A Service Deployment Method and System Based on Federated Learning

Abstract

This invention discloses a service deployment method and system based on federated learning, relating to the field of edge computing service deployment. The method includes: acquiring user personal information and service interaction data; using a clustering algorithm to cluster users into cluster sets; acquiring available edge servers within each cluster; designing multiple objective functions based on the constraints of the edge servers; using a non-dominated sorting genetic algorithm to solve the optimal solutions of the multiple objective functions to find edge cloud servers within each cluster; acquiring a global federated prediction model based on a deployed local prediction model through federated learning; and using the global federated prediction model to predict the future service request volume of each cluster based on the available edge servers within each cluster, deploying services with increasing popularity to the available edge servers. This invention improves service deployment efficiency, optimizes resource utilization, reduces communication overhead, ensures privacy and security, and enhances user experience.

Description

Technical Field

[0001] This invention relates to the field of edge computing service deployment, and in particular to a service deployment method and system based on federated learning. Background Technology

[0002] With the rapid development of 5G communication technology and the widespread adoption of smart terminals, diverse real-time application demands have emerged, such as video surveillance, smart homes, and online games. However, traditional cloud computing models, which require uploading data and tasks to remote data centers, face network congestion and service latency issues, impacting user experience. Simultaneously, data privacy and security issues are becoming increasingly prominent, with frequent data breaches and misuse incidents posing serious risks to users. Against this backdrop, Mobile Edge Computing (MEC) has successfully achieved synergistic optimization of network load balancing and data security protection by deploying virtualized computing nodes at the base station level, creating a new architecture where "data remains within the domain, and computing power is near the terminal." However, due to limited edge resources, the growing and diverse needs of users cannot be met, especially in personalized services and intelligent applications, where the processing power of edge nodes remains insufficient. How to efficiently configure and manage these edge resources has become a critical issue that urgently needs to be addressed.

[0003] The patent document with publication number CN 113296909A, entitled "A Method for Optimizing Service Deployment in Mobile Edge Computing," describes a method that estimates the number of service application failures based on historical data. It models the service deployment problem as an optimization problem maximizing service deployment variables and service request scheduling variables. This optimization problem is then simplified to a univariate optimization problem with only service deployment variables, further transformed into a set function optimization problem with these variables, and finally, a robust algorithm is used to obtain the service deployment strategy, thus achieving optimized deployment. While this method can maximize the overall utility of deployed services, its simplification of a complex optimization problem into a univariate problem may overlook real-world conditions, affecting the optimization effect. Furthermore, regarding privacy protection, this method requires transmitting user data for failure prediction, increasing the risk of data leakage. Summary of the Invention

[0004] The summary of this invention introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. This summary is not intended to limit the key features and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0005] To address the technical challenges of limited computing resources on edge servers, diverse user needs, dynamically changing service requests, and privacy leaks in distributed environments, this invention provides a service deployment method and system based on federated learning. By combining users' personal information and service interaction data, it achieves efficient resource allocation and service deployment.

[0006] Firstly, a service deployment method based on federated learning is provided, including:

[0007] User personal information and service interaction data are obtained based on edge servers;

[0008] Users are clustered based on the personal information to obtain a set of clusters;

[0009] Based on the obtained cluster set, the available edge servers in each cluster are obtained. Through the constraints of the edge servers, multiple objective functions are designed, and the optimal solutions of the multiple objective functions are solved using a non-dominated sorting genetic algorithm to find the edge cloud servers in each cluster.

[0010] Deploy local prediction models based on available edge servers and edge cloud servers within each cluster, and obtain global federated prediction models through federated learning.

[0011] Based on the available edge servers within each cluster, the global federated prediction model is used to predict the future service request volume of each cluster, and services with increasing popularity are deployed to the available edge servers.

[0012] Furthermore, the process of clustering users to obtain a set of clusters includes:

[0013] Based on the aforementioned personal information, the feature values ​​of the personal information are mapped to... On the 3D coordinate axis, obtain the user's standard coordinates;

[0014] Based on the user's standard coordinates, the similarity between different users is calculated to construct a user similarity matrix;

[0015] The responsibility matrix and availability matrix are initialized based on the user similarity matrix, and the responsibility matrix and availability matrix are iteratively updated until the iteration threshold is reached to obtain the final responsibility matrix and final availability matrix;

[0016] All cluster centers are selected based on the final responsibility matrix and the final availability matrix. The remaining data points that are not cluster centers are assigned to the clusters to which the nearest cluster center belongs, thus obtaining a cluster set.

[0017] The personal information includes: age, gender, occupation, geographical location, device type, interests and preferences, and access time.

[0018] Furthermore, the initialization of the responsibility matrix and availability matrix refers to initializing the responsibility matrix and availability matrix as zero matrices with the same dimension as the user similarity matrix;

[0019] The user similarity matrix is ​​represented as follows:

[0020] ;

[0021] ;

[0022] in, For the user similarity matrix, For users and users The similarity between them For users Standardized user coordinates after processing For users Standardized user coordinates Distance threshold Total number of users The median of the user similarity matrix is ​​represented as: ;

[0023] The iterative update of the responsibility matrix and availability matrix is ​​calculated using the following formula:

[0024] ;

[0025] ;

[0026] in, This is a responsibility matrix, representing users. For users Suitability as a cluster center For users For users The degree of support for becoming a cluster center This is the availability matrix, representing users. For users The degree of support as a cluster center; At that time, i.e., user With users Not the same person, indicating the user For users The degree of support for becoming a cluster center, when At that time, i.e., user With users "Being the same person" indicates self-attribution and is used when only considering other users' relationships with users. The degree of support for becoming a cluster center;

[0027] The selection of all cluster centers refers to the process based on judgment. If the value is greater than the iteration threshold, then select the user. If one user is selected as the cluster center, another user will be used to try until all cluster centers are found.

[0028] The set of clusters is represented as follows: ,in, For the first Each cluster contains all users belonging to that cluster.

[0029] Furthermore, the mapping of personal information feature values ​​to On the 3D coordinate axes, the user's standard coordinates are obtained, including:

[0030] The personal characteristics in the aforementioned personal information are converted from non-numeric features to numeric feature values ​​through feature encoding, and then the feature values ​​are mapped to... The coordinates are plotted on the 3D coordinate axes and standardized to obtain the user's standard coordinates.

[0031] The user standard coordinates are represented as follows:

[0032] ;

[0033] in, For users User standard coordinates, For users The The eigenvalues ​​on the 3D coordinate axes after standardization;

[0034] The standardization process refers to deviation standardization, which scales the feature values ​​on each coordinate axis to the [0,1] interval. The calculation formula is as follows:

[0035] ;

[0036] in, For users The The eigenvalues ​​on the 3D coordinate axes are standardized eigenvalues. For the first The largest eigenvalue on the 3D coordinate axis For the first The smallest eigenvalue on the 3D coordinate axis For users The Eigenvalues ​​on the 3D coordinate axes.

[0037] Furthermore, the calculation of similarity between different users refers to the user similarity obtained by calculating the Euclidean distance between the standard coordinates of different users. The formula for calculating user similarity is:

[0038] ;

[0039] ;

[0040] in, Indicates user With users similarity, Represents the user based on the user's standard coordinates. With users The Euclidean distance between them For users The The eigenvalues ​​on the 3D coordinate axes are standardized eigenvalues. For users The The eigenvalues ​​on the 3D coordinate axes are standardized eigenvalues.

[0041] Furthermore, the constraints include: bandwidth constraints, power constraints, load balancing constraints, and signal-to-noise ratio constraints;

[0042] The bandwidth constraint refers to the fact that data transmission does not exceed the bandwidth limit of the edge server, expressed as: ,in, This is the data transmission objective function used to compute edge servers. Data transmission rate, For edge servers bandwidth;

[0043] The power constraint refers to the fact that the transmission power of the edge server cannot exceed the maximum power of the edge server, expressed as: ,in, For edge servers The transmission power, For edge servers Maximum power;

[0044] The load balancing constraint refers to the requirement that the load on the edge servers should be within the maximum load range to avoid overloading them, expressed as: ,in, For edge servers The load, For edge servers Maximum load;

[0045] The signal-to-noise ratio constraint refers to the requirement that the signal-to-noise ratio of the edge server must not be lower than a minimum signal-to-noise ratio to ensure that the signal quality meets the data transmission requirements. This constraint is expressed as: ;

[0046] in, For edge servers The transmission power, To improve channel gain for connecting terminal devices and edge servers, This refers to the signal-to-noise power generated during data transmission. This represents the lowest signal-to-noise ratio.

[0047] Furthermore, the plurality of objective functions include: a data transmission objective function, a computing power objective function, and a load balancing objective function;

[0048] The data transmission objective function is calculated using the following formula:

[0049] ;

[0050] in, The objective function for data transmission is... For edge servers The transmission power, To improve channel gain for connecting terminal devices and edge servers, For edge servers bandwidth, This refers to the signal-to-noise power generated during data transmission.

[0051] The objective function for computing power is calculated using the following formula:

[0052] ;

[0053] in, The objective function for computational capability is... For edge servers Number of CPU cores The clock speed of each CPU core, For edge servers computing power For edge servers memory size, These are the relevant parameters for CPU, GPU, and memory, respectively.

[0054] The load balancing objective function is calculated using the following formula:

[0055] ;

[0056] ;

[0057] in, Let the load balancing objective function be... This represents the average load on the edge servers. For edge servers The load, For the first Each cluster For clusters The number of edge servers.

[0058] Furthermore, the step of using a non-dominated sorting genetic algorithm to solve for the optimal solutions of the multiple objective functions and to find edge cloud servers within each cluster includes:

[0059] Based on the multiple objective functions designed, the performance data of the edge servers in each cluster are input into the multiple objective functions. Under the constraints, the Pareto optimal solution in each cluster is obtained using a non-dominated sorting genetic algorithm. The edge server corresponding to the Pareto optimal solution is used as the edge cloud server of that cluster.

[0060] Furthermore, the step of using a non-dominated sorting genetic algorithm to obtain the Pareto optimal solution for each cluster includes:

[0061] S11: Calculate the comprehensive score of each available edge server based on the performance data of available edge servers within each cluster;

[0062] S12: Based on the calculated comprehensive score, select the top 20% of available edge servers as the current population;

[0063] S13: Calculate the function values ​​of the multiple objective functions for each individual in the current population, perform non-dominated sorting to classify individual ranks, and assign fitness values ​​to each individual;

[0064] S14: Rank the individuals according to the superiority of the fitness values;

[0065] S15: Dynamically adjust the crossover rate and mutation rate, perform crossover and mutation operations on each individual in the current population to generate new candidate individuals, and introduce a local search strategy to fine-tune some individuals to obtain a newly generated offspring population.

[0066] S16: Merge the individuals in the current population with the newly generated offspring population, and retain the top N individuals as the next generation population based on non-dominated sorting and crowding distance selection.

[0067] S17: Repeat S13-S16 until the maximum number of iterations is reached, and select the individual with the highest fitness as the Pareto optimal solution.

[0068] Furthermore, the calculation of the comprehensive score for each available edge server in step S11 refers to a comprehensive score calculated based on the edge server's performance data, including computing power, energy consumption, latency, bandwidth, and storage capacity. The calculation formula is as follows:

[0069] ;

[0070] in, Edge servers Its computing power, energy consumption, latency, bandwidth, and storage capacity. This represents the maximum computing power of the edge servers within the cluster. The maximum bandwidth of the communication link. This refers to the storage capacity of the edge servers within the cluster. These are the weighted parameters for computing power, energy consumption, latency, bandwidth, and storage capacity, respectively.

[0071] The dynamic adjustment of crossover rate and mutation rate mentioned in step S15 means that if the fitness converges slowly, the mutation rate is increased, and if the convergence is fast, the mutation rate is decreased.

[0072] Furthermore, the deployment of the local prediction model includes:

[0073] The service interaction data obtained from the available edge servers and edge cloud servers within each cluster are converted into time series data and used as training data.

[0074] The Autoformer model is trained based on the training data, the mean squared error function is used as the loss function, and the Adam optimizer is used to update the parameters to obtain the local prediction model.

[0075] The service interaction data refers to all data entries containing the interaction services between each user and the edge server. These data entries include: service ID, service name, service access time, service request volume, service response time, service quality, data transmission volume, user identity information, and user behavior data.

[0076] Furthermore, the step of obtaining a global federated prediction model through federated learning includes:

[0077] S21: Based on the available edge servers within each cluster of the cluster set, upload the parameters of the deployed local prediction model to the cloud server;

[0078] S22: The edge cloud server uses a federated averaging algorithm to aggregate and update the parameters of the local prediction model uploaded by each available edge server to obtain a global model.

[0079] S23: The edge cloud server distributes the parameters of the global model to each available edge server in the same cluster. Each available edge server aggregates and updates the local prediction model. Based on the interaction data obtained by the available edge servers, cross-validation is used to train and fine-tune the updated local prediction model. The parameters of the trained and updated local prediction model are then uploaded to the edge cloud server.

[0080] S24: Repeat steps S22-S23 until the change in the loss value of the global model is less than the convergence threshold, and obtain the global federated prediction model.

[0081] Furthermore, the cross-validation method mentioned in step S23 refers to converting the local service interaction data obtained by the available edge servers in each cluster into time series data and using it as training data. A portion of the samples in the training data are then selected to test the local prediction model.

[0082] Furthermore, the federated averaging algorithm refers to aggregating the parameters of the local prediction models of each participant through weighted averaging, thereby enabling collaborative training of the global model while protecting data privacy. The calculation formula for updating the global model by weighted averaging is as follows:

[0083] ;

[0084] in, for The parameters of the global model on the edge cloud server during round training. for Participants in the round of training Parameters of the local prediction model As a participant The weight, As a participant The number of local data samples, The total number of data samples from all participants. For the number of training rounds, This represents the total number of training rounds.

[0085] Furthermore, the provision that the available edge servers within each cluster utilize the global federated prediction model to predict the future service request volume of each cluster, and deploy services with increasing popularity to the available edge servers, includes:

[0086] Each available edge server within the same cluster downloads the parameters of the global federated prediction model from the edge cloud server and loads them into the deployed local prediction model to update the model parameters;

[0087] Based on the local service interaction data obtained from available edge servers, the local prediction model with updated parameters is fine-tuned, and then the service request volume of each service in each cluster is predicted for future time.

[0088] The service popularity is calculated based on the predicted service request volume, and then sorted in descending order of the service popularity to obtain the future... The service popularity queue at any given moment;

[0089] Based on the service popularity queue and service heat range, identify services with popularity trends;

[0090] Subdivide the services with popular trends and find services with increasing popularity;

[0091] Deploy the services with popularity increments within each cluster to available edge servers.

[0092] Furthermore, the service popularity is calculated using the following formula:

[0093] ;

[0094] ;

[0095] in, For the future 24-hour service popularity For the present 24-hour service popularity For the present 24-hour service The number of service requests, For service collection, For the future 24-hour service The number of service requests, For the present From Moment to the Future 24-hour service Incremental service requests For the present 24-hour service The ratio of the service popularity ranking in the service popularity queue to the total number of service requests. These are the current service request percentage factor, trend strength factor, and ranking decay factor, respectively.

[0096] The service popularity queue refers to a service queue in which services are arranged in descending order of service popularity at the same time.

[0097] The service popularity range is represented as follows: ,in, The lower threshold for popular services For niche services, the upper limit threshold is set, and ;

[0098] The search for services with popular trends includes:

[0099] Get the future 24-hour service The ratio of the service popularity queue to the total number of service requests. ;

[0100] The Compare with the upper and lower thresholds of the service popularity range, if This indicates the service Its popularity is too low; it's a niche service with no trend. Remove and upload the logs to the cloud server; if This indicates the service This is a popular service, so the service will be... Backups are needed to handle sudden traffic spikes; if This indicates the service It has a popular trend;

[0101] The segmentation of services with popular trends, and the identification of services with increasing popularity, includes:

[0102] Compare and The size, if This indicates the service For services with increasing popularity, if This indicates the service For services that do not have a popular growth rate, and the service will be... Migrate to adjacent available edge servers to reduce load and resource consumption. For the present 24-hour service The ratio of the service popularity ranking in the service popularity queue to the total number of service requests.

[0103] Furthermore, the deployment of services with popularity increments within each cluster to available edge servers includes:

[0104] Each available edge server in each cluster checks whether the service with popularity increment has been deployed. If so, it continues to be deployed; otherwise, it downloads and deploys the service with popularity increment from the cloud server. The deployment process includes: loading the service image, distributing the configuration file, and initializing the runtime environment.

[0105] Secondly, a service deployment system based on federated learning is provided, including: a data acquisition module, a clustering module, an edge cloud server selection module, a global federated prediction model construction module, and a service deployment module;

[0106] The data acquisition module is used to acquire the user's personal information and service interaction data based on the edge server;

[0107] The clustering module is used to cluster users based on the personal information to obtain a set of clusters;

[0108] The edge cloud server selection module is used to obtain available edge servers in each cluster based on the acquired cluster set, design multiple objective functions through the constraints of the edge servers, and use a non-dominated sorting genetic algorithm to solve the optimal solutions of the multiple objective functions to find the edge cloud servers in each cluster.

[0109] The global federated prediction model construction module is used to deploy local prediction models based on the available edge servers and edge cloud servers in each cluster, and obtain a global federated prediction model through federated learning.

[0110] The service deployment module is used to predict the future service request volume of each cluster based on the available edge servers within each cluster using the global federated prediction model, and to deploy services with popularity increments to the available edge servers.

[0111] The beneficial effects of the invention are:

[0112] This invention provides a service deployment method based on federated learning, which assigns users with similar needs to the same cluster by performing affinity clustering on users, thereby reducing redundant computation and resource waste, and optimizing the distribution and deployment efficiency of services.

[0113] This invention provides a service deployment method based on federated learning, which employs multiple objective optimization methods and combines the computing power, data transmission efficiency, and load balancing of edge servers to rationally allocate service resources, avoid excessive concentration or resource waste, and improve the overall performance of the system.

[0114] This invention provides a service deployment method based on federated learning, which continuously optimizes the selection process of edge servers through the dynamic crossover and mutation strategy of a non-dominated sorting genetic algorithm, enhances the system's adaptability to different environments and loads, and ensures the system's robustness in complex network environments. Attached Figure Description

[0115] Figure 1This is a flowchart illustrating a service deployment method based on federated learning provided in one embodiment of the present invention.

[0116] Figure 2 This is a schematic diagram of a service deployment system based on federated learning provided in one embodiment of the present invention. Detailed Implementation

[0117] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0118] Many specific details are set forth in the following description in order to provide a full understanding of the invention, but the invention may also be practiced in other ways different from those described herein; obviously, the embodiments in the specification are only some embodiments of the invention, and not all embodiments.

[0119] The specific implementation of the technical solution of the present invention includes the following:

[0120] Example 1

[0121] The service deployment method based on federated learning involved in Embodiment 1 of the present invention includes:

[0122] The system acquires user personal information and service interaction data using edge servers; it clusters users based on the personal information to obtain cluster sets; based on the obtained cluster sets, it identifies available edge servers within each cluster, designs multiple objective functions based on the constraints of the edge servers, and uses a non-dominated sorting genetic algorithm to find the optimal solutions for the multiple objective functions, thereby identifying edge cloud servers within each cluster; it deploys local prediction models based on the available edge servers and the edge cloud servers within each cluster, and obtains a global federated prediction model through federated learning; based on the available edge servers within each cluster, it uses the global federated prediction model to predict the future service request volume of each cluster, and deploys services with increasing popularity to the available edge servers.

[0123] Specifically, Figure 1 The diagram illustrates a service deployment method based on federated learning in Embodiment 1 of the application, including:

[0124] S1. Obtain user personal information and service interaction data based on edge servers.

[0125] The personal information mentioned in step S1 includes: age, gender, occupation, geographical location, device type, interests and preferences, and access time.

[0126] The service interaction data mentioned in step S1 refers to all data entries containing the interaction services between each user and the edge server. The data entries include: service ID, service name, service access time, service request volume, service response time, service quality, data transmission volume, user identity information, and user behavior data.

[0127] For example, edge servers acquire personal information and service interaction data of all users within a region via the network. This includes: firstly, accessing the application programming interface (API) of user devices to extract basic personal information for each user, including user ID, age, gender, occupation, geographic address, device type, interests, and access time. Simultaneously, service interaction data between users and various edge servers is recorded. Data entries include, but are not limited to: service ID (e.g., "Service A"), service name (e.g., "Video Streaming Service"), service access time (e.g., "2024-10-20 10:30:00"), service access frequency (e.g., "5 times / day"), service response time (e.g., "200ms"), service quality (e.g., "4.5 / 5"), data transfer volume (e.g., "100MB"), and user identity information and user behavior data (e.g., "click rate"). All collected data is stored in a backend database for subsequent analysis and processing.

[0128] S2. Cluster users based on the personal information to obtain a set of clusters.

[0129] The step S2, which involves clustering users to obtain a set of clusters, includes:

[0130] Based on the aforementioned personal information, the feature values ​​of the personal information are mapped to... On the 3D coordinate axis, obtain the user's standard coordinates;

[0131] Based on the user's standard coordinates, the similarity between different users is calculated to construct a user similarity matrix;

[0132] The responsibility matrix and availability matrix are initialized based on the user similarity matrix, and the responsibility matrix and availability matrix are iteratively updated until the iteration threshold is reached to obtain the final responsibility matrix and final availability matrix;

[0133] All cluster centers are selected based on the final responsibility matrix and the final availability matrix. The remaining data points that are not cluster centers are assigned to the clusters to which the nearest cluster center belongs, thus obtaining a cluster set.

[0134] The personal information refers to information that includes a user's personal characteristics, including: age, gender, occupation, geographical location, device type, interests and preferences, and access time.

[0135] Specifically, the initialization of the responsibility matrix and availability matrix refers to initializing the responsibility matrix and availability matrix as zero matrices with the same dimension as the user similarity matrix;

[0136] The user similarity matrix is ​​represented as follows:

[0137] ;

[0138] ;

[0139] in, For the user similarity matrix, For users and users The similarity between them For users Standardized user coordinates For users Standardized user coordinates Distance threshold Total number of users The median of the user similarity matrix is ​​represented as: ;

[0140] The iterative update of the responsibility matrix and availability matrix is ​​calculated using the following formula:

[0141] ;

[0142] ;

[0143] in, This is a responsibility matrix, representing users. For users Suitability as a cluster center For users For users The degree of support for becoming a cluster center This is the availability matrix, representing users. For users The degree of support as a cluster center; At that time, i.e., user With users Not the same person, indicating the user For users The degree of support for becoming a cluster center, when At that time, i.e., user With users "Being the same person" indicates self-attribution and is used when only considering other users' relationships with users. The degree of support for becoming a cluster center;

[0144] The selection of all cluster centers refers to the process based on judgment. If the value is greater than the iteration threshold, then select the user. If one user is selected as the cluster center, another user will be used to try until all cluster centers are found.

[0145] Represented as: ,in, For the first Each cluster contains all users belonging to that cluster.

[0146] Specifically, the mapping of personal information feature values ​​to On the 3D coordinate axes, the user's standard coordinates are obtained, including:

[0147] The personal characteristics in the aforementioned personal information are converted from non-numeric features to numeric feature values ​​through feature encoding, and then the feature values ​​are mapped to... The coordinates are plotted on the 3D coordinate axes and standardized to obtain the user's standard coordinates.

[0148] The user standard coordinates are represented as follows:

[0149] ;

[0150] in, For users User standard coordinates, For users The The eigenvalues ​​on the 3D coordinate axes after standardization;

[0151] The standardization process refers to deviation standardization, which scales the feature values ​​on each coordinate axis to the [0,1] interval. The calculation formula is as follows:

[0152] ;

[0153] in, For users The The eigenvalues ​​on the 3D coordinate axes are standardized eigenvalues. For the first The largest eigenvalue on the 3D coordinate axis For the first The smallest eigenvalue on the 3D coordinate axis For users The Eigenvalues ​​on the 3D coordinate axes.

[0154] Specifically, calculating the similarity between different users refers to obtaining user similarity by calculating the Euclidean distance between the standard coordinates of different users. The formula for calculating user similarity is:

[0155] ;

[0156] ;

[0157] in, Indicates user With users similarity, Represents the user based on the user's standard coordinates. With users The Euclidean distance between them For users The The eigenvalues ​​on the 3D coordinate axes are standardized eigenvalues. For users The The eigenvalues ​​on the 3D coordinate axes are standardized eigenvalues.

[0158] S3. Based on the acquired cluster set, obtain the available edge servers within each cluster. Through the constraints of the edge servers, design multiple objective functions, and use a non-dominated sorting genetic algorithm to solve the optimal solutions of the multiple objective functions to find the edge cloud servers within each cluster.

[0159] The constraints mentioned in step S3 include: bandwidth constraints, power constraints, load balancing constraints, and signal-to-noise ratio constraints;

[0160] The bandwidth constraint refers to the fact that data transmission does not exceed the bandwidth limit of the edge server, expressed as: ,in, This is the data transmission objective function used to compute edge servers. Data transmission rate, For edge servers bandwidth;

[0161] The power constraint refers to the fact that the transmission power of the edge server cannot exceed the maximum power of the edge server, expressed as: ,in, For edge servers The transmission power, For edge servers Maximum power;

[0162] The load balancing constraint refers to the requirement that the load on the edge servers should be within the maximum load range to avoid overloading them, expressed as: ,in, For edge servers The load, For edge servers Maximum load;

[0163] The signal-to-noise ratio constraint refers to the requirement that the signal-to-noise ratio of the edge server must not be lower than a minimum signal-to-noise ratio to ensure that the signal quality meets the data transmission requirements. This constraint is expressed as: ;

[0164] in, For edge servers The transmission power, To improve channel gain for connecting terminal devices and edge servers, This refers to the signal-to-noise power generated during data transmission. This represents the lowest signal-to-noise ratio.

[0165] The multiple objective functions mentioned in step S3 include: a data transmission objective function, a computing power objective function, and a load balancing objective function;

[0166] The data transmission objective function is calculated using the following formula:

[0167] ;

[0168] in, The objective function for data transmission is... For edge servers The transmission power, To improve channel gain for connecting terminal devices and edge servers, For edge servers bandwidth, This refers to the signal-to-noise power generated during data transmission.

[0169] The objective function for computing power is calculated using the following formula:

[0170] ;

[0171] in, The objective function for computational capability is... For edge servers Number of CPU cores The clock speed of each CPU core, For edge servers computing power For edge servers memory size, These are the relevant parameters for CPU, GPU, and memory, respectively.

[0172] The load balancing objective function is calculated using the following formula:

[0173] ;

[0174] ;

[0175] in, Let the load balancing objective function be... This represents the average load on the edge servers. For edge servers The load, For the first Each cluster For clusters The number of edge servers.

[0176] Step S3, which involves using a non-dominated sorting genetic algorithm to find the optimal solutions to the multiple objective functions and locate edge cloud servers within each cluster, includes:

[0177] Based on the multiple objective functions designed, the performance data of the edge servers in each cluster are input into the multiple objective functions. Under the constraints, the Pareto optimal solution in each cluster is obtained using a non-dominated sorting genetic algorithm. The edge server corresponding to the Pareto optimal solution is used as the edge cloud server of that cluster.

[0178] Specifically, the method of using a non-dominated sorting genetic algorithm to obtain the Pareto optimal solution for each cluster includes:

[0179] S11: Calculate the comprehensive score of each available edge server based on the performance data of available edge servers within each cluster;

[0180] S12: Based on the calculated comprehensive score, select the top 20% of available edge servers as the current population;

[0181] S13: Calculate the function values ​​of the multiple objective functions for each individual in the current population, perform non-dominated sorting to classify individual ranks, and assign fitness values ​​to each individual;

[0182] S14: Rank the individuals according to the superiority of the fitness values;

[0183] S15: Dynamically adjust the crossover rate and mutation rate, perform crossover and mutation operations on each individual in the current population to generate new candidate individuals, and introduce a local search strategy to fine-tune some individuals to obtain a newly generated offspring population.

[0184] S16: Merge the individuals in the current population with the newly generated offspring population, and retain the top N individuals as the next generation population based on non-dominated sorting and crowding distance selection.

[0185] S17: Repeat S13-S16 until the maximum number of iterations is reached, and select the individual with the highest fitness as the Pareto optimal solution.

[0186] Furthermore, the calculation of the comprehensive score for each available edge server in step S11 refers to a comprehensive score calculated based on the edge server's performance data, including computing power, energy consumption, latency, bandwidth, and storage capacity. The calculation formula is as follows:

[0187] ;

[0188] in, Edge servers Its computing power, energy consumption, latency, bandwidth, and storage capacity. This represents the maximum computing power of the edge servers within the cluster. The maximum bandwidth of the communication link. This refers to the storage capacity of the edge servers within the cluster. These are the weighted parameters for computing power, energy consumption, latency, bandwidth, and storage capacity, respectively.

[0189] The dynamic adjustment of crossover rate and mutation rate mentioned in step S15 means that if the fitness converges slowly, the mutation rate is increased, and if the convergence is fast, the mutation rate is decreased.

[0190] S4. Deploy local prediction models based on available edge servers and edge cloud servers within each cluster, and obtain a global federated prediction model through federated learning.

[0191] In step S4, deploying the local prediction model includes:

[0192] The service interaction data obtained from the available edge servers and edge cloud servers within each cluster are converted into time series data and used as training data.

[0193] The Autoformer model is trained based on the training data, the mean squared error function is used as the loss function, and the Adam optimizer is used to update the parameters to obtain the local prediction model.

[0194] The service interaction data refers to all data entries containing the interaction services between each user and the edge server. These data entries include: service ID, service name, service access time, service request volume, service response time, service quality, data transmission volume, user identity information, and user behavior data.

[0195] In step S4, obtaining the global federated prediction model through federated learning includes:

[0196] S21: Based on the available edge servers within each cluster of the cluster set, upload the parameters of the deployed local prediction model to the cloud server;

[0197] S22: The edge cloud server uses a federated averaging algorithm to aggregate and update the parameters of the local prediction model uploaded by each available edge server to obtain a global model.

[0198] S23: The edge cloud server distributes the parameters of the global model to each available edge server in the same cluster. Each available edge server aggregates and updates the local prediction model. Based on the interaction data obtained by the available edge services, the updated local prediction model is trained using cross-validation. The parameters of the trained and updated local prediction model are then uploaded to the edge cloud server.

[0199] S24: Repeat steps S22-S23 until the change in the loss value of the global model is less than the convergence threshold, and obtain the global federated prediction model.

[0200] Specifically, the cross-validation method mentioned in step S23 refers to converting the local service interaction data obtained from the available edge servers in each cluster into time series data and using it as training data. A portion of the samples in the training data are then selected to test the local prediction model.

[0201] Specifically, the federated averaging algorithm refers to aggregating the parameters of the local prediction models of each participant through weighted averaging, thereby enabling collaborative training of the global model while protecting data privacy. The calculation formula for updating the global model by weighted averaging is as follows:

[0202] ;

[0203] in, for The parameters of the global model on the edge cloud server during round training. for Participants in the round of training Parameters of the local prediction model As a participant The weight, As a participant The number of local data samples, The total number of data samples from all participants. For the number of training rounds, This represents the total number of training rounds.

[0204] S5. Based on the available edge servers in each cluster, the global federated prediction model is used to predict the future service request volume of each cluster, and services with popularity increments are deployed to the available edge servers.

[0205] Specifically, step S5, which involves using the global federated prediction model to predict the future service request volume of each cluster based on the available edge servers within each cluster, and deploying services with increasing popularity to the available edge servers, includes:

[0206] Each available edge server within the same cluster downloads the parameters of the global federated prediction model from the edge cloud server and loads them into the deployed local prediction model to update the model parameters;

[0207] Based on the local service interaction data obtained from available edge servers, the local prediction model with updated parameters is fine-tuned, and then the service request volume of each service in each cluster is predicted for future time.

[0208] The service popularity is calculated based on the predicted service request volume, and then sorted in descending order of the service popularity to obtain the future... The service popularity queue at any given moment;

[0209] Based on the service popularity queue and service heat range, identify services with popularity trends;

[0210] Further segment the services with popular trends to find services with increasing popularity;

[0211] Deploy the services with popularity increments within each cluster to available edge servers.

[0212] Specifically, the service popularity is calculated using the following formula:

[0213] ;

[0214] ;

[0215] in, For the future 24-hour service popularity For the present 24-hour service popularity For the present 24-hour service The number of service requests, For service collection, For the future 24-hour service The number of service requests, For the present From Moment to the Future 24-hour service Incremental service requests For the present 24-hour service The ratio of the service popularity ranking in the service popularity queue to the total number of service requests. These are the current request percentage factor, trend strength factor, and ranking decay factor, respectively.

[0216] The service popularity queue refers to a service queue in which services are arranged in descending order of service popularity at the same time.

[0217] The service popularity range is represented as follows: ,in, The lower threshold for popular services For niche services, the upper limit threshold is set, and ;

[0218] The search for services with popular trends includes:

[0219] Get the future 24-hour service The ratio of the service popularity queue to the total number of service requests. ;

[0220] The Compare with the upper and lower thresholds of the service popularity range, if This indicates the service Its popularity is too low; it's a niche service with no trend. Remove and upload the logs to the cloud server; if This indicates the service This is a popular service, so the service will be... Backups are needed to handle sudden traffic spikes; if This indicates the service It has a popular trend;

[0221] The segmentation of services with popular trends, and the identification of services with increasing popularity, includes:

[0222] Compare and The size, if This indicates the service For services with increasing popularity, if This indicates the service For services that do not have a popular growth rate, and the service will be... Migrate to adjacent available edge servers to reduce load and resource consumption. For the present 24-hour service The ratio of the service popularity ranking in the service popularity queue to the total number of service requests.

[0223] Specifically, deploying the services with popularity increments within each cluster to available edge servers includes:

[0224] Each available edge server in each cluster checks whether the service with popularity increment has been deployed. If so, it continues to be deployed; otherwise, it downloads and deploys the service with popularity increment from the cloud server. The deployment process includes: loading the service image, distributing the configuration file, and initializing the runtime environment.

[0225] Example 2

[0226] like Figure 2 As shown in Embodiment 2 of this application, a service deployment system based on federated learning includes: a data acquisition module, a clustering module, an edge cloud server selection module, a global federated prediction model construction module, and a service deployment module.

[0227] Specifically, the data acquisition module is used to acquire user personal information and service interaction data based on the edge server;

[0228] The clustering module is used to cluster users based on the personal information to obtain a set of clusters;

[0229] The edge cloud server selection module is used to obtain available edge servers in each cluster based on the acquired cluster set, design multiple objective functions through the constraints of the edge servers, and use a non-dominated sorting genetic algorithm to solve the optimal solutions of the multiple objective functions to find the edge cloud servers in each cluster.

[0230] The global federated prediction model construction module is used to deploy local prediction models based on available edge servers and edge cloud servers in each cluster, and obtain a global federated prediction model through federated learning.

[0231] The service deployment module is used to predict the future service request volume of each cluster based on the available edge servers within each cluster using the global federated prediction model, and to deploy services with popularity increments to the available edge servers.

[0232] The specific implementation method of this embodiment is the same as that of Embodiment 1, and will not be repeated here. Please refer to the description of Embodiment 1 for details.

[0233] Those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments.

[0234] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A service deployment method based on federated learning, characterized in that, include: User personal information and service interaction data are obtained based on edge servers; Based on the aforementioned personal information, the feature values ​​of the personal information are mapped to... On the 3D coordinate axis, obtain the user's standard coordinates; Based on the user's standard coordinates, the similarity between different users is calculated to construct a user similarity matrix; The responsibility matrix and availability matrix are initialized based on the user similarity matrix, and the responsibility matrix and availability matrix are iteratively updated until the iteration threshold is reached to obtain the final responsibility matrix and final availability matrix; Based on the final responsibility matrix and the final availability matrix, cluster centers are selected, and data points that are not at cluster centers are assigned to the clusters to which the nearest cluster center belongs, thus obtaining a cluster set. Based on the obtained cluster set, the available edge servers within each cluster are obtained, and multiple objective functions are designed based on the constraints of the edge servers. Based on the constraints of the edge servers, the performance data of the edge servers are input into multiple objective functions to calculate a comprehensive score. The top 20% of available edge servers are selected as the initial population. A non-dominated sorting and crowding distance selection mechanism is adopted, combined with dynamically adjusted crossover and mutation rates, and a local search strategy is introduced to iteratively solve the Pareto optimal solution. The edge server corresponding to the Pareto optimal solution is determined as the edge cloud server of this cluster. A local prediction model is obtained based on the local training parameters of the Autoformer model, and the local prediction model is deployed to the available edge servers and edge cloud servers in each cluster. The parameters of the deployed local prediction model are uploaded to the corresponding edge cloud server. The edge cloud server uses a federated averaging algorithm to aggregate the parameters, update the global model and distribute it. The aggregation and distribution are repeated until the model loss value converges to obtain the global federated prediction model. The parameters of the global federated prediction model are downloaded to available edge servers to predict service request volume, calculate service popularity, generate a service popularity queue, identify services with popularity trends based on service popularity intervals, further compare request increments to identify services with popularity increments, and download and deploy the services with popularity increments from the cloud server to available edge servers. This includes: each cluster of available edge servers checking whether the services with popularity increments have been deployed; if so, maintaining them; otherwise, downloading and deploying the services with popularity increments from the cloud server. The service popularity queue refers to a service queue in which services are arranged in descending order of service popularity at the same time. The service popularity range is represented as follows: ,in, The lower threshold for popular services For niche services, the upper limit threshold is set, and .

2. The service deployment method based on federated learning according to claim 1, characterized in that, The personal information includes: age, gender, occupation, geographical location, device type, interests and preferences, and access time.

3. The service deployment method based on federated learning according to claim 1, characterized in that, The mapping of personal information feature values ​​to On the 3D coordinate axes, the user's standard coordinates are obtained, including: The personal characteristics in the aforementioned personal information are converted from non-numeric features to numeric feature values ​​through feature encoding, and then the feature values ​​are mapped to... The coordinates are plotted on the 3D coordinate axes and standardized to obtain the user's standard coordinates. The user standard coordinates are represented as follows: ； in, For users User standard coordinates, For users The The eigenvalues ​​on the 3D coordinate axes after standardization; The standardization process refers to deviation standardization, which scales the feature values ​​on each coordinate axis to the [0,1] interval. The calculation formula is as follows: ； in, For users The The eigenvalues ​​on the 3D coordinate axes are standardized eigenvalues. For the first The largest eigenvalue on the 3D coordinate axis For the first The smallest eigenvalue on the 3D coordinate axis For users The Eigenvalues ​​on the 3D coordinate axes.

4. The service deployment method based on federated learning according to claim 1, characterized in that, The calculation of similarity between different users refers to the user similarity obtained by calculating the Euclidean distance between the standard coordinates of different users. The formula for calculating user similarity is: ； ； in, Indicates user With users similarity, Represents the user based on the user's standard coordinates. With users The Euclidean distance between them For users The The eigenvalues ​​on the 3D coordinate axes are standardized eigenvalues. For users The The eigenvalues ​​on the 3D coordinate axes after standardization; The user similarity matrix is ​​represented as follows: ； ； in, For the user similarity matrix, For users and users The similarity between them For users Standardized user coordinates after processing For users Standardized user coordinates Distance threshold Total number of users The median of the user similarity matrix is ​​represented as: ; The iterative update of the responsibility matrix and availability matrix is ​​calculated using the following formula: ； ； in, This is a responsibility matrix, representing users. For users Suitability as a cluster center For users For users The degree of support for becoming a cluster center This is the availability matrix, representing users. For users The degree of support as a cluster center; At that time, i.e., user With users Not the same person, indicating the user For users The degree of support for becoming a cluster center, when At that time, i.e., user With users "Being the same person" indicates self-attribution and is used when only considering other users' relationships with users. The degree of support for becoming the center of a cluster.

5. A service deployment method based on federated learning according to claim 1, characterized in that, The constraints include: bandwidth constraints, power constraints, load balancing constraints, and signal-to-noise ratio constraints. The bandwidth constraint refers to the fact that data transmission does not exceed the bandwidth limit of the edge server, expressed as: ,in, This is the data transmission objective function used to compute edge servers. Data transmission rate, For edge servers bandwidth; The power constraint refers to the fact that the transmission power of the edge server cannot exceed the maximum power of the edge server, expressed as: ,in, For edge servers The transmission power, For edge servers Maximum power; The load balancing constraint refers to the requirement that the load on the edge servers should be within the maximum load range to avoid overloading them, expressed as: ,in, For edge servers The load, For edge servers Maximum load; The signal-to-noise ratio constraint refers to the requirement that the signal-to-noise ratio of the edge server must not be lower than a minimum signal-to-noise ratio to ensure that the signal quality meets the data transmission requirements. This constraint is expressed as: ;in, For edge servers The transmission power, To improve channel gain for connecting terminal devices and edge servers, This refers to the signal-to-noise power generated during data transmission. This represents the lowest signal-to-noise ratio.

6. The service deployment method based on federated learning according to claim 1, characterized in that, The multiple objective functions include: a data transmission objective function, a computing power objective function, and a load balancing objective function; The data transmission objective function is calculated using the following formula: ； in, The objective function for data transmission is... For edge servers The transmission power, To connect terminal devices and edge servers Channel gain, For edge servers bandwidth, This refers to the signal-to-noise power generated during data transmission. The objective function for computing power is calculated using the following formula: ； in, The objective function for computational capability is... For edge servers Number of CPU cores The clock speed of each CPU core, For edge servers computing power For edge servers memory size, These are the relevant parameters for CPU, GPU, and memory, respectively. The load balancing objective function is calculated using the following formula: ； ； in, Let the load balancing objective function be... This represents the average load on the edge servers. For edge servers The load, For the first Each cluster For clusters The number of edge servers.

7. A service deployment method based on federated learning according to claim 1, characterized in that, The process involves inputting the performance data of edge servers into multiple objective functions to calculate a comprehensive score, selecting the top 20% of usable edge servers as the initial population, employing non-dominated sorting and crowding distance selection mechanisms, combined with dynamically adjusted crossover and mutation rates, and introducing a local search strategy to iteratively solve for the Pareto optimal solution, including: S11: Calculate the comprehensive score of each available edge server based on the performance data of available edge servers within each cluster; S12: Based on the calculated comprehensive score, select the top 20% of available edge servers as the current population; S13: Calculate the function values ​​of the multiple objective functions for each individual in the current population, perform non-dominated sorting to classify individual ranks, and assign fitness values ​​to each individual; S14: Rank the individuals according to the superiority of the fitness values; S15: Dynamically adjust the crossover rate and mutation rate, perform crossover and mutation operations on each individual in the current population to generate new candidate individuals, and introduce a local search strategy to fine-tune some individuals to obtain a newly generated offspring population. S16: Merge the individuals in the current population with the newly generated offspring population, and based on non-dominated sorting and crowding distance selection, retain the previous... Each individual serves as the next generation of the population; S17: Repeat S13-S16 until the maximum number of iterations is reached, and select the individual with the highest fitness as the Pareto optimal solution; The formula for calculating the comprehensive score is as follows: ； in, Edge servers Its computing power, energy consumption, latency, bandwidth, and storage capacity. This represents the maximum computing power of the edge servers within the cluster. The maximum bandwidth of the communication link. This refers to the storage capacity of the edge servers within the cluster. These are the weighted parameters for computing power, energy consumption, latency, bandwidth, and storage capacity, respectively.

8. A service deployment method based on federated learning according to claim 1, characterized in that, The process of obtaining a local prediction model based on local training parameters of the Autoformer model includes: The service interaction data obtained from the available edge servers and edge cloud servers within each cluster are converted into time series data and used as training data. The Autoformer model is trained based on the training data, the mean squared error function is used as the loss function, and the Adam optimizer is used to update the parameters to obtain the local prediction model. The service interaction data refers to all data entries containing the interaction services between each user and the edge server. These data entries include: service ID, service name, service access time, service request volume, service response time, service quality, data transmission volume, user identity information, and user behavior data.

9. A service deployment method based on federated learning according to claim 1, characterized in that, The process involves uploading the deployed local prediction model parameters to the corresponding edge cloud server. The edge cloud server then aggregates the parameters using a federated averaging algorithm, updates the global model, and distributes it. This aggregation and distribution process is repeated until the model loss value converges, resulting in a global federated prediction model, including: S21: Based on the available edge servers within each cluster in the cluster set, upload the parameters of the deployed local prediction model to the edge cloud server; S22: The edge cloud server uses a federated averaging algorithm to aggregate and update the parameters of the local prediction model uploaded by each available edge server to obtain a global model. S23: The edge cloud server distributes the parameters of the global model to each available edge server in the same cluster. Each available edge server aggregates and updates the local prediction model. Based on the interaction data obtained by the available edge services, the updated local prediction model is trained using cross-validation. The parameters of the trained and updated local prediction model are then uploaded to the edge cloud server. S24: Repeat steps S22-S23 until the change in the loss value of the global model is less than the convergence threshold, and obtain the global federated prediction model.

10. A service deployment method based on federated learning according to claim 9, characterized in that, The federated averaging algorithm refers to aggregating the parameters of the local prediction models of each participant through weighted averaging, thereby enabling collaborative training of the global model while protecting data privacy. The formula for calculating the weighted average update of the global model is as follows: ； in, for The parameters of the global model on the edge cloud server during round training. for Participants in the round of training Parameters of the local prediction model As a participant The weight, As a participant The number of local data samples, The total number of data samples from all participants. For the number of training rounds, This represents the total number of training rounds.

11. A service deployment method based on federated learning according to claim 1, characterized in that, The process of downloading the parameters of the global federated prediction model to available edge servers, predicting service request volume, calculating service popularity, generating a service popularity queue, identifying services with popularity trends based on service popularity intervals, further comparing request increments to identify services with popularity increments, and downloading and deploying the services with popularity increments from the cloud server to available edge servers includes: Each available edge server within the same cluster downloads the parameters of the global federated prediction model from the edge cloud server and loads them into the deployed local prediction model to update the model parameters; Based on the local service interaction data obtained from available edge servers, the local prediction model with updated parameters is fine-tuned, and then the service request volume of each service in each cluster is predicted for future time. The service popularity is calculated based on the predicted service request volume, and then sorted in descending order of the service popularity to obtain the future... The service popularity queue at any given moment; Based on the service popularity queue and service heat range, identify services with popularity trends; Subdivide the services with popular trends and find services with increasing popularity; Deploy the services with popularity increments within each cluster to available edge servers.

12. A service deployment method based on federated learning according to claim 11, characterized in that, The service popularity is calculated using the following formula: ； ； in, For the future 24-hour service popularity For the present 24-hour service popularity For the present 24-hour service The number of service requests, For service collection, For the future 24-hour service The number of service requests, For the present From Moment to the Future 24-hour service Incremental service requests For the present 24-hour service The ratio of the service popularity queue order to the total number of service requests. These are the current request percentage factor, trend strength factor, and ranking decay factor, respectively. The search for services with popular trends includes: Get the future 24-hour service The ratio of the service popularity queue to the total number of service requests. ; The Compare with the upper and lower thresholds of the service popularity range, if This indicates the service Its popularity is too low; it's a niche service with no trend. Remove and upload the logs to the cloud server; if This indicates the service This is a popular service, so the service will be... Backups are needed to handle sudden traffic spikes; if This indicates the service It has a popular trend; The segmentation of services with popular trends, and the identification of services with increasing popularity, includes: Compare and The size, if This indicates the service For services with increasing popularity, if This indicates the service For services that do not have a popular growth rate, and the service will be... Migrate to adjacent available edge servers to reduce load and resource consumption. For the present 24-hour service The ratio of the service popularity ranking in the service popularity queue to the total number of service requests.

13. A service deployment method based on federated learning according to claim 11, characterized in that, The step of deploying the services with popularity increments within each cluster to available edge servers includes: The process of downloading and deploying the service with popular increments from the cloud server includes: loading the service image, distributing the configuration file, and initializing the runtime environment.

14. A service deployment system based on federated learning, characterized in that, Implementing a service deployment method based on federated learning according to any one of claims 1-13 specifically includes: The data acquisition module is used to acquire users' personal information and service interaction data based on the edge server; The clustering module is used to map the feature values ​​of the personal information to... On the 3D coordinate axis, the user's standard coordinates are obtained; based on the user's standard coordinates, the similarity between different users is calculated to construct a user similarity matrix; based on the user similarity matrix, the responsibility matrix and availability matrix are initialized, and the responsibility matrix and availability matrix are iteratively updated until an iteration threshold is reached to obtain the final responsibility matrix and final availability matrix; according to the final responsibility matrix and final availability matrix, cluster centers are selected, and data points that are not at the cluster center are assigned to the cluster to which the nearest cluster center belongs, thus obtaining a cluster set; The edge cloud server selection module is used to obtain available edge servers within each cluster based on the acquired cluster set, design multiple objective functions based on the constraints of the edge servers, input the performance data of the edge servers into the multiple objective functions to calculate a comprehensive score, select the top 20% of available edge servers as the initial population, adopt a non-dominated sorting and crowding distance selection mechanism, combine dynamically adjusted crossover and mutation rates, and introduce a local search strategy to iteratively solve for the Pareto optimal solution, and determine the edge server corresponding to the Pareto optimal solution as the edge cloud server of that cluster. The global federated prediction model building module is used to obtain a local prediction model based on the local training parameters of the Autoformer model, and deploy the local prediction model to the available edge servers and edge cloud servers in each cluster; the parameters of the deployed local prediction model are uploaded to the corresponding edge cloud server, the edge cloud server uses the federated averaging algorithm to aggregate the parameters, update the global model and distribute it, and repeat the aggregation and distribution until the model loss value converges to obtain the global federated prediction model. The service deployment module is used to download the parameters of the global federated prediction model to available edge servers, predict the service request volume, calculate the service popularity, generate a service popularity queue, identify services with popularity trends based on service popularity intervals, further compare request increments to identify services with popularity increments, and download and deploy the services with popularity increments from the cloud server to available edge servers. This includes: each cluster of available edge servers checking whether the services with popularity increments have already been deployed; if so, maintaining the deployment; otherwise, downloading and deploying the services with popularity increments from the cloud server. The service popularity queue refers to a service queue in which services are arranged in descending order of service popularity at the same time. The service popularity range is represented as follows: ,in, The lower threshold for popular services For niche services, the upper limit threshold is set, and .

Images