A cloud-network converged service management system based on edge computing

By generating control microservice orchestration blueprints through the cloud-based policy orchestration layer, and through the synergy of the edge policy compilation layer and the real-time execution layer, the problems of low deployment efficiency and control interruption of control policies across heterogeneous edge nodes in existing technologies are solved, achieving efficient and stable control execution and optimization.

CN123123145APending Publication Date: 2026-11-17SHANGHAI LIGUS TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611597749.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-15
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing industrial control systems, control strategies are tied to specific hardware platforms, requiring manual rewriting for deployment across heterogeneous edge nodes, and cannot be automatically adapted; control logic runs as a monolithic program, and changes to local functions require overall redeployment; containerized execution lacks deterministic communication guarantees, resulting in large control latency jitter; the lack of a stateless snapshot mechanism leads to control interruptions; there is a lack of health scoring and dynamic scheduling methods, making it impossible to dynamically adjust resources based on execution quality; and strategy execution data cannot be transmitted back to the cloud to drive optimization without the original data leaving the edge nodes.

Method used

The cloud-based strategy orchestration layer generates a blueprint for controlling microservice orchestration. The edge strategy compilation layer combines hardware-protocol profiles to compile microservice deployment packages. The edge real-time execution layer executes in the container runtime and communicates through a deterministic communication network. Control state snapshots ensure migration continuity, health scores enable dynamic scheduling, and federated learning drives strategy optimization.

Benefits of technology

It enables automatic adaptation and deployment of control policies across heterogeneous edge nodes, reducing deployment time, ensuring control execution latency and resource isolation, avoiding control interruptions, and continuously improving control quality through federated learning optimization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123145A_ABST
    Figure CN123123145A_ABST
Patent Text Reader

Abstract

This invention discloses a cloud-network converged service management system based on edge computing, belonging to the interdisciplinary field of industrial automation control and edge computing. The system includes a cloud-based policy orchestration layer, an edge policy compilation layer, an edge real-time execution layer, and a cloud-network converged transmission layer. The cloud-based policy orchestration layer generates a control microservice orchestration blueprint. The edge policy compilation layer, combined with local hardware-protocol profiles, compiles the blueprint into microservice deployment packages. The edge real-time execution layer executes the control microservices in a real-time container runtime and communicates via a deterministic communication network. This system enables automatic adaptation and deployment of control policies across heterogeneous edge nodes, ensuring deterministic control execution and migration continuity, and improving the robustness of control services.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of industrial automation control and edge computing, specifically to a cloud-network converged service management system based on edge computing. Background Technology

[0002] Existing industrial control systems typically use programmable logic controllers (PLCs) as the core control device. Control strategies are written in ladder diagrams or structured text and embedded in specific hardware platforms. With the development of edge computing technology, some solutions deploy edge nodes in the industrial field, with control strategies generated in the cloud and then distributed to the edge nodes for execution. In these solutions, the edge nodes handle data acquisition, protocol conversion, and lightweight inference, while the generation and distribution of control commands still rely on cloud servers or dedicated controllers. Other solutions allocate computing resources in the cloud, distributing computing tasks based on the load of the edge nodes. In recent years, container technology has been gradually introduced into the edge computing field, with some solutions deploying applications as containers on edge nodes to achieve isolated operation and independent scaling. In the above solutions, the description, compilation, deployment, and execution of control strategies are completed by different modules, lacking a unified description and compilation mechanism between these modules.

[0003] However, the aforementioned existing technologies have the following shortcomings. First, control strategies are tied to specific hardware platforms, requiring manual rewriting when deployed across heterogeneous edge nodes, and cannot automatically adapt to different computing power levels, protocol stacks, and container runtimes. Second, control logic runs as a monolithic program; changes to local functions require complete redeployment, lacking atomic decomposition and dynamic reconfiguration. Third, containerized execution lacks deterministic communication guarantees, resulting in significant control latency fluctuations. Fourth, there is no state snapshot mechanism during microservice migration, leading to control interruptions. Fifth, there is a lack of health scoring and dynamic scheduling methods, making it impossible to dynamically adjust resources based on execution quality. Sixth, policy execution data cannot be transmitted back to the cloud to drive optimization without the original data leaving the edge nodes. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention discloses a cloud-network converged service management system based on edge computing. The cloud-based policy orchestration layer generates and controls the microservice orchestration blueprint, the edge policy compilation layer compiles the microservice deployment package by combining hardware-protocol profiles, the edge real-time execution layer executes in the container runtime and communicates through a deterministic communication network, control state snapshots ensure migration continuity, health scores enable dynamic scheduling, and federated learning drives policy optimization.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a cloud-network converged service management system based on edge computing, comprising a cloud policy orchestration layer, an edge policy compilation layer, an edge real-time execution layer, and a cloud-network converged transmission layer. The cloud policy orchestration layer is used to generate a control microservice orchestration blueprint, which defines atomic control capabilities, data flows and triggering relationships between capabilities, and service quality objectives. The edge policy compilation layer is used to combine local hardware-protocol profiles to compile the control microservice orchestration blueprint into a microservice deployment package. The edge real-time execution layer is used to execute the control microservices in the microservice deployment package in a container runtime and to communicate between microservices through a deterministic communication network, wherein the container runtime is a real-time container runtime or a non-real-time container runtime. The cloud-network converged transmission layer is used to transmit policy descriptions, execution quality data, and compilation artifacts between the cloud and edge nodes. The edge real-time execution layer is also used to maintain a control state snapshot to restore the control state during control microservice migration.

[0006] Preferably, the cloud-based strategy orchestration layer uses a declarative domain-specific language when generating the control microservice orchestration blueprint. The control microservice orchestration blueprint adopts a three-layer structure of capability layer, orchestration layer, and target layer. The capability layer declares atomic control capabilities, the orchestration layer declares data flow, triggering relationships, and timing constraints between capabilities, and the target layer declares service quality goals.

[0007] Preferably, the cloud policy orchestration layer further includes a semantic verification module, which performs integrity verification, constraint satisfiability verification, and service quality reachability verification on the control microservice orchestration blueprint.

[0008] Preferably, the edge policy compilation layer includes hardware, a protocol profile acquisition module, and a policy compiler. The hardware-protocol profile acquisition module acquires computing power level, protocol stack, container runtime, and control execution capabilities. The policy compiler performs blueprint parsing and microservice mapping, hardware adaptation and compilation target selection, microservice communication configuration generation, and degradation migration contingency plan generation.

[0009] Preferably, the hardware adaptation and compilation target selection includes: if the local machine has IEC 61131-3 runtime and the microservice requires hard real-time control, generating structured text code and deploying it to the PLC runtime; if the local machine supports real-time containers and the microservice is soft real-time or computationally intensive, generating containerized microservice instance configuration and deploying it to the real-time container; if the local computing power is insufficient or the protocol is incompatible, generating remote call stub code and incorporating it into the degradation plan.

[0010] Preferably, the edge real-time execution layer includes a real-time container runtime, at least one control microservice instance, a time-sensitive network communication module, and a control state snapshot management module; the real-time container runtime binds CPU cores through cgroup cpuset and locks memory through mlock; the time-sensitive network communication module guarantees deterministic latency based on IEEE 802.1AS time synchronization, 802.1Qbv traffic shaping, and 802.1Qbu frame preemption.

[0011] Preferably, when migrating the control microservice, the control state snapshot management module synchronizes the PID integral term, filter historical data, and timer status to the target node through a time-sensitive network channel. The target node executes the control microservice in parallel in shadow mode, and completes the takeover after the deviation between the actual control state and the target control state is less than a preset threshold.

[0012] Preferably, it also includes a microservice health score and dynamic scheduling module. The microservice health score and dynamic scheduling module calculates the health score based on latency jitter, resource utilization, control accuracy deviation and communication quality, and triggers resource quota adjustment, local rescheduling, neighbor migration and remote call degradation in order of priority when the health score is lower than the threshold.

[0013] Preferably, the cloud-network converged transmission layer includes a network status awareness module, which monitors network jitter, latency, and bandwidth in real time and provides network status information to the cloud policy orchestration layer. The microservice orchestration engine of the cloud policy orchestration layer adjusts and controls the deployment location and communication compression level of microservices according to the network status.

[0014] Preferably, the cloud-based policy orchestration layer further includes a federated learning optimization module. The federated learning optimization module receives policy execution effect, hardware profile, and compilation configuration triplet data uploaded by edge nodes. The original control data does not leave the edge nodes; only the desensitization effect index, configuration parameters, and encryption model gradient are uploaded. After aggregation, the module drives iterative optimization of template parameters, orchestration logic, and service quality target thresholds for declarative domain-specific languages.

[0015] Compared with existing technologies, this invention provides a cloud-network converged service management system based on edge computing, which has the following beneficial effects: This invention generates a control microservice orchestration blueprint through a cloud-based policy orchestration layer, and compiles it into a microservice deployment package using an edge policy compilation layer that combines hardware-protocol profiling. This enables control policies to automatically adapt and deploy across heterogeneous edge nodes without manual rewriting. The edge real-time execution layer executes the control microservices within a real-time container runtime and communicates via a deterministic communication network, ensuring control execution latency and resource isolation. The control state snapshot management module restores the control state during migration, preventing control interruptions. The health scoring and dynamic scheduling module enables dynamic resource adjustment and degradation handling. The federated learning optimization module drives continuous policy optimization without the original data leaving the edge nodes. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the system connection of the present invention. Detailed Implementation

[0017] 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.

[0018] refer to Figure 1 This invention provides a cloud-network converged service management system based on edge computing.

[0019] Example 1 Purpose of implementation: This embodiment is used to solve the problem that control service strategies cannot be automatically ported across nodes and the deployment efficiency is low when multiple heterogeneous edge nodes coexist in a discrete manufacturing workshop.

[0020] Implementation System: This embodiment employs the cloud-network converged service management system based on edge computing described in the overall system implementation above. Three heterogeneous edge nodes are set up within the workshop: Node A is an industrial control computer supporting real-time containers and TSN; Node B is a traditional PLC supporting IEC61131-3; and Node C is an embedded controller with insufficient computing power but a Modbus interface. The cloud-based policy orchestration layer generates a control microservice orchestration blueprint. The capability layer declares three atomic control capabilities: "vibration monitoring," "threshold judgment," and "feed rate adjustment." The orchestration layer declares the data flow of "vibration monitoring → threshold judgment → feed rate adjustment" with an end-to-end latency ≤5ms. The target layer declares a control accuracy deviation ≤2%. The hardware-protocol profiling module of the edge policy compilation layer collects data showing that Node A supports real-time containers, Node B supports IEC 61131-3, and Node C has insufficient computing power. The policy compiler performs blueprint parsing and microservice mapping, mapping "vibration monitoring" and "threshold judgment" to containerized microservice templates, and mapping "feed rate adjustment" to a structured text template. During hardware adaptation and compilation target selection, containerized microservice instance configurations are generated and deployed to the real-time container for node A; structured text code is generated and deployed to the PLC runtime for node B; and remote call stub code is generated for node C due to insufficient computing power and included in the degradation plan. When generating inter-microservice communication configurations, a time-sensitive network stream is configured between node A and node B, with a priority of 7 and a time slot of 250μs. The edge real-time execution layer runs the "vibration monitoring" and "threshold judgment" microservices in the real-time container of node A, and the "feed rate adjustment" microservice in the PLC of node B. The microservices communicate with each other via time-sensitive network. The control state snapshot management module saves PID integral terms and filter history on nodes A and B respectively. The control microservice health scoring and dynamic scheduling module continuously calculates the health score.

[0021] Implementation results: The control strategy is automatically compiled and deployed across three types of heterogeneous nodes, eliminating the need for manual rewriting and reducing deployment time from 3 days in the traditional solution to 2 hours. When the health score of node A falls below the threshold, a neighbor migration is triggered, migrating the "vibration monitoring" microservice to a backup node. The status snapshot is synchronized through a time-sensitive network channel, and the target node runs in shadow mode. It takes over after the deviation is less than 0.5%, and the migration process is uninterrupted.

[0022] Example 2 Purpose of implementation: This embodiment addresses the problem that control service strategies cannot be uniformly arranged and control accuracy is affected by protocol differences when multiple industrial communication protocols coexist in the process industry.

[0023] Implementation System: This embodiment employs the cloud-network converged service management system based on edge computing described in the overall system implementation above. Edge node D supports OPC UA and Profinet and has real-time containers; edge node E supports Modbus TCP and EtherCAT and has IEC 61131-3 runtime capability; edge node F supports TSN but has limited computing power. The cloud-based policy orchestration layer generates a control microservice orchestration blueprint. The capability layer declares four atomic control capabilities: "temperature acquisition," "PID regulation," "valve control," and "alarm." The orchestration layer declares "temperature acquisition → PID regulation → valve control, with parallel alarm monitoring." The target layer declares a response time ≤10ms and jitter ≤2ms. After the semantic verification module passes the verification, the cloud-network converged transport layer distributes the blueprint to the edge policy compilation layer. The hardware-protocol profile acquisition module collects the hardware-protocol profiles of nodes D, E, and F. The policy compiler performs blueprint parsing and microservice mapping, mapping "temperature acquisition" to an OPC UA client microservice template, "PID regulation" to a containerized microservice template, "valve control" to a structured text template, and "alarm" to a lightweight microservice template. During hardware adaptation and compilation target selection, node D generates containerized microservice instance configurations, node E generates structured text code, and node F generates lightweight microservice instance configurations and configures remote call stubs as a fallback plan. When generating inter-microservice communication configurations, microservices within node D communicate via shared memory; communication between node D and node E is configured with a priority of 6 and a time slot of 500μs via time-sensitive network streams; and communication between node F and node D is configured with a priority of 5 and a buffer size of 1MB via message queues. The edge real-time execution layer runs the "temperature acquisition" and "PID regulation" microservices in the real-time container on node D, the "valve control" microservice in the PLC on node E, and the "alarm" microservice on node F. The control microservice health scoring and dynamic scheduling module continuously monitors the latency jitter, resource utilization, control accuracy deviation, and communication quality of each microservice.

[0024] Implementation results: The system supports heterogeneous nodes using multiple protocols such as OPC UA, Profinet, Modbus TCP, and EtherCAT, reducing the deployment time of the control service to 1.5 hours and decreasing the control accuracy deviation from 3% in traditional solutions to 1.5%. When the health score of node F falls below the threshold, resource quota adjustments, local rescheduling, migration to neighboring node D, and remote degradation calls are triggered sequentially according to priority, ensuring uninterrupted alarm functionality.

[0025] Example 3 Purpose of implementation: This embodiment addresses the issues of declining control service quality due to network status changes and the inability to continuously optimize control strategies based on actual operational data in cross-plant scenarios of large enterprise groups.

[0026] Implementation System: This embodiment employs the edge computing-based cloud-network converged service management system described in the overall system implementation above. Edge nodes are deployed in both Plant 1 and Plant 2, and are uniformly managed by the cloud-based policy orchestration layer. Edge node G in Plant 1 runs the "Temperature PID Regulation" microservice, while edge node H in Plant 2 runs the "Pressure Monitoring" microservice. The network status awareness module of the cloud-network converged transmission layer monitors the increase in network jitter in Plant 1 in real time and provides this network status information to the microservice orchestration engine of the cloud-based policy orchestration layer. The microservice orchestration engine adjusts the deployment location of the control microservice based on the network status, deciding to migrate the "Temperature PID Regulation" microservice to edge node H in Plant 2. The control status snapshot management module of the edge real-time execution layer synchronizes the PID integral term, filter historical data, and timer status to node H via a time-sensitive network channel. Node H executes the control microservice in parallel in shadow mode, taking over once the deviation between the actual control state and the target control state is less than 0.5%. Simultaneously, each edge node uploads the policy execution effect, hardware profile, and compilation configuration triplet data to the federated learning optimization module of the cloud-based policy orchestration layer. The raw control data remains within the edge nodes; only the anonymization effectiveness metrics, configuration parameters, and encrypted model gradients are uploaded. The federated learning optimization module aggregates this data to drive iterative optimization of template parameters, orchestration logic, and service quality target thresholds for the declarative domain-specific language.

[0027] Implementation results: The cross-plant migration time was reduced from 30 minutes in the traditional solution to 2 minutes, with no interruption in the control process. Federated learning optimization reduced the control accuracy deviation month by month, from 1.8% to 1.0% after three months. After optimization of the template parameters of the declarative domain-specific language, the default proportional coefficient of temperature PID regulation was adjusted from 1.2 to 1.5, and the jitter tolerance in the service quality target was adjusted from 2ms to 1.5ms. The optimized version was sent to the edge nodes in a grayscale manner after semantic verification, and was fully released after verification.

[0028] Comparative Example Purpose of implementation: This comparative example is used to compare with the above embodiments to illustrate the advantages of the system of the present invention over traditional cloud-based centralized control schemes in terms of deployment efficiency, control accuracy, latency jitter, and migration continuity.

[0029] Implementation System: The comparison uses a traditional cloud-based centralized control scheme. The cloud server generates control commands, which are then sent to the edge gateway via ordinary Ethernet. The edge gateway forwards these commands to the PLC for execution. Control policies are tied to specific PLC brands; changing a PLC requires manually rewriting the ladder diagram. Control logic is not atomically generated into control microservices; any change to a function requires a complete redeployment. Communication between microservices lacks time-sensitive network guarantees, resulting in significant latency and jitter. There is no control state snapshot migration mechanism, leading to control interruption during migration. There is no control microservice health scoring and dynamic scheduling module. There is no federated learning optimization module.

[0030] Implementation results: In the comparison, the control strategy takes 3 days to deploy across heterogeneous nodes, the control accuracy deviation is about 3%, the latency jitter is about 5ms, the migration interruption time is about 30 seconds, and there is no self-optimization capability.

[0031] Summary of beneficial effects: The three embodiments described above are compared with the comparative example. Regarding deployment time, Embodiment 1 took 2 hours, Embodiment 2 took 1.5 hours, and Embodiment 3 had a migration time of 2 minutes, while the comparative example took 3 days. Regarding control accuracy deviation, Embodiment 1 had a deviation of 1.5%, Embodiment 2 had a deviation of 1.5%, and Embodiment 3, after federated learning optimization, reduced the deviation from 1.8% to 1.0%, while the comparative example had a deviation of 3%. Regarding latency jitter, Embodiment 1 had a latency of no more than 1ms, Embodiment 2 had a latency of no more than 2ms, and Embodiment 3 had a latency of no more than 1.5ms, while the comparative example had a latency of approximately 5ms. Regarding migration interruption, all three embodiments were uninterrupted, while the comparative example experienced an interruption of approximately 30 seconds. Regarding self-optimization capability, all three embodiments supported federated learning optimization, while the comparative example did not.

[0032] As can be seen from the above comparison, the system of the present invention generates a control microservice orchestration blueprint through a cloud-based policy orchestration layer, compiles it into a microservice deployment package by combining a hardware-protocol profile with an edge policy compilation layer, executes the control microservice in real-time container runtime through time-sensitive network communication, ensures migration continuity through a control state snapshot management module, realizes dynamic scheduling through a control microservice health scoring and dynamic scheduling module, and realizes policy self-optimization through a federated learning optimization module. This significantly improves the deployment efficiency, control accuracy, and robustness of control services on heterogeneous edge nodes.

[0033] 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 cloud-network converged service management system based on edge computing, characterized in that, It includes a cloud policy orchestration layer, an edge policy compilation layer, an edge real-time execution layer, and a cloud-network converged transmission layer. The cloud policy orchestration layer is used to generate a control microservice orchestration blueprint. The control microservice orchestration blueprint defines atomic control capabilities, data flows and triggering relationships between capabilities, and service quality objectives. The edge policy compilation layer is used to compile the control microservice orchestration blueprint into a microservice deployment package by combining the local hardware-protocol profile. The edge real-time execution layer is used to execute the control microservices in the microservice deployment package in the container runtime and to conduct inter-microservice communication through a deterministic communication network. The container runtime can be a real-time container runtime or a non-real-time container runtime. The cloud-network converged transport layer is used to transmit policy descriptions, execution quality data, and compilation artifacts between the cloud and edge nodes. The edge real-time execution layer is also used to maintain control state snapshots so that the control state can be restored through the control state snapshots when the control microservices are migrated.

2. The cloud-network converged service management system based on edge computing according to claim 1, characterized in that: When the cloud-based policy orchestration layer generates the control microservice orchestration blueprint, it uses a declarative domain-specific language. The control microservice orchestration blueprint adopts a three-layer structure of capability layer, orchestration layer, and target layer. The capability layer declares atomic control capabilities, the orchestration layer declares data flow, triggering relationships, and timing constraints between capabilities, and the target layer declares service quality goals.

3. The cloud-network converged service management system based on edge computing according to claim 1, characterized in that: The cloud-based policy orchestration layer also includes a semantic verification module, which performs integrity verification, constraint satisfiability verification, and service quality reachability verification on the control microservice orchestration blueprint.

4. The cloud-network converged service management system based on edge computing according to claim 1, characterized in that: The edge policy compilation layer includes hardware, a protocol profile acquisition module, and a policy compiler. The hardware-protocol profile acquisition module collects computing power level, protocol stack, container runtime, and control execution capabilities. The policy compiler performs blueprint parsing and microservice mapping, hardware adaptation and compilation target selection, microservice communication configuration generation, and degradation migration contingency plan generation.

5. The cloud-network converged service management system based on edge computing according to claim 4, characterized in that: The hardware adaptation and compilation target selection includes: if the local machine has IEC 61131-3 runtime and the microservice requires hard real-time control, generating structured text code and deploying it to the PLC runtime; if the local machine supports real-time containers and the microservice is soft real-time or compute-intensive, generating containerized microservice instance configuration and deploying it to the real-time container; if the local computing power is insufficient or the protocol is incompatible, generating remote call stub code and incorporating it into the degradation plan.

6. The cloud-network converged service management system based on edge computing according to claim 1, characterized in that: The edge real-time execution layer includes a real-time container runtime, at least one control microservice instance, a time-sensitive network communication module, and a control state snapshot management module. The real-time container runtime binds CPU cores through cgroup cpuset and locks memory through mlock. The time-sensitive network communication module guarantees deterministic latency based on IEEE 802.1AS time synchronization, 802.1Qbv traffic shaping, and 802.1Qbu frame preemption.

7. The cloud-network converged service management system based on edge computing according to claim 6, characterized in that: When migrating control microservices, the control state snapshot management module synchronizes the PID integral term, filter historical data, and timer status to the target node through a time-sensitive network channel. The target node executes the control microservice in parallel in shadow mode, and completes takeover after the deviation between the actual control state and the target control state is less than a preset threshold.

8. The cloud-network converged service management system based on edge computing according to claim 1, characterized in that: It also includes a microservice health score and dynamic scheduling module. The microservice health score and dynamic scheduling module calculates the health score based on latency jitter, resource utilization, control accuracy deviation and communication quality. When the health score is lower than the threshold, it triggers resource quota adjustment, local rescheduling, neighbor migration and remote call degradation in order of priority.

9. The cloud-network converged service management system based on edge computing according to claim 1, characterized in that: The cloud-network converged transmission layer includes a network status awareness module, which monitors network jitter, latency, and bandwidth in real time and provides network status information to the cloud policy orchestration layer. The microservice orchestration engine of the cloud policy orchestration layer adjusts and controls the deployment location and communication compression level of microservices based on the network status.

10. A cloud-network converged service management system based on edge computing according to claim 1, characterized in that: The cloud-based policy orchestration layer also includes a federated learning optimization module. This module receives policy execution results, hardware profiles, and compilation configuration triplet data uploaded by edge nodes. The original control data does not leave the edge nodes; only the desensitization effect indicators, configuration parameters, and encryption model gradients are uploaded. After aggregation, the module drives iterative optimization of template parameters, orchestration logic, and service quality target thresholds in a declarative domain-specific language.