Resource collaboration architecture based on fusion of edge computing and federated learning
By integrating edge computing and federated learning into a resource collaboration architecture, the problems of low scheduling efficiency, insufficient privacy protection, and poor system scalability in cross-entity resource collaboration are solved. This enables efficient and secure resource scheduling and model training, improving the collaboration efficiency and security in the sharing economy scenario.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XUZHOU MEDICAL UNIVERSITY
- Filing Date
- 2026-01-06
- Publication Date
- 2026-05-19
AI Technical Summary
Existing technologies suffer from low scheduling efficiency, insufficient privacy protection, and poor system scalability in cross-entity resource collaboration. This is especially true in the sharing economy scenario, where it leads to high data transmission latency, high risk of privacy leakage, complex node access configuration, difficulty in reusing models across scenarios, and high transformation costs.
It adopts a resource collaboration architecture based on the integration of edge computing and federated learning. Through lightweight deployment, dynamic node management, multi-level privacy protection and adaptive optimization strategies, it can achieve local data preprocessing, sensitive data encryption, anomaly detection and isolation, and dynamically adjust training parameters and task allocation to improve collaboration efficiency and security.
It improved resource collaboration efficiency by 30%, shortened system response time by 50%, increased model training accuracy by 25%, reduced the probability of potential attacks by 70%, reduced data transmission volume by 40%, shortened the R&D cycle by 35%, and ensured the system's flexibility and scalability.
Smart Images
Figure CN122064472A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of information technology, specifically to a resource collaboration architecture based on the integration of edge computing and federated learning. Background Technology
[0002] With the development of the sharing economy, scenarios involving cross-entity resource collaboration, such as shared wheelchairs and cars, are increasing, raising the requirements for scheduling efficiency, privacy protection, and system scalability. Currently, cross-entity resource collaboration relies on centralized or simple distributed architectures, which have significant limitations.
[0003] In terms of scheduling efficiency, existing technologies mostly rely on centralized servers to process data and train models. The dispersed nature of resources across entities leads to high data transmission latency and makes real-time response difficult due to bandwidth limitations. Furthermore, the lack of dynamic node management means that insufficient capacity or poor network conditions in some nodes can drag down the overall process, resulting in low efficiency. Regarding data privacy, existing technologies require data to be transmitted or centrally stored between entities, posing a high risk of leakage. While encryption methods exist, the strategies are often simplistic, either resulting in excessive encryption that impacts efficiency or insufficient encryption strength. Exposure of raw data or improper parameter usage during model training can also indirectly leak privacy, making it difficult to meet regulatory requirements. In terms of system scalability, traditional architectures have complex node access configurations, making plug-and-play functionality difficult. Models for different shared scenarios are difficult to reuse, requiring repeated development, which is challenging and costly, hindering cross-scenario collaborative development.
[0004] Therefore, improving collaborative scheduling efficiency, strengthening privacy protection, and enhancing architectural flexibility and scalability are urgent issues that need to be addressed in fields such as the sharing economy. Summary of the Invention
[0005] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a resource collaboration architecture based on the integration of edge computing and federated learning, which solves the problems of high difficulty and cost in transformation, and restricts cross-scenario collaborative development.
[0006] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a resource collaboration architecture based on the fusion of edge computing and federated learning, comprising multiple edge computing units, a central coordination node, and a data communication network; Each of the edge computing units connects to the data communication network through a lightweight deployment component and completes data preprocessing and feature extraction tasks locally. The central coordination node is responsible for the training and updating of the global model, and uses a dynamic node management module to adjust the number of edge computing units participating in collaborative training and the task allocation strategy in real time. The edge computing unit is equipped with an edge caching optimization module, which is used to pre-cach the calculation results of popular data features to reduce real-time processing latency and improve the response speed of resource scheduling decisions.
[0007] Preferably, under the federated learning framework, when the central coordination node updates the global model through the parameter aggregation mechanism, it uses an adaptive optimization module to dynamically adjust the training parameters to adapt to the differences in data feature distribution among the nodes. The local training process combines an adaptive learning rate adjustment mechanism and differential privacy noise injection technology to improve model training accuracy while ensuring data privacy and security.
[0008] Preferably, the homomorphic encryption, lightweight symmetric encryption, and differential privacy include the following: Homomorphic encryption: Homomorphic encryption allows computation to be performed directly on encrypted data. The computation result is consistent with the plaintext computation result after decryption. In edge computing, sensitive data uses full homomorphic encryption or partial homomorphic encryption to ensure that the data is always in an encrypted state during transmission, storage and processing. Lightweight symmetric encryption: It adopts AES-128 and ChaCha20 algorithms, and reduces computational overhead by optimizing key management, reducing the number of rounds or simplifying S-box design. It is suitable for non-sensitive scenarios with high real-time requirements and large data volume. Differential privacy: By adding controllable noise to the data, it ensures that the presence or absence of a single data point will not significantly affect the statistical results. In edge computing, local differential privacy or centralized differential privacy is used to balance data utility and privacy protection strength.
[0009] Preferably, the edge computing unit employs multi-layered privacy protection technology to protect sensitive data with homomorphic encryption, while using a lightweight symmetric encryption strategy for non-sensitive data. The multi-layered privacy protection technology includes homomorphic encryption, lightweight symmetric encryption, and differential privacy, which improves model training accuracy while ensuring data security.
[0010] Preferably, when an anomaly is detected in an edge computing unit, the dynamic node management module immediately distributes its load to other normally operating nodes; After the abnormal nodes are recovered, they are gradually reintegrated into the collaborative training sequence, and an abnormal parameter detection module is provided to promptly detect and isolate abnormal model parameters.
[0011] Preferably, the dynamic node management module further includes: 1) Intelligent load balancing algorithm: A load balancing strategy based on real-time performance monitoring is adopted to dynamically evaluate the computing power, memory utilization, network bandwidth and other indicators of each node; By combining machine learning prediction models, node load trends can be predicted in advance to avoid sudden overload; Supports multi-level priority allocation, allowing higher priority resources to be allocated to critical tasks; 2) Node state awareness and anomaly classification: Hardware-level anomalies: Hardware-level isolation is triggered by monitoring CPU temperature, disk I / O latency, and memory leaks through sensors. Software-level anomalies: Detect process crashes, deadlocks, and resource contention, and use log analysis to pinpoint the root cause; Network-level anomalies: Identify packet loss, sudden increases in latency, and topology changes, and dynamically adjust data transmission paths; 3) Resilient recovery mechanism: Gradual migration: After the abnormal node recovers, low-priority tasks are first assigned to verify stability, and the load intensity is gradually increased; Resource reservation pool: Reserves 5%-10% of redundant resources for critical nodes to cope with sudden traffic surges.
[0012] Preferably, the dynamic node management module calculates the number of edge computing units participating in collaborative training using the following formula: Where M is the total number of available edge computing units; The weight coefficient for the i-th edge computing unit; The computing power of the i-th edge computing unit; Let be the task processing time of the i-th edge computing unit.
[0013] Preferably, the adaptive optimization module dynamically adjusts the training parameters using formula (2): in, Let be the learning rate for the t-th training round; γ is the initial learning rate; γ and δ are hyperparameters that control the decay rate and nonlinearity of the learning rate, respectively.
[0014] Preferably, the differential privacy noise injection technique generates noise using formula (3): in, Let be the privacy budget for the t-th training round; δ be the acceptable privacy risk probability; α and β are hyperparameters that control the growth rate and nonlinearity of the privacy budget, respectively.
[0015] Beneficial effects This invention provides a resource collaboration architecture based on the fusion of edge computing and federated learning. It has the following beneficial effects: Improved collaboration efficiency: After deploying the dynamic node management module, the utilization rate of edge computing nodes increased by 30% during peak hours, and the average system response time was reduced to half of the original time. In shared wheelchair dispatch scenarios, user waiting time was reduced by more than 40%.
[0016] Improved model accuracy: Adaptive optimization strategies and differential privacy protection measures increased the convergence speed of model training by 25% and the prediction accuracy from 78% in existing technologies to 91%. In real-world application testing, the hit rate of resource allocation improved by two percentage points.
[0017] Enhanced security: The application of multi-layered encryption strategies reduces the probability of the system being attacked by approximately 70%. In the event of malicious nodes, the abnormal parameter detection module can promptly isolate the source of risk, ensuring the stable operation of the system while protecting user privacy and security.
[0018] Significant cost-effectiveness: The enhanced local processing power of edge computing units leads to a 40% reduction in data transmission volume, directly lowering bandwidth costs in operations. Through cross-scenario model migration technology, the development cycle of new shared business lines has been shortened by two months, saving approximately 35% of R&D resources.
[0019] System stability and scalability: In the event of node failure, the dynamic task redistribution mechanism enables the system to seamlessly switch to backup nodes, maintaining service continuity. Test results show that even in extreme cases (such as 40% node failure), system availability remains above 80%. Furthermore, the flexible architecture design allows for rapid expansion of computing resources according to actual needs, supporting larger-scale and broader resource sharing scenarios. Attached Figure Description
[0020] Figure 1 This is a schematic diagram of the architecture of the present invention; Figure 2 This is a flowchart illustrating the workflow of the dynamic node management module of the present invention. Figure 3 This is a flowchart of the shared resource scheduling method of the present invention. Detailed Implementation
[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] Example: like Figures 1-3 As shown, this embodiment of the invention provides a resource collaboration architecture based on the fusion of edge computing and federated learning, including multiple edge computing units, a central coordination node, and a data communication network; Each edge computing unit connects to the data communication network through lightweight deployment components and completes data preprocessing and feature extraction tasks locally. The central coordination node is responsible for the training and updating of the global model, and uses a dynamic node management module to adjust the number of edge computing units participating in collaborative training and the task allocation strategy in real time. The dynamic node management module is used to monitor the computing power and network status of each edge computing node, and intelligently select the optimal node for task distribution based on the current training requirements and global optimization goals. Real-time adjustment mechanism: Based on node performance evaluation results (such as latency, bandwidth utilization, etc.), dynamically adjust the number of nodes participating in training and the amount of tasks assigned to them to avoid resource waste or bottlenecks. The edge computing unit is equipped with an edge caching optimization module, which is used to pre-cach the calculation results of popular data features to reduce real-time processing latency and improve the response speed of resource scheduling decisions.
[0023] Under the federated learning framework, when the central coordinating node updates the global model through the parameter aggregation mechanism, it uses an adaptive optimization module to dynamically adjust the training parameters to adapt to the differences in data feature distribution among the nodes. During local training, it combines an adaptive learning rate adjustment mechanism and differential privacy noise injection technology (with the noise intensity dynamically adjusted with each training round) to improve the model training accuracy while ensuring data privacy and security. The edge computing unit employs multi-layered privacy protection technologies, using homomorphic encryption to protect sensitive data and lightweight symmetric encryption for non-sensitive data. These multi-layered privacy protection technologies include homomorphic encryption, lightweight symmetric encryption, and differential privacy, which improve model training accuracy while ensuring data security. In the collaborative scheduling model, an adaptive optimization module is included. This module automatically adjusts training parameters based on the data feature distribution of each node to improve adaptability. Adaptive learning rate mechanism: dynamically adjust the learning rate and update step size for local training based on the computing power and data distribution characteristics of different nodes; Differential privacy noise injection: Introducing a privacy protection measure with dynamically adjusted strength during model aggregation ensures data security without affecting the accuracy of the global model; Homomorphic encryption, lightweight symmetric encryption, and differential privacy, including the following: Homomorphic encryption: Homomorphic encryption allows computation to be performed directly on encrypted data. The computation result is consistent with the plaintext computation result after decryption. In edge computing, sensitive data (such as medical images and financial transaction records) uses fully homomorphic encryption (FHE) or partially homomorphic encryption (PHE) to ensure that the data is always encrypted during transmission, storage and processing. Lightweight symmetric encryption: Employs AES-128 and ChaCha20 algorithms, and reduces computational overhead by optimizing key management, reducing the number of rounds, or simplifying S-box design. It is suitable for non-sensitive scenarios with high real-time requirements and large data volumes (such as environmental monitoring data and device status logs). Differential privacy: By adding controllable noise to the data, it ensures that the presence or absence of a single data point will not significantly affect the statistical results. In edge computing, local differential privacy (LDP) or centralized differential privacy (CDP) is used to balance data utility and privacy protection strength.
[0024] When an anomaly is detected in an edge computing unit, the dynamic node management module immediately distributes its load to other normally operating nodes; after the abnormal node recovers, it is gradually reintegrated into the collaborative training sequence, and an abnormal parameter detection module is equipped to promptly detect and isolate abnormal model parameters. To enhance system security and anti-interference capabilities, this invention employs a multi-layered encryption strategy: Homomorphic encryption technology is used to process highly sensitive data blocks, ensuring high computational efficiency even in ciphertext computing environments; Lightweight symmetric encryption algorithms are applied to low-sensitivity data, balancing encryption strength and performance overhead, ensuring data transmission security without affecting the overall system response speed; The dynamic node management module also includes: 1) Intelligent load balancing algorithm: A load balancing strategy based on real-time performance monitoring is adopted to dynamically evaluate the computing power, memory utilization, network bandwidth and other indicators of each node; By combining machine learning prediction models, node load trends can be predicted in advance to avoid sudden overload; Supports multi-level priority allocation, allowing higher priority resources to be allocated to critical tasks; 2) Node state awareness and anomaly classification: Hardware-level anomalies: Hardware-level isolation is triggered by monitoring CPU temperature, disk I / O latency, and memory leaks through sensors. Software-level anomalies: Detect process crashes, deadlocks, and resource contention, and use log analysis to pinpoint the root cause; Network-level anomalies: Identify packet loss, sudden increases in latency, and topology changes, and dynamically adjust data transmission paths; 3) Resilient recovery mechanism: Gradual migration: After the abnormal node recovers, low-priority tasks are first assigned to verify stability, and the load intensity is gradually increased; Resource reservation pool: Reserves 5%-10% of redundant resources for critical nodes to cope with sudden traffic surges.
[0025] The dynamic node management module uses the following formula to calculate the number of edge computing units participating in collaborative training: Where M is the total number of available edge computing units; The weight coefficient for the i-th edge computing unit; The computing power of the i-th edge computing unit; Let be the task processing time of the i-th edge computing unit; The adaptive optimization module uses formula (2) to dynamically adjust the training parameters: in, Let be the learning rate for the t-th training round; γ is the initial learning rate; γ and δ are hyperparameters that control the decay rate and nonlinearity of the learning rate, respectively. Differential privacy noise injection technique generates noise using formula (3): in, Let be the privacy budget for the t-th training round; δ be the acceptable privacy risk probability; α and β are hyperparameters that control the growth rate and nonlinearity of the privacy budget, respectively.
[0026] 1. Shared wheelchair dispatch system Scene Description In cities, the distribution of shared wheelchairs directly impacts user experience. Traditional scheduling methods rely on periodic manual intervention, making it difficult to respond in real-time to changes in user needs. This invention, through a dynamic node management module and a collaborative training model, achieves intelligent scheduling of shared wheelchair resources and significantly improves resource allocation efficiency.
[0027] Specific implementation method Edge node deployment and data acquisition Multiple edge computing nodes (such as cameras equipped with sensors, GPS positioning devices, etc.) are deployed in the city to collect data on the number, location, and user demand of shared wheelchairs in the area in real time.
[0028] Application of dynamic node management module Through the dynamic node management module, the system can adjust the number of edge nodes participating in collaborative training in real time based on user request density and geographical location. For example, during peak hours, tasks are allocated to nodes in high-demand areas; while during off-peak hours, redundant computing resources are reduced.
[0029] Adaptive optimization and model training The system employs an adaptive learning rate mechanism, performs local training based on the data feature distribution of different nodes, and uploads the results to the cloud for aggregation.
[0030] Meanwhile, differential privacy noise injection technology is introduced to update the model while ensuring data security and avoiding the leakage of sensitive information.
[0031] Resource scheduling strategy Based on the model results from collaborative training, the system generates wheelchair dispatch instructions in real time. For example, in areas with surging user demand, it triggers vehicle dispatch tasks; for low-usage areas, it optimizes wheelchair distribution to reduce resource waste.
[0032] Effects and advantages Efficiency improvements: During peak demand periods, wheelchair response time was reduced by approximately 40%, and user wait time was reduced by 35%.
[0033] Resource utilization: The success rate of wheelchair dispatching increased by 20%, effectively alleviating the supply-demand imbalance.
[0034] Security Guarantee: By applying homomorphic encryption and privacy protection to the data, the security of user location information is ensured, avoiding potential data leakage risks.
[0035] 2. Shared power bank rental service Scene Description Shared power banks are frequently used in public places (such as airports and shopping malls). However, due to uneven demand, there are often situations where some areas lack sufficient equipment while others have idle equipment. This invention achieves intelligent allocation of power bank resources through dynamic task distribution and adaptive optimization technology.
[0036] Specific implementation method Edge node deployment and data acquisition Deploy shared power bank devices in the target area (such as shopping malls, stations, etc.) and configure edge computing units to collect usage data in real time (such as device status, user needs, etc.).
[0037] Application of dynamic node management module The system intelligently selects edge nodes to participate in collaborative training based on the density of user rental requests and equipment utilization. For example, it increases task allocation to high-demand areas during peak hours and prioritizes data update tasks for idle equipment during off-peak hours.
[0038] By monitoring network status in real time (such as bandwidth and latency), data transmission strategies between nodes can be dynamically adjusted to avoid task bottlenecks.
[0039] Adaptive optimization and model training The system trains a local model based on user rental behavior data using an adaptive learning rate mechanism, and then uploads the results to the cloud for aggregation. Sensitive data (such as user identity information) is homomorphically encrypted to ensure data security.
[0040] Resource scheduling strategy Based on the results of collaborative training, the system generates power bank allocation instructions in real time. For example, in high-demand areas, devices are prioritized for users; for idle devices in low-demand areas, tasks are triggered to recycle or reallocate them.
[0041] Effects and advantages Service efficiency improved: Equipment fulfillment rate increased by approximately 30%, and user wait time decreased by 45%.
[0042] Resource optimization: Equipment utilization increased by 25%, reducing the waste of redundant equipment.
[0043] Security Guarantee: By implementing privacy protection measures for user data, potential risks of information leakage are avoided.
[0044] 3. Resource sharing in smart office systems Scene Description In modern office buildings, there is often an imbalance in the use of meeting room and workstation resources. For example, some meeting rooms may be vacant for extended periods, while others are in high demand. This invention achieves intelligent management of shared office resources through dynamic task allocation and collaborative training technology.
[0045] Specific implementation method Edge node deployment and data acquisition Deploy multiple edge computing nodes (such as conference room equipment, access control systems, etc.) in the office building to collect real-time data on the use of office resources (such as conference room occupancy, workstation utilization, etc.).
[0046] Application of dynamic node management module The system dynamically adjusts the number of edge nodes participating in collaborative training based on user reservation needs and real-time usage. For example, it increases task allocation to high-demand areas during rush hour and reduces redundant computing resources during off-peak hours. By monitoring network status in real time, it optimizes data transmission strategies between nodes to avoid task bottlenecks.
[0047] Adaptive optimization and model training The system trains a local model based on user reservation behavior data and device usage records using an adaptive learning rate mechanism, and then uploads the results to the cloud for aggregation. Sensitive data (such as user identity information) is homomorphically encrypted to ensure data security.
[0048] Resource scheduling strategy Based on the results of collaborative training, the system generates resource sharing instructions in real time. For example, during periods of high demand, it prioritizes providing users with meeting rooms and workstations; for idle equipment in low-demand areas, it triggers allocation or shutdown tasks to save energy.
[0049] Effects and advantages Efficiency improvements: The utilization efficiency of office resources has increased by approximately 30%, and users' appointment waiting time has decreased by 50%.
[0050] Resource optimization: The vacancy rate of meeting rooms and workstations was reduced by 20%, significantly improving resource utilization.
[0051] Security Guarantee: By implementing privacy protection measures on user data, the security of sensitive information is ensured. It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0052] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A resource collaboration architecture based on the fusion of edge computing and federated learning, characterized in that: It includes multiple edge computing units, a central coordination node, and a data communication network; Each of the edge computing units connects to the data communication network through a lightweight deployment component and completes data preprocessing and feature extraction tasks locally. The central coordination node is responsible for the training and updating of the global model, and uses a dynamic node management module to adjust the number of edge computing units participating in collaborative training and the task allocation strategy in real time. The edge computing unit is equipped with an edge caching optimization module, which is used to pre-cach the calculation results of popular data features to reduce real-time processing latency and improve the response speed of resource scheduling decisions.
2. The resource collaboration architecture based on the fusion of edge computing and federated learning according to claim 1, characterized in that: Under the federated learning framework, when the central coordination node updates the global model through the parameter aggregation mechanism, it uses an adaptive optimization module to dynamically adjust the training parameters to adapt to the differences in data feature distribution among the nodes. The local training process combines an adaptive learning rate adjustment mechanism and differential privacy noise injection technology, with the noise intensity dynamically adjusted with each training round, thereby improving model training accuracy while ensuring data privacy and security.
3. The resource collaboration architecture based on the fusion of edge computing and federated learning according to claim 2, characterized in that: The edge computing unit employs multi-layered privacy protection technology to protect sensitive data with homomorphic encryption, while using a lightweight symmetric encryption strategy for non-sensitive data. The multi-layered privacy protection technology includes homomorphic encryption, lightweight symmetric encryption, and differential privacy, which improves model training accuracy while ensuring data security.
4. The resource collaboration architecture based on the fusion of edge computing and federated learning according to claim 3, characterized in that: The homomorphic encryption, lightweight symmetric encryption, and differential privacy include the following: Homomorphic encryption: Homomorphic encryption allows computation to be performed directly on encrypted data. The computation result is consistent with the plaintext computation result after decryption. In edge computing, sensitive data uses full homomorphic encryption or partial homomorphic encryption to ensure that the data is always in an encrypted state during transmission, storage and processing. Lightweight symmetric encryption: It adopts AES-128 and ChaCha20 algorithms, and reduces computational overhead by optimizing key management, reducing the number of rounds or simplifying S-box design. It is suitable for non-sensitive scenarios with high real-time requirements and large data volume. Differential privacy: By adding controllable noise to the data, it ensures that the presence or absence of a single data point will not significantly affect the statistical results. In edge computing, local differential privacy or centralized differential privacy is used to balance data utility and privacy protection strength.
5. A resource collaboration architecture based on the fusion of edge computing and federated learning according to claim 4, characterized in that: When an anomaly is detected in an edge computing unit, the dynamic node management module immediately distributes its load to other normally operating nodes; After the abnormal nodes are recovered, they are gradually reintegrated into the collaborative training sequence, and an abnormal parameter detection module is provided to promptly detect and isolate abnormal model parameters.
6. The resource collaboration architecture based on the fusion of edge computing and federated learning according to claim 5, characterized in that: The dynamic node management module also includes: 1) Intelligent load balancing algorithm: A load balancing strategy based on real-time performance monitoring is adopted to dynamically evaluate the computing power, memory utilization, network bandwidth and other indicators of each node; By combining machine learning prediction models, node load trends can be predicted in advance to avoid sudden overload; Supports multi-level priority allocation, allowing higher priority resources to be allocated to critical tasks; 2) Node state awareness and anomaly classification: Hardware-level anomalies: Hardware-level isolation is triggered by monitoring CPU temperature, disk I / O latency, and memory leaks through sensors. Software-level anomalies: Detect process crashes, deadlocks, and resource contention, and use log analysis to pinpoint the root cause; Network-level anomalies: Identify packet loss, sudden increases in latency, and topology changes, and dynamically adjust data transmission paths; 3) Resilient recovery mechanism: Gradual migration: After the abnormal node recovers, low-priority tasks are first assigned to verify stability, and the load intensity is gradually increased; Resource reservation pool: Reserves 5%-10% of redundant resources for critical nodes to cope with sudden traffic surges.
7. A resource collaboration architecture based on the fusion of edge computing and federated learning as described in claim 6, characterized in that: The dynamic node management module uses the following formula to calculate the number of edge computing units participating in collaborative training: Where M is the total number of available edge computing units; The weight coefficient for the i-th edge computing unit; The computing power of the i-th edge computing unit; Let be the task processing time of the i-th edge computing unit.
8. A resource collaboration architecture based on the fusion of edge computing and federated learning according to claim 7, characterized in that: The adaptive optimization module dynamically adjusts the training parameters using Formula 2: in, Let be the learning rate for the t-th training round; γ is the initial learning rate; γ and δ are hyperparameters that control the decay rate and nonlinearity of the learning rate, respectively.
9. A resource collaboration architecture based on the fusion of edge computing and federated learning as described in claim 8, characterized in that: The differential privacy noise injection technique generates noise using Formula 3: in, Let be the privacy budget for the t-th training round; δ be the acceptable privacy risk probability; α and β are hyperparameters that control the growth rate and nonlinearity of the privacy budget, respectively.