploitation. This process further optimizes the solution through the following position update formula:

[0059]

[0060] where τ 4 is another random number within the range of [0, 1], and r1, r2 are random numbers within the range of [0, 1];

[0061] In some stages of the optimization process, the algorithm will re-evaluate the stored excellent solutions to check whether they are still superior. In each iteration, the algorithm updates the known global optimal solution to ensure that the algorithm can continuously track and record the current best solution. When the algorithm reaches the maximum number of iterations, the iteration process will stop and return the optimal solution found currently.

[0062] Furthermore, step 2.3 is the specific step of the mobile edge server deployment method based on the improved starling optimization strategy:

[0063] (1) Data collection and preprocessing:

[0064] First, collect the detailed geographical location information of the base stations. This data will serve as the basis for determining the deployment locations of the edge servers, and standardize the collected base station data to provide reliable data input for subsequent analysis;

[0065] (2) Construct the distance matrix and preliminary clustering:

[0066] Based on the data of the base stations, calculate the generalized neighborhood similarity between the base stations, so as to analyze the distribution of the base stations and identify the clustering centers in the high-density areas, and based on these identified clustering centers, determine the number and initial deployment locations of the edge servers;

[0067] (3) Analysis of clustering results and determination of the initial deployment plan:

[0068] The preliminarily determined server locations represent the clustering centers in the high-density areas, ensuring that the servers can effectively cover the areas with higher demand. These initial locations serve as the starting solutions for the DPC-NOA algorithm, and on this basis, further optimize the deployment locations of the edge servers;

[0069] (4) Location optimization by the Starling optimization algorithm:

[0070] The Starling optimization algorithm combines global search and local optimization capabilities. According to the access delay and load balancing performance metrics, dynamically adjust the positions of the servers. In each round of iteration, the DPC-NOA algorithm gradually optimizes the positions of each server based on the deployment positions of the previous round of iteration. When the maximum number of iterations is reached, the algorithm will return the optimal solution found currently.

[0071] Furthermore, in step 3.1, the experimental parameters are used to evaluate the robustness of the DPC-NOA algorithm in the edge server deployment model through experiments. First, the operating environment of the experiment and five comparison algorithms are introduced. Then, by analyzing the experimental results, it is proved that the DPC-NOA algorithm is efficient in optimizing the task access delay and server load balancing.

[0072] Furthermore, in step 3.2 for evaluating the performance of the DPC-NOA algorithm, five algorithms are selected for comparative experiments, namely the Grey Wolf Optimization algorithm, the Genetic Algorithm, the Genetic Simulated Annealing algorithm, the Particle Swarm Optimization algorithm, and the K-means clustering algorithm. By comparing the performances of these algorithms in terms of the average access delay and load balancing degree of the servers, their advantages and disadvantages are evaluated.

[0073] The advantages and positive effects of the present invention are:

[0074] The present invention mainly designs a method for deploying mobile edge servers based on the starling optimization strategy. In this method, aiming at the server deployment problem faced in optimizing the user experience in a smart city, an improved starling optimization algorithm combining generalized neighborhood similarity and density peak clustering is proposed. First, a system model of access delay and load balancing is constructed, and the server deployment problem is transformed into an optimal solution problem. Density peak clustering based on generalized neighborhood similarity locates the densely populated base station areas as potential deployment points by redefining local density and generalized neighborhood similarity. Then, the improved starling optimization algorithm optimizes the delay and load balancing by simulating the foraging and food storage behaviors of starlings and dynamically adjusting the server positions according to the fitness function. Experimental results show that compared with other algorithms, the method proposed in this paper performs well in optimizing access delay and load balancing and has certain value. Description of the Drawings

[0075] Figure 1 is the number of edge servers under different base station coverage ranges;

[0076] Figure 2 is the relationship between the number of iterations and the fitness for n = 100 (a), n = 300 (b), and n = 500 (c);

[0077] Figure 3 is the relationship between the fitness and the number of base stations for w = 0.3 (a), w = 0.5 (b), and w = 0.7 (c);

[0078] Figure 4 is the relationship between the weight parameter w and the fitness for n = 100 (a), n = 300 (b), and n = 500 (c);

[0079] Figure 5 is the relationship between the number of iterations and the fitness for w = 0 (a) and w = 1 (b);

[0080] Figure 6 is the flow chart of the method for deploying mobile edge servers based on the starling optimization strategy of the present invention. Detailed Embodiments

[0081] Example 1

[0082] The method designed in this embodiment was simulated and run on an Acer LAPTOP-PC3V82OF device equipped with a 2.5GHz Intel(R) i5-13500HX processor, 16GB of memory, and a 256GB SSD. To more comprehensively evaluate the performance of the DPC-NOA algorithm, five algorithms were selected for comparative experiments, namely the Grey Wolf Optimization algorithm, the Genetic algorithm, the Genetic Simulated Annealing algorithm, the Particle Swarm Optimization algorithm, and the K-means clustering algorithm. By comparing the performance of each algorithm in terms of the average access delay and load balancing degree of the server, their advantages and disadvantages were evaluated.

[0083] See the appendix Figure 6 , the method for deploying a mobile edge server based on the starling optimization strategy in this embodiment mainly includes the following key steps:

[0084] Step 1: Construction of the system model:

[0085] Step 1.1: Establish a network model;

[0086] Step 1.2: Establish a communication model;

[0087] Step 1.3: Establish an access delay model;

[0088] Step 1.4: Establish a load balancing model;

[0089] Step 2: Algorithm design:

[0090] Step 2.1: DPC clustering algorithm based on generalized neighborhood similarity;

[0091] Step 2.2: Starling optimization algorithm;

[0092] Step 2.3: Method for deploying a mobile edge server based on the starling optimization strategy;

[0093] Step 3: Experiment:

[0094] Step 3.1: Experiment parameters;

[0095] Step 3.2: Actual simulation settings.

[0096] In step 1.1, a network model was established, abstracting the server deployment problem into a network topology structure and modeling it using an undirected graph G=(V, E). The network consists of nodes and edges, where the nodes V represent network entities such as base stations and edge servers, and the edges E represent the connection relationships between these entities.

[0097] The method for establishing the communication model in step 1.2 is as follows. The communication between the edge server and the base station is transmitted through a channel. Using Shannon's theorem, the maximum transmission rate between the base station and the edge server is calculated based on the channel bandwidth and signal-to-noise ratio;

[0098] Let D be the channel bandwidth, H / N be the signal-to-noise ratio, where H represents the signal power received during communication and N represents the noise power received at the receiving end during communication. Then, the maximum channel data transmission rate C under an additive white Gaussian noise channel satisfies:

[0099]

[0100] The communication rate between the base station and its affiliated edge server is obtained according to Shannon's theorem as:

[0101]

[0102] where C ij is the maximum data transmission rate between the base station and the edge server, D is the communication channel bandwidth between the base station and the edge server, W is the channel gain for task transmission to the edge server, Q j is the transmit power of the local device of base station b j and N 0 is the white noise power.

[0103] The method for establishing the access delay model in step 1.3 is as follows: In the delay model of server deployment, the access delay includes propagation delay, transmission delay, and queuing delay. Propagation delay refers to the delay generated when the base station sends the load to the edge server through the fiber optic network. In an actual scenario, assume that an edge server s i is responsible for a base station b j and the communication distance is d i,j . If the edge server provides computing services for the base station, then the calculation formula for d i,j is:

[0104]

[0105] where R is the radius of the earth, and represent the longitude and latitude of base station b j , and represent the longitude and latitude of edge server s i , in radians;

[0106] Therefore, the calculation formula for the average propagation delay of all base stations that edge server s i needs to be responsible for is:

[0107]

[0108] where v is the data propagation speed in the network, LG i,j is a binary variable;

[0109] Edge server s i The access delay T j is the sum of the propagation delay, transmission delay, and queuing delay, and its calculation formula is:

[0110]

[0111] In the edge server deployment model, the average access delay T of the edge server is:

[0112]

[0113] The method for establishing the load balancing model in Step 1.4 is as follows: During the edge server deployment process, load balancing is always a key factor. The goal of load balancing is to reasonably allocate the load of each server. By modeling and analyzing load balancing, the reasonable allocation of resources can be achieved, and the server deployment effect can be optimized.

[0114] In Step 2.1, the DPC clustering algorithm based on generalized neighborhood similarity is used. The DPC algorithm automatically determines the number of clusters by calculating the local density ρ of each data point and the minimum distance δ to the high-density points. This enables the DPC algorithm to adapt to the actual distribution of the data. Regardless of how the data volume and distribution change, it can be dynamically adjusted to accurately identify the dense areas;

[0115] The local density ρ in the DPC algorithm is calculated by setting a distance threshold dc. This distance threshold dc is a user-defined parameter used to determine which points are considered neighbors. By setting an appropriate distance threshold dc, the points that affect the local density value of a certain data point can be effectively controlled, and the size of the density threshold directly determines the range of the neighborhood;

[0116] The steps of the DPC clustering algorithm based on generalized neighborhood similarity are as follows:

[0117] First, calculate the local density of each data point. The improved algorithm no longer relies solely on the direct distance between data points but introduces generalized neighborhood similarity to redefine the local density ρ i The calculation formula is:

[0118]

[0119] where Gns(i,j) is the generalized neighborhood similarity between data points i and j;

[0120] Then, calculate the relative distance δ of the data point to the high-density point i . To enhance the recognition of low-density points, the information of adjacent points is considered in the new relative distance definition, enabling points in a low-density environment to obtain compensation in the distance calculation and increasing the possibility of becoming a clustering center. The definition is as follows:

[0121]

[0122] where d ij is the distance between points i and j, KNN i and KNN j are the sets of the nearest neighbor points of points i and j respectively, and p and q represent any points in the neighborhood;

[0123] After determining the local density and relative distance, calculate the decision value γ of each point, and select the point with the highest decision value as the clustering center. The calculation formula of the decision value γ is:

[0124] γ = ρ × δ (9)

[0125] Then, the algorithm enters the data point assignment phase. First, assign the core points based on the generalized neighborhood similarity. For the unassigned data point q, if q is the nearest neighbor of an assigned point p and Gns(p, q) is higher than the average similarity between p and its neighbors, then assign q to the same cluster as p. For the still unassigned points, perform auxiliary assignment through the support coefficient, which is used to measure the support strength of a point for its adjacent points. The support coefficient is defined as:

[0126]

[0127] The algorithm described in step 2.2 is the starling optimization algorithm. The specific steps of the starling optimization algorithm are as follows:

[0128] First, the algorithm randomly generates a population in the solution space. Each individual represents a potential solution, and initializes the relevant parameters of the algorithm, such as the population size and the maximum number of iterations. The position X i of each individual represents a point in the solution space. These initial positions are usually randomly generated to ensure coverage of a wide solution space;

[0129] In each iteration of the algorithm, first calculate the fitness f(X i ) of each individual to evaluate the quality of its current position. This step determines the survival ability of the individual in the population. The starling adjusts its position continuously according to the surrounding environment during the foraging phase. The evaluation formula is as follows:

[0130]

[0131] where τ 1 , τ 2 and τ 3 are random numbers in the range of [0, 1] used to determine the position update rule, represents the new position of the starling in the (t + 1)-th iteration, μ is a random number generated based on the flight distribution, XA and X B are the positions of different individuals randomly selected from the population;

[0132] Then, the algorithm simulates the behavior of the starling storing food, stores the excellent solutions found by each individual, and the storage process is realized through spatial memory and surrounding reference objects. The starling will return to the storage location when necessary, and the storage formula is:

[0133]

[0134] where, represents the position of the first reference point of the starling in the t-th iteration;

[0135] In the recovery stage, the starling will return to the storage location to find food, thus achieving a balance between wide search and local exploitation. This process further optimizes the solution through the following position update formula:

[0136]

[0137] where τ 4 is another random number within the range of [0, 1], and r1, r2 are random numbers within the range of [0, 1];

[0138] In some stages of the optimization process, the algorithm re-evaluates the stored excellent solutions to check whether they are still superior. In each iteration, the algorithm updates the known global optimal solution to ensure that the algorithm can continuously track and record the current best solution. When the algorithm reaches the maximum number of iterations, the iteration process stops and returns the optimal solution found currently.

[0139] Step 2.3 is the specific step of the mobile edge server deployment method based on the improved starling optimization strategy:

[0140] (1) Data collection and preprocessing:

[0141] First, collect the detailed geographical location information of the base stations. These data will serve as the basis for determining the deployment location of the edge servers, and standardize the collected base station data to provide reliable data input for subsequent analysis;

[0142] (2) Construct the distance matrix and preliminary clustering:

[0143] Based on the data of the base stations, calculate the generalized neighborhood similarity between the base stations, thereby analyzing the distribution of the base stations and identifying the clustering centers in the high-density areas, and based on these identified clustering centers, determine the number and initial deployment locations of the edge servers;

[0144] (3) Analysis of clustering results and determination of the initial deployment plan:

[0145] The preliminarily determined server locations represent the clustering centers in the high-density areas, ensuring that the servers can effectively cover the areas with higher demand. These initial locations serve as the starting solutions for the DPC-NOA algorithm, and based on this, the deployment locations of the edge servers are further optimized.

[0146] (4) Location optimization by the Starling optimization algorithm:

[0147] The Starling optimization algorithm combines global search and local optimization capabilities. According to the access delay and load balancing performance metrics, it dynamically adjusts the positions of the servers. In each iteration, the DPC-NOA algorithm gradually optimizes the positions of each server based on the deployment positions of the previous iteration. When the maximum number of iterations is reached, the algorithm returns the optimal solution found currently.

[0148] In step 3.1, the experimental parameters are used to evaluate the robustness of the DPC-NOA algorithm in the edge server deployment model through experiments. First, the operating environment of the experiment and five comparison algorithms are introduced. Then, by analyzing the experimental results, the effectiveness of the DPC-NOA algorithm in optimizing the task access delay and server load balancing is demonstrated.

[0149] In step 3.2 for evaluating the performance of the DPC-NOA algorithm, five algorithms are selected for comparative experiments, namely the Grey Wolf optimization algorithm, the Genetic algorithm, the Genetic Simulated Annealing algorithm, the Particle Swarm Optimization algorithm, and the K-means clustering algorithm. By comparing the performance of each algorithm in terms of the average server access delay and load balancing degree, their advantages and disadvantages are evaluated.

[0150] Based on the above embodiments, the present invention mainly designs a mobile edge server deployment method based on the Starling optimization strategy. In this method, aiming at the server deployment problem faced in optimizing the user experience in smart cities, an improved Starling optimization algorithm combining the generalized neighborhood similarity and density peak clustering is proposed. First, a system model of access delay and load balancing is constructed, and the server deployment problem is transformed into an optimal solution problem. The density peak clustering based on the generalized neighborhood similarity locates the densely populated base station areas as potential deployment points by redefining the local density and generalized neighborhood similarity. Then, the improved Starling optimization algorithm dynamically adjusts the server positions according to the fitness function by simulating the foraging and food storage behaviors of starlings, thereby optimizing the delay and load balancing. The experimental results show that compared with other algorithms, the method proposed in this paper performs well in optimizing the access delay and load balancing and has certain value.

[0151] In this example, the present invention is tested based on the real base station network dataset provided by the Tianjin Telecommunications Bureau. This dataset includes detailed information on user access to base stations in the city center area. The processed and filtered dataset contains 6,358 base stations, and the base station information includes the ID, longitude and latitude of the base station, and the number of served users. In order to simulate the arrival of user tasks, in the delay model simulation, each base station is assigned an arrival rate generated according to the Poisson distribution, and the performance of different algorithms in the real scenario is compared.

[0152] The simulation experiment results of this example are as follows:

[0153] 1. Relationship between the number of edge servers and the base station coverage range under different numbers of base stations

[0154] Appendix Figure 1 shows that as the base station coverage range gradually expands from 0.02 km to 0.1 km, the required number of edge servers decreases accordingly. This change is mainly because as the coverage range expands, the distance threshold of the clustering algorithm also increases, resulting in the need for fewer servers to cover the same area. After comprehensively considering the base station load and the maximum load of the edge server, this paper selects 0.1 km as the optimal coverage range of the base station.

[0155] 2. Relationship between the number of algorithm iterations and the server fitness

[0156] Appendix Figure 2 indicates that during the process of gradually increasing the number of base stations, the DPC-NOA algorithm demonstrates a faster iteration speed and higher stability. Compared with other algorithms, DPC-NOA can always find a lower fitness value. Although the PSO algorithm shows a similar performance in fitness calculation to DPC-NOA, its poor search ability results in the need for more iteration times. The traditional SA and GA algorithms show instability, and the number of iterations required to reach the lowest fitness is relatively large, making it difficult to meet the requirements of edge server deployment. The fitness calculation ability of the GWO algorithm is comparable to that of DPC-NOA, but its number of iterations is relatively high. The K-means algorithm has never been able to reach the fitness level of DPC-NOA. In summary, the DPC-NOA algorithm shows good adaptability in dealing with different scenarios of changes in the number and distribution of base stations and is suitable for the edge server deployment task.

[0157] 3. Relationship between the number of base stations and the server fitness function

[0158] From the appendix Figure 3It can be concluded that under different settings of the parameter w, the performance of the DPC-NOA algorithm is very stable. Even when the number of base stations changes significantly, the fluctuation of the minimum fitness value calculated by the DPC-NOA algorithm is very small. At the same time, when w = 0.3, that is, when the proportion of load balancing is much higher than the time delay, the GWO algorithm has the problem of falling into the local optimal solution, resulting in an overly large calculated fitness value. In contrast, the minimum fitness of the PSO, SA, GA, and K-means algorithms is generally higher than the result of the DPC-NOA algorithm. Especially, the fitness of the GA algorithm fluctuates greatly and its stability is poor.

[0159] 4. Relationship between different weight parameters w and server fitness

[0160] According to the appendix Figure 4 It can be concluded that the DPC-NOA and K-means algorithms show high stability when calculating fitness. However, the K-means algorithm sometimes falls into the local optimal solution, resulting in a significant increase in fitness. The SA algorithm performs well in terms of fitness stability, but usually calculates a relatively high fitness. The PSO and GWO algorithms only show better performance in specific scenarios, and the fitness of these two algorithms fluctuates greatly, indicating their limited adaptability in different scenarios. This difference in performance reflects the potential limitations of the algorithms when dealing with complex optimization problems.

[0161] 5. Algorithm performance under extreme parameter conditions

[0162] Appendix Figure 5 Indicates that under the condition of weights, the fitness of the PSO algorithm is slightly lower than that of the DPC-NOA algorithm after multiple iterations; under the condition of, the GWO algorithm performs slightly better than the DPC-NOA algorithm after multiple iterations, but the number of iterations is too large. At the same time, the SA and GA algorithms do not perform well under extreme conditions, and the calculated fitness is relatively high. Generally speaking, the DPC-NOA algorithm shows strong stability under various extreme weight conditions, can maintain a low fitness level in different scenarios, and its performance is similar to that under normal conditions.

[0163] The present invention verifies the performance of the DPC-NOA algorithm through simulation experiments and actual scenario experiments, and compares it with the SA, GA, K-means, PSO, and GWO algorithms. The research focus is on the influence of the number of base stations, the number of iterations, and the weight parameters on the fitness of edge server deployment. The experimental results in the real scenario further prove the advantages of the algorithm of the present invention in these aspects.

Claims

1. A mobile edge server deployment method based on starfinch optimization strategy, characterized in that The method mainly includes the following steps:

1. Construction of system model: Section 1.

1. Establish a network model; 1.

2. Establish a communication model; 1.

3. Establish access delay model; Section 1.

4. Establish a load balancing model; Second, algorithm design: Section 2.1, DPC clustering algorithm based on generalized neighborhood similarity; Section 2.2, Starfinch Optimization Algorithm; Section 2.3, Mobile edge server deployment method based on starbird optimization strategy; 3. Experiment: Section 3.1, Experimental parameters; Section 3.2, Actual simulation setup.

2. The mobile edge server deployment method based on the star bird optimization strategy according to claim 1 is characterized in that: In step 1.1, a network model is established, the server deployment problem is abstracted into a network topology structure, and an undirected graph G = (V, E) is used for modeling. The network consists of nodes and edges, where the node V represents network entities such as base stations and edge servers, and the edge E represents the connection relationship between these entities.

3. The mobile edge server deployment method based on the star bird optimization strategy according to claim 1 is characterized in that: The method of establishing the communication model in step 1.2 is as follows: the communication between the edge server and the base station is transmitted through the channel, and the Shannon theorem is used to calculate the maximum transmission rate between the base station and the edge server according to the channel bandwidth and the signal-to-noise ratio; Let D be the channel bandwidth, H / N be the signal-to-noise ratio, H be the signal power received during the communication process, and N be the noise power received by the receiving end during the communication process. Then the maximum transmission rate C of the channel data under the additive white Gaussian noise channel satisfies: The communication rate between the base station and its edge server is obtained according to Shannon's theorem: Among them C ij is the maximum data transmission rate between the base station and the edge server, D is the communication channel bandwidth between the base station and the edge server, W is the channel gain for task transmission to the edge server, Q j Base station b j The local device transmit power, N0 is the white noise power.

4. The mobile edge server deployment method based on the star bird optimization strategy according to claim 1, characterized in that: The method for establishing the access delay model in step 1.3 is as follows: In the server deployment delay model, the access delay includes propagation delay, transmission delay and queuing delay. The propagation delay refers to the delay caused when the base station sends the load to the edge server through the optical fiber network. In the actual scenario, assuming that an edge server s i A base station b j The communication distance is d i,j , the edge server provides computing services for the base station, then d i,j The calculation formula is: Where R is the radius of the Earth, and Indicates base station b j The longitude and latitude of and Represents edge servers i The longitude and latitude of , in radians; Therefore, the edge servers i The formula for calculating the average propagation delay of all base stations to be responsible is: Where v is the speed of data transmission in the network, LG i,j is a binary variable; Edge Servers i The access delay T j It is the sum of propagation delay, transmission delay and queuing delay, and its calculation formula is: The average access delay T of the edge server in the edge server deployment model is:

5. The mobile edge server deployment method based on the star bird optimization strategy according to claim 1, characterized in that: The method for establishing a load balancing model in step 1.4 is as follows: During edge server deployment, the goal of load balancing is to reasonably distribute the load of each server. By modeling and analyzing load balancing, reasonable allocation of resources can be achieved and server deployment effects can be optimized.

6. The method for deploying a mobile edge server based on the starfinch optimization strategy according to claim 1, characterized in that: In step 2.1, the DPC clustering algorithm based on generalized neighborhood similarity is used. The DPC algorithm automatically determines the number of clusters by calculating the local density ρ of each data point and the minimum distance δ to the high-density point. This enables the DPC algorithm to adapt to the actual distribution of the data and dynamically adjust regardless of how the data volume and distribution change, thereby accurately identifying dense areas. The local density ρ in the DPC algorithm is calculated by setting a distance threshold dc, which is a custom parameter used to determine which points are considered neighbors. By setting an appropriate distance threshold dc, it is effective to control which points affect the local density value of a data point. The size of the density threshold directly determines the range of the neighborhood. The steps of the DPC clustering algorithm based on generalized neighborhood similarity are: First, the local density of each data point is calculated. The improved algorithm no longer relies solely on the direct distance between data points, but introduces generalized neighborhood similarity to redefine the local density. The local density ρ i The calculation formula is: Among them, Gns(i,j) is the generalized neighborhood similarity between data points i and j; Then, calculate the relative distance δ from the data point to the high-density point i In order to enhance the recognition of low-density points, the new relative distance definition takes into account the information of adjacent points, so that points in low-density environments can be compensated in the distance calculation, increasing the possibility of becoming a cluster center. The definition is as follows: Among them, d ij is the distance between points i and j, KNN i and KNN j are the nearest neighbor points of point i and j respectively, and p and q represent any point in the neighborhood; After determining the local density and relative distance, the decision value γ of each point is calculated, and the point with the highest decision value is selected as the cluster center. The calculation formula of the decision value γ is: γ=v×δ (9) Then, the algorithm enters the data point allocation stage. First, core points are allocated based on generalized neighborhood similarity. For an unassigned data point q, if q is the nearest neighbor of an assigned point p, and Gns(p,q) is higher than the average similarity between p and its neighbors, q is assigned to the same cluster as p. For unassigned points, auxiliary allocation is performed through the support coefficient, which is used to measure the support strength of a point for its neighboring points. The support coefficient is defined as:

7. The mobile edge server deployment method based on the starfinch optimization strategy according to claim 1, characterized in that: The algorithm described in step 2.2 is the Starfinch Optimization Algorithm. The specific steps of the Starfinch Optimization Algorithm are as follows: First, the algorithm randomly generates a population in the solution space, each individual represents a potential solution, and initializes the relevant parameters of the algorithm, such as the population size and the maximum number of iterations, the position X of each individual i represents a point in the solution space. These initial positions are usually randomly generated to ensure that a wide range of solution space is covered; In each iteration of the algorithm, the fitness f(X i ), which is used to evaluate the pros and cons of its current position. This step determines the survival ability of the individual in the population. During the foraging stage, the starfinch will constantly adjust its position according to the surrounding environment. The evaluation formula is as follows: Among them, τ1, τ2 and τ3 are random numbers in the range of [0,1], which are used to determine the rules for position update. represents the new position of the starfinch in the t+1th iteration, μ is a random number generated based on the flight distribution, and X A and X B are the positions of different individuals randomly selected from the population; Then, the algorithm simulates the behavior of star finches storing food, and stores the excellent solutions found by each individual. The storage process is achieved through spatial memory and surrounding reference objects. The star finches will return to the storage location when necessary. The storage formula is: in, represents the position of the first reference point of the starfinch in the tth iteration; During the recovery phase, the starfinch will return to the storage location to look for food, thus achieving a balance between extensive search and local utilization. This process is further optimized by the following position update formula: Among them, τ4 is another random number in the range of [0,1], r1 and r2 are random numbers in the range of [0,1]; At certain stages of the optimization process, the algorithm re-evaluates the stored excellent solutions to check whether they are still superior. In each iteration, the algorithm updates the known global optimal solution to ensure that the algorithm can continuously track and record the current best solution. When the algorithm reaches the maximum number of iterations, the iterative process stops and returns the currently found optimal solution.

8. The method for deploying a mobile edge server based on the starfinch optimization strategy according to claim 1, characterized in that: Step 2.3 is the specific steps of the mobile edge server deployment method based on the improved starbird optimization strategy: (1) Data collection and preprocessing: First, collect detailed geographic location information of base stations. This data will serve as the basis for determining the deployment location of edge servers. Standardize the collected base station data to provide reliable data input for subsequent analysis. (2) Construct distance matrix and preliminary clustering: Based on the data of base stations, the generalized neighborhood similarity between base stations is calculated to analyze the distribution of base stations and identify the cluster centers in high-density areas. Based on these identified cluster centers, the number and initial deployment locations of edge servers are determined. (3) Clustering result analysis and initial deployment plan determination: The initially determined server locations represent cluster centers in high-density areas, ensuring that the servers can effectively cover areas with high demand. These initial locations serve as the starting solutions for the DPC-NOA algorithm, and on this basis, the deployment locations of edge servers are further optimized; (4) Position optimization of Starfinch optimization algorithm: The Starfinch optimization algorithm combines global search and local optimization capabilities, and dynamically adjusts the location of servers based on access latency and load balancing performance indicators. In each iteration, the DPC-NOA algorithm gradually optimizes the location of each server based on the deployment location of the previous iteration. When the maximum number of iterations is reached, the algorithm returns the optimal solution currently found.

9. The method for deploying a mobile edge server based on the starfinch optimization strategy according to claim 1, characterized in that: Step 3.1 is the experimental parameters. The robustness of the DPC-NOA algorithm in the edge server deployment model is evaluated through experiments. First, the experimental operating environment and five comparison algorithms are introduced. Then, by analyzing the experimental results, the efficiency of the DPC-NOA algorithm in optimizing task access latency and server load balancing is proved.

10. The method for deploying a mobile edge server based on the starfinch optimization strategy according to claim 1, characterized in that: In step 3.2, in order to evaluate the performance of the DPC-NOA algorithm, five algorithms were selected for comparative experiments, namely, the gray wolf optimization algorithm, the genetic algorithm, the genetic simulated annealing algorithm, the particle swarm optimization algorithm, and the K-means clustering algorithm. The advantages and disadvantages of each algorithm were evaluated by comparing their performance in terms of average server access delay and load balancing.

Images