Intelligent routing execution device, method and system in collaborative optimization system

Abstract

The invention discloses an intelligent routing execution device, method and system in a collaborative optimization system, and belongs to the technical field of collaborative optimization of complex systems. As a key execution component of the collaborative optimization system, the device is characterized in that a strategy interface is configured to receive a first optimization signal from a closed-loop learning component, and the signal comprises a routing strategy parameter cooperatively generated based on the same single global performance evaluation value; the collaborative evaluator generates a routing evaluation result based on the strategy parameters and the multi-dimensional context information; and the task allocator routes the task request to the target functional component accordingly. Wherein internal logic of the collaborative evaluator is configured or updated by the closed loop learning component based on optimization of the single global performance evaluation value. According to the method, the routing decision is constructed into the accurate execution of the global collaborative optimization strategy, the reliable landing and feedback closed loop of the optimization instruction in the system are ensured, and the method is a key hardware / software carrier for realizing the deep collaborative evolution of the system component.

Description

Technical Field

[0001] This invention belongs to the field of complex system collaborative optimization technology, specifically relating to an intelligent routing execution device, method, and system for collaborative optimization systems. More particularly, it relates to a core execution component that receives policy parameters generated by a closed-loop learning component based on global collaborative optimization, and makes real-time, accurate routing decisions accordingly to drive the overall collaborative evolution of the system. Background Technology

[0002] In modern complex distributed systems (such as microservice architectures, cloud-native applications, and industrial cyber-physical systems), the overall system performance depends on deep collaboration among components. Our prior application (parent application, application number: 2025115379095) proposes a system collaborative optimization paradigm based on a composite optimization space. The core of this paradigm lies in: constructing a composite optimization space that integrates routing policy parameters and internal parameters of functional components; and, based on the same global performance evaluation value, executing a unified optimization process through a closed-loop learning component to collaboratively generate two types of optimization signals that update routing policies and functional component parameters, thereby achieving system-level collaborative evolution.

[0003] In the aforementioned collaborative optimization paradigm, the "dynamic routing component" plays a crucial execution role: it must route external requests to the most suitable functional component in real time. However, existing routing components (such as API gateways and load balancers) generally suffer from the following limitations: they are typically statically configured or can only be easily adjusted based on local rules and limited objectives (such as minimum latency and round-robin). These components cannot deeply integrate into an optimization loop driven by global objectives, requiring routing strategies to be tightly coupled with downstream functional parameters and to evolve collaboratively. Their decision-making logic is isolated, unable to respond to changes in the global optimization state, and unable to provide high-quality, accurately attributable (e.g., correlated through globally unique identifiers) feedback on the decision-making effect of the optimization loop. At a deeper level, the monitoring data or contextual information that traditional routing components rely on has its format, semantics, and collection points defined by each component or team, forming "data silos." This leads to a fundamental discrepancy between the optimization hub (if it exists) and the execution components' understanding of the "system state," making any attempt at global collaborative optimization difficult to sustain at the data level.

[0004] Therefore, in the engineering practice of realizing the aforementioned collaborative optimization paradigm, a specific technical problem that urgently needs to be solved is: how to design and implement an intelligent routing execution component that can accurately, reliably, and verifiably execute the routing strategy derived by the closed-loop learning component through global collaborative optimization, and become the key hub connecting "optimization decision-making" with "system execution" and "effect feedback", thereby ensuring the efficient operation of the entire collaborative optimization closed loop.

[0005] This application is a specific implementation, optimization, and safeguard scheme for the key execution component, the "dynamic routing component," within the collaborative optimization paradigm proposed in the invention patent application number 2025115379095 (hereinafter referred to as the "parent application"). This application aims to provide an intelligent routing execution device that strictly adheres to this collaborative optimization paradigm and meets the mandatory architectural requirements of a high-reliability system. The claims and description of this application should be understood as providing independent and complete protection for the intelligent routing execution device, method, and system within the context of this collaborative optimization paradigm. Summary of the Invention

[0006] (a) Technical problems to be solved This invention aims to solve the technical problem of the disconnect between the routing execution component and the global optimization loop in a collaborative optimization system. Specifically, it provides a dedicated intelligent routing execution device and method that can seamlessly integrate into the collaborative optimization closed loop, receive and execute strategies generated by global optimization, and provide accurate feedback data, thereby fundamentally solving the problems of "execution mismatch" and "inaccurate feedback" in traditional routing components in collaborative architectures.

[0007] (II) Technical Solution To achieve the above objectives, this invention proposes an intelligent routing execution device, method, and system. The core of this invention lies in the fact that the device is not an independent, autonomous router, but rather designed as a controlled and evolving key execution component within a collaborative optimization system.

[0008] According to a first aspect of the present invention, an intelligent routing execution method is provided. Key steps of the method include: receiving a first optimization signal from a closed-loop learning component, the signal containing routing policy parameters generated through global collaborative optimization; generating a routing evaluation result through a collaborative evaluator based on these policy parameters and real-time context information; and finally routing the task to the target functional component. The internal logic of the collaborative evaluator is itself configured or updated by the closed-loop learning component based on a global performance objective. This ensures that the execution logic of the routing decision remains consistent with the overall optimization objective of the system.

[0009] According to a second aspect of the present invention, a corresponding intelligent routing execution device is provided. The device includes: a policy interface for receiving a global optimization policy; an information input interface for receiving real-time context information; a collaborative evaluator for collaborative evaluation based on the above two; and a task assigner for executing routes. The structural feature of this device is that the "intelligence" of its decision-making core (collaborative evaluator) originates from a closed-loop learning component, and its behavior is an extension of the global optimization will. Further, the intelligent routing execution device receives optimization instructions through a cryptographically enforced policy interface. This interface is configured to only accept and execute instructions issued by the closed-loop learning component and verified by digital signatures; any update requests that fail verification or are of unknown origin are architecturally rejected. This design, through cryptographic primitives, physically deprives the device of parameter autonomy, transforming the authority of the "logically unique decision source" into a verifiable and non-repudiable technical enforcement force, fundamentally eliminating the risk of "pseudo-cooperation" and local optimization regression.

[0010] According to a third aspect of the present invention, a collaborative optimization system incorporating the above-described apparatus is provided. This system clarifies the collaborative relationship between the intelligent routing execution device, the closed-loop learning component, and the functional components: the decision-making effect of the routing device is fed back to the learning component to drive the next round of collaborative optimization of the routing strategy and the parameters of the functional components, thereby forming a complete, self-reinforcing collaborative evolution closed loop.

[0011] (III) Beneficial technical effects By implementing the above technical solution, the present invention can achieve the following significant technical advancements: 1. Achieve precise implementation of global optimization strategies: Transform abstract optimization results (strategy parameters) into specific and reliable routing actions, ensuring that collaborative optimization decisions can be transformed from "blueprints" into "real-world scenarios".

[0012] 2. Provides high-quality optimization feedback: Since the routing decision strictly follows the optimized strategy and the execution process is traceable, the resulting effect data has high causal consistency and attributability, providing high-quality, attributable training data for the closed-loop learning component, effectively solving the "feedback inaccuracy" problem mentioned in the background technology.

[0013] 3. Improved system coordination efficiency and reliability: Efficient routing decisions are achieved through a specialized hardware / software architecture, reducing decision latency. Simultaneously, its "logically controlled, autonomous without backend" design eliminates system friction and oscillations caused by conflicts between local decisions and global objectives, thus improving the overall stability and reliability of the system.

[0014] 4. Empowering execution components to evolve: The evaluation logic within the device can be continuously updated by the global optimization process, enabling the routing execution capability to evolve in sync with the capabilities of downstream functional components, jointly driving the system performance to a higher level. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the logical location and data flow of the intelligent routing execution device of the present invention in the collaborative optimization system.

[0016] Figure 2 This is a flowchart of the intelligent routing execution method of the present invention.

[0017] Figure 3 This is a schematic diagram of three exemplary implementation structures of the internal logic of the collaborative evaluator and their optimization relationship with the closed-loop learning component.

[0018] Figure 4 This is a schematic diagram illustrating the principle comparison between the device of this invention and traditional routing components in a collaborative optimization closed loop. The diagram is a principle-based comparison curve generated based on the different behavioral logics and optimization mechanisms of the two types of components (traditional local self-tuning vs. the global collaborative execution of this invention) under the same external disturbance conditions. Its purpose is to intuitively and qualitatively reveal and compare the inherent trend differences between the two schemes in maintaining the overall stability and performance convergence of the system, rather than to present quantitative data from a specific experiment. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, a detailed description will be provided below in conjunction with the accompanying drawings and specific embodiments. Those skilled in the art should understand that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0020] [Overall Architecture and System Interaction] like Figure 1As shown, the intelligent routing execution device (200) of this invention is located at the core of the collaborative optimization system. It receives a first optimization signal (Sig1) from the closed-loop learning component (100) through a policy interface (201). This signal carries routing policy parameters (e.g., a set of weights, thresholds, or a model file) obtained through global collaborative optimization. At the same time, the device receives real-time multi-dimensional context information through an information input interface (202). The collaborative evaluator (203) integrates the policy parameters and context information and outputs the routing evaluation result. The task allocator (204) makes a final routing decision based on this and sends the request to the target functional components (300a, 300b, 300c). The execution result of the functional component and the routing decision itself are associated with a globally unique identifier (such as a request tracking ID) and recorded, and sent back to the closed-loop learning component (100) as causal feedback data to start the next round of optimization cycle.

[0021] [Internal Logic Implementation and Controlled Update Mechanism of the Collaborative Evaluator] The collaborative evaluator (203) is the core of the device. Its internal logic is the process of "instantiating" policy parameters into decision rules. For example... Figure 3 As shown, its implementation can take many forms, but all follow the principle of "logic controlled by closed-loop learning components": 1. Rule engine based on optimizable thresholds: The strategy parameters optimized by the closed-loop learning component (100) are embodied in dynamic rules. For example, the threshold T in the rule "IF request latency budget < T THEN route to edge node" is optimized and issued by the closed-loop learning component (100). The device is only responsible for executing this rule.

[0022] 2. Parameterized multi-attribute decision model: For example, a weighted scoring function `Score = w1*f1(context) + w2*f2(context)`. The weight vector `W = [w1, w2]` is the policy parameters obtained by the closed-loop learning component (100) in the composite optimization space.

[0023] 3. Machine learning classifier trained by the closed-loop learning component: The overall parameters of the classifier model (such as a neural network) are trained by the closed-loop learning component (100) using its global historical data, with the goal of directly optimizing the global performance evaluation value. The trained model parameters are sent as policy parameters to the device (200) for loading.

[0024] All updates to the above logic are performed through the policy interface (201). This interface is designed to accept only digitally signed instructions from the closed-loop learning component (100), thereby architecturally depriving the device of parameter autonomy and ensuring that its behavior is absolutely consistent with the global objective.

[0025] [Cooperative Implementation Mechanism of Standardized System State Interfaces] The "standardized system status interface" (corresponding to information input interface 202) is a key technical design for achieving deep integration between the routing component and the global collaborative optimization architecture. Its core lies in the fact that the data semantics, format specifications, and update mechanism of the "multi-dimensional context information" received by this interface are not defined by the routing component itself, but are uniformly managed by the closed-loop learning component (100) which serves as the source of global optimization decisions.

[0026] 1. Enforcement of Data Source Homology: The closed-loop learning component (100) maintains and publishes a unified set of contextual information metadata specifications. This specification explicitly defines all state attributes available for routing decisions, including but not limited to: attribute name, data type, value range, collection frequency, normalization method, and semantic label. This specification maintains strict consistency with the data model upon which the closed-loop learning component itself relies to construct the "same effect feedback dataset" and calculate the "single global performance evaluation value," thereby ensuring data source homogeneity between optimization decisions and execution awareness from the root.

[0027] 2. Integration with the parent architecture: In a preferred embodiment, this specification management function can be implemented by integrating with or calling a parameter federation management system (or similar service) as described in the parent specification. This system, serving as the foundation for the "composite optimization space," is responsible for aggregating and standardizing configurable parameters and status indicators across the entire system. During startup or runtime, the routing component dynamically obtains or subscribes to the latest contextual information specifications from this service, ensuring that its perception dimension is synchronized with the global optimization view in real time.

[0028] 3. Dynamic Synchronization and Evolution Mechanism: When system business changes or functional components iterate, the required contextual information may change. The closed-loop learning component (100) can dynamically adjust and release new metadata specification versions based on optimization goals. The routing component (200) receives relevant update instructions through its policy interface (201) and refreshes the parsing logic of its information input interface (202) accordingly. This mechanism ensures that the "system state profile" on which routing decisions depend can always evolve in tandem with the "state pattern" that the global optimization process focuses on.

[0029] Through the above design, the standardized interface ensures that routing decisions are no longer based on local, heterogeneous monitoring data, but on an authoritative system state view that is isomorphic and of the same origin as the global optimization target. This fundamentally eliminates the optimization friction caused by "state understanding bias" in traditional systems at the data input level. It is a prerequisite for the routing component to accurately execute collaborative optimization strategies and ensures that its decision-making perspective is absolutely aligned with the global optimization center.

[0030] [Cryptographic Enforcement Mechanism for Parameter Writing Interface] The "parameter write interface" (corresponding to policy interface 201) is a key security barrier ensuring that the routing device (200) is absolutely controlled by the closed-loop learning component (100). Its characteristic of "only accepting and executing update instructions issued by the closed-loop learning component and verified by cryptographic signatures" is achieved through the following technical mechanism: 1. Command Issuance and Signature Generation: After the closed-loop learning component (100) generates the first optimized signal (containing routing policy parameters), it digitally signs the signal using its private key. The signature algorithm can be an industry-standard asymmetric encryption algorithm (such as RSA, ECDSA, etc.). The signing process includes elements to prevent replay attacks, such as command content, timestamp, and sequence number. The signed command and the original command together constitute a complete and verifiable update command package.

[0031] 2. Interface Layer Mandatory Verification Logic: The policy interface (201) of the routing device (200) integrates a mandatory verification module at the firmware or software level. This module is pre-configured with a public key certificate corresponding to the private key of the closed-loop learning component (100). For any incoming update command, the verification module must perform the following checks: (a) verify the integrity of the command packet structure; (b) verify the validity of the digital signature using the pre-configured public key; (c) verify the freshness of the timestamp or sequence number to prevent command replay; (d) optionally, verify the legitimacy of the issuer's certificate chain.

[0032] 3. Rejection and Audit Mechanism: Any instruction that fails any of the above verifications will be immediately rejected by the interface and will not enter the execution process. Simultaneously, such rejection events, along with information such as the instruction's source and the reason for failure, will be recorded in an immutable security audit log for post-event traceability and compliance verification.

[0033] 4. Integration with Hardware Security Module (Preferred Embodiment): In scenarios with higher security requirements, the private key of the closed-loop learning component (100) can be stored in the Hardware Security Module (HSM) or Trusted Platform Module (TPM) to ensure that the private key cannot be exported. Correspondingly, the verification module of the routing device (200) can also be implemented based on the HSM or a trusted execution environment with secure boot capability to prevent the verification logic from being bypassed or tampered with.

[0034] Through the aforementioned cryptographic mandatory verification mechanism, the technical basis is ensured that the parameter update rights of the routing device (200) belong entirely and exclusively to the closed-loop learning component (100). This essentially transforms the architectural principle of "logically unique decision source" into an undeniable technical enforcement, completely stripping the device of its parameter autonomy at the physical interface level, making it a purely controlled execution terminal.

[0035] Example 1: Cloud-Edge Collaborative Video Analytics System In this embodiment, the system needs to handle a large number of real-time video stream analysis requests. The functional components include a lightweight model deployed at the edge (handling low-latency requests) and a fine-grained model deployed in the cloud (handling high-precision requests).

[0036] The closed-loop learning component (100) calculates a single global performance evaluation value (G) based on metrics such as bandwidth cost, analysis latency, and accuracy, and optimizes routing strategy parameters: an edge routing threshold T_edge and a set of edge node weights.

[0037] The intelligent routing execution device (200) receives T_edge and weight parameters. For each video stream request, the device reads the latency requirement and data size (context) it carries, and applies the rule: "IF Request latency requirement < T_edge THEN Select an edge node according to the weight; ELSE Route to the cloud".

[0038] • After a period of time, the network conditions change, and the closed-loop learning component (100) optimizes T_edge, for example, from 100ms to 80ms based on the new feedback data, and updates the parameters of the device (200) and the cloud model via atomic publishing. The device's behavior immediately adapts without human intervention.

[0039] [Example 2: Integration of Microservice API Gateway and Enforcement Constraints] This embodiment demonstrates the operation of the device in a highly reliable system that complies with mandatory architectural specifications.

[0040] The device (200) is implemented as a smart agent component (e.g., an enhanced version of the Envoy Sidecar) in a service mesh architecture within a container orchestration platform (e.g., Kubernetes).

[0041] All its routing rules (such as the proportion of canary publishing traffic based on the `x-user-tier` request header) are configured to be read-only. This read-only status is enforced through security policies during the container image's build phase or deployment. Updates are only permitted via a controlled "Configuration Push API".

[0042] The closed-loop learning component (100) runs in a secure environment with TPM. When it decides to adjust the "canary traffic ratio from 5% to 10%", the decision and the corresponding microservice replica count adjustment instruction are packaged into an atomic transaction, digitally signed, and then published.

[0043] Upon receiving this transaction packet, the configuration push API of device (200) first verifies the signature to confirm that the source is a legitimate closed-loop learning component (100), and then atomically updates the local routing configuration. This verification process specifically implements the cryptographic enforcement verification logic described in the aforementioned [0017B] section. Any attempt to modify the configuration directly through the platform's configuration management objects (such as Kubernetes ConfigMap) or local files will be rejected or overwritten.

[0044] [Example 3: Comparison of the underlying principles and effects with traditional routing components] like Figure 4 As shown, to illustrate the key role of the device of the present invention in the collaborative optimization closed loop from a principle perspective, this embodiment constructs a principle-based comparative analysis: Comparison Logic and Prerequisites: For the purpose of principle analysis, assuming the same system architecture and initial state, we will examine the logical behavior of the two routing mechanisms respectively: The first mechanism (corresponding to the traditional solution): adopt a traditional smart router with local self-adjustment capabilities (such as a load balancer that can dynamically adjust weights based on its own monitored latency).

[0045] The second mechanism (corresponding to the present invention): employing the intelligent routing execution device (200) of the present invention.

[0046] To clearly demonstrate the core principles, Figure 4 The curve is illustrative data generated based on the following behavioral logic: Traditional routers have an inherent conflict between their local optimization goal (reducing the latency they perceive) and their global goal (balancing latency and cost); while the device of the present invention strictly follows the strategy generated by global optimization.

[0047] Explanation of the principle and effect (corresponding) Figure 4 (Trend of the curve): When encountering a sudden flow (corresponding to the disturbance at t=4 in the figure): 1. Behavioral deduction of the first mechanism (traditional routing): Its local optimization algorithm, in response to perceived increases in latency, may frequently switch routing targets, leading to oscillations in the routing policy. These oscillations can transmit unstable loads to downstream functional components, potentially triggering a chain reaction that causes significant fluctuations and degradation in the overall system performance (G value), such as... Figure 4 As shown by the solid line in the middle.

[0048] 2. Behavioral deduction of the second mechanism (the device of this invention): Its routing strategy parameters are generated by the closed-loop learning component based on global objective optimization. When faced with the same disturbance, the device strictly and stably executes the current global optimization strategy. Simultaneously, the closed-loop learning component can, based on feedback, publish an updated cooperative strategy in the next cooperative cycle (while simultaneously adjusting routing parameters and functional component parameters). Therefore, the system performance can smoothly transition and quickly converge to a new, better state, such as... Figure 4 As shown by the dashed line.

[0049] Industrial applicability The intelligent routing execution device, method, and system provided by this invention, as a key infrastructure execution component for achieving global collaborative optimization, possess high versatility and broad industrial application prospects. Its core value lies in providing a standardized, controllable, and evolvable routing execution carrier for any distributed system or collaborative control architecture that needs to translate globally optimized strategic decisions into concrete execution actions in real time, reliably, and verifiably.

[0050] The application scenarios of this invention are defined by the following two levels, demonstrating its universality as a cross-domain foundational technology component: First level: As the core execution unit of the closed-loop collaborative optimization system This device is naturally adapted to various systems that follow a global collaborative optimization architecture (as defined by the parent design and its related sub-designs). In such systems, this device plays a dual role as an "intelligent executor" and a "precise feedback sensor" in the collaborative optimization loop. It receives collaborative optimization instructions from the global optimization hub (closed-loop learning component) and completes intelligent routing of requests, tasks, or data within milliseconds, while providing high-quality, causally attributable feedback on the execution effect of the optimization loop. This mode is the advanced and preferred operating mode of the device of this invention, which can systematically eradicate the problems of "optimization internal friction" and "feedback inaccuracy" caused by the separation of decision-making and execution in complex systems.

[0051] Second level: Enhanced implementation as an independent intelligent router with collaborative potential Even in systems lacking a complete global optimization loop, this device can be deployed independently as a high-performance, externally configurable intelligent router. Its standardized state interface, modular collaborative evaluator structure, and support for atomic instructions provide seamless upgrade space for future system evolution to higher-order collaborative forms, protecting users' technological investments.

[0052] In summary, the device of this invention constitutes a key technological bridge from "strategy optimization" to "physical execution." Its industrial applications are not limited to specific industries, but broadly cover all scenarios with intelligent and collaborative needs for task scheduling, resource allocation, traffic management, or control command routing. It provides an indispensable standardized execution layer hardware / software solution for building next-generation intelligent systems with deterministic, efficient, and continuously evolving capabilities.

[0053] Legal Statement This specification provides a full, complete, and clear disclosure of the technical solutions claimed in the claims. Those skilled in the art can implement this invention and achieve the stated technical effects based on the content of this specification and common general knowledge. The scope of protection of this invention is determined by the content of the claims. The specification and drawings are used to interpret the claims but do not constitute a limitation thereof. Any equivalent substitutions or modifications based on the inventive concept should be included within the scope of protection of this invention.

Claims

1. A method for intelligent routing execution in a collaborative optimization system, characterized in that, The collaborative optimization system includes a closed-loop learning component and multiple functional components, and the method includes: Receive a first optimization signal from the closed-loop learning component, the first optimization signal containing routing policy parameters collaboratively generated by the closed-loop learning component through a unified optimization process based on the same global performance evaluation value; Receive task requests and their associated multi-dimensional context information; Based on the routing policy parameters and the multi-dimensional context information, a collaborative evaluator is used to process the data and generate a routing evaluation result. Based on the routing evaluation results, a target functional component is selected from the plurality of functional components, and the task request is routed to the target functional component. The internal logic of the collaborative evaluator is configured or updated by the closed-loop learning component based on the optimization of the single global performance evaluation value.

2. The method according to claim 1, characterized in that, The internal logic of the collaborative evaluator is embodied in at least one of the following optimized structures: (a) A rule engine based on an optimizable threshold, wherein the threshold parameter in the rule conditions is used as part of the routing strategy parameters and is optimized by the closed-loop learning component; (b) A parameterized multi-attribute decision model, wherein the weight vector or function parameters used to fuse multi-dimensional information are used as part of the routing strategy parameters and are optimized by the closed-loop learning component; (c) A machine learning classifier whose model parameters are trained and optimized by the closed-loop learning component based on historical performance data with causal consistency.

3. The method according to claim 1 or 2, characterized in that, The multi-dimensional context information is received through a standardized system state interface; the data semantics and format specifications followed by the interface are uniformly managed, published and synchronized by the closed-loop learning component, which is the logically unique global optimization decision source, to ensure that the context information on which the routing decision is based is of the same source and isomorphic as the state data used by the closed-loop learning component for optimization calculation.

4. The method according to claim 1, characterized in that, It also includes a feedback step: providing the routing decision and execution result of the task request as effect feedback data to the closed-loop learning component for the next round of collaborative optimization.

5. An intelligent routing execution device in a collaborative optimization system, characterized in that, include: The policy interface is configured to receive a first optimization signal from the closed-loop learning component, the first optimization signal containing routing policy parameters collaboratively generated by the closed-loop learning component through a unified optimization process based on the same global performance evaluation value. The information input interface is configured to receive task requests and associated multi-dimensional context information. A collaborative evaluator, coupled to the policy interface and the information input interface, is configured to process the routing policy parameters and the multi-dimensional context information to generate a routing evaluation result. A task assigner, coupled to the collaborative evaluator, is configured to output a decision to route the task request to a target functional component based on the routing evaluation result; The internal logic of the collaborative evaluator is configured or updated by the closed-loop learning component based on the optimization of the single global performance evaluation value.

6. The apparatus according to claim 5, characterized in that, The internal logic of the collaborative evaluator is embodied in at least one of the following optimized structures: (a) A weighted fusion computing unit, wherein the weight coefficients of each dimension are determined by the closed-loop learning component based on global performance evaluation optimization; (b) A rule engine based on optimizable thresholds, wherein the threshold parameters in the rules are determined by the closed-loop learning component based on global performance evaluation; (c) A machine learning model inference engine, whose model parameters are obtained by the closed-loop learning component through optimization training based on global performance evaluation.

7. The apparatus according to claim 5, characterized in that, The information input interface is a standardized system status interface, and the metadata specifications of the context information attributes it receives are uniformly managed and published by the closed-loop learning component.

8. The apparatus according to claim 5, characterized in that, The parameter write interface exposed by the device is configured to only accept and execute update instructions issued by the closed-loop learning component and verified by cryptographic signature.

9. A collaborative optimization system, characterized in that, include: The intelligent routing execution device as described in any one of claims 5 to 8; The closed-loop learning component is configured to collaboratively generate the first optimization signal and a second optimization signal for updating configurable parameters within the plurality of functional components; Multiple functional components, whose internal configurable parameters are updated by the second optimization signal; The routing decision performance of the intelligent routing execution device, as part of the performance feedback data, is fed back to the closed-loop learning component for the next round of collaborative optimization.

10. The system according to claim 9, characterized in that, The first optimization signal and the second optimization signal are encapsulated in the same atomic publishing unit and synchronously published to the intelligent routing execution device and the functional component in a transactional manner.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 4.

Images