Edge calculation optimization method, system and device based on double-loop feedback iteration

Abstract

The invention discloses an edge calculation optimization method, system and device based on double-loop feedback iteration. The method comprises the following steps: firstly, setting a work target based on cross-scene task feature transfer learning through outer circulation, and decomposing the work target into work tasks to be distributed to an edge node group; secondly, working tasks are executed through internal circulation, and execution data are collected in real time; feedback analysis is carried out based on the execution data, deviation types are judged, and inner loop iteration or outer loop iteration is carried out; and finally, simulating an extreme working condition through an iteration effect mirror image evaluation mechanism, and correcting an iteration direction. According to the edge calculation optimization method based on the double-loop feedback iteration model, efficient and self-adaptive optimization of the edge calculation environment can be achieved, and safety, reliability and efficiency are improved.

Description

Technical Field

[0001] This invention belongs to the field of edge computing, specifically relating to an edge computing optimization method, system, and device based on dual-loop feedback iteration. Background Technology

[0002] Edge computing, as an important extension of cloud computing, effectively reduces latency, improves response speed, and supports real-time applications such as the Internet of Things (IoT) and smart homes by offloading computing tasks to the network edge. However, with the expansion of edge computing scale and the diversification of application scenarios, the edge environment faces many challenges: First, the heterogeneous and widely distributed resources of edge nodes lead to low efficiency in task allocation and collaboration, and are prone to load imbalance; second, the edge environment is dynamic and changeable, with frequent extreme conditions such as network latency and security attacks, which existing static optimization methods are difficult to adapt to; third, insufficient cross-scenario task migration capabilities result in high repetitive learning costs and a large deviation between work objectives and actual execution.

[0003] Existing edge computing optimization methods mostly focus on a single dimension, such as load balancing or security hardening, lacking an overall collaborative mechanism. For example, some methods employ centralized scheduling strategies, but cannot address the autonomy requirements of edge nodes; others rely on machine learning for prediction, but the models are rigid and difficult to adjust in real time. Furthermore, existing methods lack effective feedback and iteration mechanisms, failing to distinguish between tactical execution problems and deviations from work objectives, resulting in limited optimization effectiveness. Therefore, there is an urgent need to propose an edge computing optimization method that can consider both strategic and tactical aspects and possess adaptive iterative capabilities. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide an edge computing optimization method, system and device based on dual-loop feedback iteration.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: An edge computing optimization method based on dual-loop feedback iteration includes the following steps: The work objectives are set through the outer loop, and then broken down into multiple work tasks and assigned to the corresponding nodes of the edge node group. The task is executed through an inner loop, and execution data of the edge node group is collected in real time. The execution data includes security performance, reliability indicators and efficiency data. The execution data is analyzed to obtain and judge the execution deviation. If the execution deviation is due to tactical problems, the inner loop iteration is entered. The iteration results trigger the inner loop ant colony heuristic weight adjustment to optimize the collaborative weight between pairs of edge nodes. If the deviation in execution stems from strategic issues, then an external loop iteration is initiated, employing an external loop restructuring strategy to calibrate work objectives and re-decompose work tasks. A virtual twin scenario is constructed to simulate the optimization effect under extreme working conditions. The direction of the iterative cycle is corrected by comparing the virtual and real environments, and the continuous iterative optimization of the edge computing environment is achieved through the direction of the iterative cycle.

[0006] As one possible implementation, the inner-loop ant colony heuristic weight adjustment mechanism includes the following steps: Based on the security performance, reliability indicators, and efficiency data of edge nodes, the corresponding pheromone concentration is obtained. By autonomously optimizing the collaborative weights between pairs of edge nodes through pheromone concentration, if an edge node exhibits better performance in a specific task, the pheromone concentration is increased, resulting in task-appropriate bias. For edge nodes with high risk fluctuation frequency, the weight ratio is automatically reduced through the pheromone evaporation mechanism to achieve distributed self-optimization.

[0007] As one possible implementation method, the external circulation recombination strategy includes the following steps: The task characteristics in different scenarios are broken down into basic units, wherein the basic units include safety factors, reliability factors and efficiency factors; New feature combinations are generated based on basic units and through cross-fusion, wherein the cross-fusion method includes biological gene recombination; A historical feature gene library is constructed. When new task types or scenario changes occur in the edge computing environment, the historical feature gene library is retrieved through new feature combinations to obtain feature models that adapt to the new requirements, and then fed back to the weight adjustment mechanism of the inner loop.

[0008] As one possible implementation method, the construction of a virtual twin scenario to simulate the optimization effect under extreme working conditions, and the correction of the direction of iterative iteration through virtual-real comparison, includes the following steps: A virtual twin scenario is constructed for each iteration, and extreme working conditions are simulated in the virtual twin scenario, including high load, network latency, and security attacks; The optimization effect is evaluated by comparing the simulation results with the real data, and the iteration direction is continuously corrected to ensure that the collaborative weighting maintains forward-looking adaptability to complex edge environments. The collaborative weighting is dynamically weighted by the execution data of the edge nodes, and the optimal performance solution is achieved through optimization.

[0009] As one possible implementation method, the step of performing feedback analysis based on execution data to obtain and determine execution deviations includes the following steps: Judgment is made based on execution deviation. If the tactical execution does not meet expectations but the work objectives are still reasonable, it is judged as a tactical problem and enters the inner loop iteration. If tactical execution is satisfactory but work objectives are not achieved, it is judged to be a strategic problem and enters the external loop iteration; The execution deviations include feature usage rate, user operation time, and number of negative feedback reviews.

[0010] As one possible implementation, the work objective is set through an outer loop based on cross-scenario task feature transfer learning, including extracting task feature genes from historical scenarios and adapting them to new scenarios through transfer learning, thereby reducing the cost of repeated learning.

[0011] As one possible implementation, the inner loop executes work tasks in a distributed autonomous manner, obtains weight status through collaborative weighting, and synchronizes pheromone concentration and weight status among edge nodes through a consensus protocol.

[0012] An edge computing optimization system based on dual-loop feedback iteration includes a target decomposition module, a data collection module, an inner loop module, an outer loop module, and an iterative evaluation module. The target decomposition module sets the work target through an outer loop, decomposes the work target into multiple work tasks, and assigns them to the corresponding nodes of the edge node group. The data collection module executes the work task through an internal loop and collects the execution data of the edge node group in real time. The execution data includes security performance, reliability indicators and efficiency data. The inner loop module analyzes the execution data to obtain and judge the execution deviation. If the execution deviation is caused by tactical problems, it enters the inner loop iteration. The iteration result triggers the inner loop ant colony heuristic weight adjustment to optimize the collaborative weight between pairs of edge nodes. If the deviation in the execution of the outer loop module originates from a strategic problem, it will enter the outer loop iteration, adopt the outer loop reorganization strategy to calibrate the work objectives and re-decompose the work tasks. The iterative evaluation module constructs a virtual twin scenario to simulate the optimization effect under extreme working conditions, corrects the direction of iterative iteration through virtual-real comparison, and achieves continuous iterative optimization of the edge computing environment through the direction of iterative iteration.

[0013] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the following method: The work objectives are set through the outer loop, and then broken down into multiple work tasks and assigned to the corresponding nodes of the edge node group. The task is executed through an inner loop, and execution data of the edge node group is collected in real time. The execution data includes security performance, reliability indicators and efficiency data. The execution data is analyzed to obtain and judge the execution deviation. If the execution deviation is due to tactical problems, the inner loop iteration is entered. The iteration results trigger the inner loop ant colony heuristic weight adjustment to optimize the collaborative weight between pairs of edge nodes. If the deviation in execution stems from strategic issues, then an external loop iteration is initiated, employing an external loop restructuring strategy to calibrate work objectives and re-decompose work tasks. A virtual twin scenario is constructed to simulate the optimization effect under extreme working conditions. The direction of the iterative cycle is corrected by comparing the virtual and real environments, and the continuous iterative optimization of the edge computing environment is achieved through the direction of the iterative cycle.

[0014] An edge computing optimization device based on dual-loop feedback iteration includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the following method: The work objectives are set through the outer loop, and then broken down into multiple work tasks and assigned to the corresponding nodes of the edge node group. The task is executed through an inner loop, and execution data of the edge node group is collected in real time. The execution data includes security performance, reliability indicators and efficiency data. The execution data is analyzed to obtain and judge the execution deviation. If the execution deviation is due to tactical problems, the inner loop iteration is entered. The iteration results trigger the inner loop ant colony heuristic weight adjustment to optimize the collaborative weight between pairs of edge nodes. If the deviation in execution stems from strategic issues, then an external loop iteration is initiated, employing an external loop restructuring strategy to calibrate work objectives and re-decompose work tasks. A virtual twin scenario is constructed to simulate the optimization effect under extreme working conditions. The direction of the iterative cycle is corrected by comparing the virtual and real environments, and the continuous iterative optimization of the edge computing environment is achieved through the direction of the iterative cycle.

[0015] The present invention has at least the following beneficial effects: This invention separates work objectives from tactical execution through a dual-loop feedback iterative model, achieving hierarchical optimization of the edge computing environment. This effectively distinguishes and resolves the discrepancy between tactical coordination issues and work objectives, improving the accuracy and efficiency of optimization. Employing an inner-loop ant colony heuristic weight adjustment mechanism, it achieves distributed self-optimization among edge nodes, eliminating the need for centralized control, reducing system complexity, and enhancing the ability to cope with dynamic environments. By introducing an outer-loop feature gene recombination strategy and enabling cross-scenario task feature transfer learning, it quickly adapts to new scenarios, reducing repetitive learning costs and improving the system's generalization ability. Through an iterative effect mirror evaluation mechanism, it simulates extreme conditions for forward-looking testing, ensuring the reliability of the optimization scheme in real-world environments. Overall, this invention achieves continuous adaptive optimization of the edge computing environment, significantly improving security, reliability, and resource utilization efficiency, providing technical support for large-scale edge computing applications. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the modules of the system of the present invention; Figure 3 This is a diagram of the overall architecture of the present invention. Detailed Implementation

[0017] To clearly illustrate the present invention and make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings, so that those skilled in the art can implement the invention based on the description. The technology of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment takes the edge computing environment of a smart factory as an example to illustrate the implementation process of the edge computing optimization method based on a dual-loop feedback iterative model described in the present invention. Those skilled in the art will understand that similar implementations can be made in other edge computing scenarios.

[0019] Example 1: An edge computing optimization method based on dual-loop feedback iteration, such as Figure 1 As shown, it includes the following steps: S100: Set work objectives through the outer loop, decompose the work objectives into multiple work tasks, and assign them to the corresponding nodes of the edge node group; S200. The work task is executed through an inner loop, and the execution data of the edge node group is collected in real time, wherein the execution data includes security performance, reliability indicators and efficiency data; S300: Perform feedback analysis on the execution data, obtain and judge the execution deviation. If the execution deviation is caused by tactical problems, enter the inner loop iteration. Trigger the inner loop ant colony heuristic weight adjustment through the iteration results to optimize the collaborative weight between pairs of edge nodes. S400. If the execution deviation stems from strategic issues, then enter the outer loop iteration, adopt the outer loop restructuring strategy to calibrate work objectives and re-decompose work tasks; S500 constructs a virtual twin scenario to simulate the optimization effect under extreme working conditions, and corrects the direction of iterative iteration through virtual-real comparison, and realizes continuous iterative optimization of the edge computing environment through the direction of iterative iteration.

[0020] like Figure 3 The overall framework diagram of this invention shown illustrates that the system achieves continuous optimization through a dual-loop feedback iterative model. This model comprises four core modules: an outer loop module, responsible for setting and calibrating long-term, low-frequency work objectives; an inner loop module, responsible for executing and rapidly adjusting short-term, high-frequency work tasks; a feedback analysis module, responsible for quantitatively analyzing execution data and determining whether the root cause of the problem lies at the tactical or strategic level; and an iterative evaluation module, responsible for proactively evaluating and correcting the optimization scheme for each iteration using virtual twin technology.

[0021] The dual-loop iterative feedback model achieves continuous optimization of edge computing across all levels, from tactical to strategic, through the division of labor and linkage between the inner and outer loops, combined with data-driven and virtual verification mechanisms. The inner loop focuses on the tactical execution layer, characterized by "high-frequency and rapid adjustment." Starting from the initiation of work tasks, it optimizes tactical execution strategies such as edge node collaborative weights through data analysis and iteration, and data feedback. The outer loop focuses on the strategic direction layer, characterized by "low-frequency and deep calibration." It conducts data analysis and iteration around work objectives, and after verification through digital twin virtual scenarios, it calibrates work objectives and reconstructs work tasks.

[0022] The core linkage layer carries the collaborative logic of a dual-loop system: on the one hand, it distinguishes between "execution-level (tactical) problems" and "direction-level (strategic) problems," optimizing and calibrating tactics or strategies around dimensions such as objectives and algorithm tools; on the other hand, through the periodic division of "short-cycle (internal loop) high-frequency adjustment" and "long-cycle (external loop) deep calibration," it relies on functional modules such as internal / external loop modules, iterative evaluation, data collection, and feedback analysis for support. Simultaneously, it leverages virtual twin scenarios to extract task features, optimize node collaborative weights, and correct iterative directions through comparison of virtual and real data. Ultimately, a continuously evolving ecosystem is formed through "four-step closed loop + dual-loop linkage," allowing the internal and external loops to mutually nourish each other (internal loop tactical optimization feeds back into strategic feasibility, while external loop strategic adjustments provide new directions for the internal loop), achieving full-level coverage from tactical execution details to strategic direction adaptation, dynamically responding to scenario changes, and continuously improving system efficiency and adaptability. Specific implementation steps are shown below.

[0023] Specifically, such as Figure 1 As shown, step S100 involves setting the work objective and decomposing the work tasks. The work objective is to extract task feature genes from historical scenarios based on cross-scenario task feature transfer learning. For example, task features, including safety factors, reliability factors, and efficiency factors, are extracted from historical data. This process is called the construction and analysis of the historical feature gene library. The specific implementation process is as follows: a historical feature gene library is constructed based on the collaborative construction of task features and task feature genes, with the two being interdependent to form a complete system. The construction begins with the systematic decomposition of task features, decomposing task features under different scenarios into basic units such as safety factors, reliability factors, and efficiency factors. These units are then abstracted into feature genes, which are stored and managed through standardized coding methods. Each feature gene corresponds to the core attribute or performance dimension of the task.

[0024] The construction process encompasses feature extraction, gene encoding, gene storage, and gene fusion: key features are extracted from historical task data and transformed into standardized feature genes, which are then uniformly encoded and stored in a structured database. Finally, through biological gene recombination, genetic algorithms, and other means, the feature genes are cross-fused to generate new feature combinations.

[0025] This gene bank features dynamic expansion capabilities. Upon completion of a new task, it automatically extracts and adds feature genes to the database. Combined with an external loop transfer learning mechanism, it achieves cross-scenario adaptation and optimization. When encountering a new task, it retrieves similar feature models from the database using a feature matching algorithm. Task features are the external representation, describing the specific attributes and requirements of the task; feature genes are the internal encoding, supporting efficient storage, retrieval, and genetic operations. The two are rapidly adapted to tasks through a bidirectional "feature-gene-feature" conversion. The overall system follows a path of "feature decomposition - gene encoding - fusion evolution - retrieval application," forming an intelligent knowledge base that supports edge computing task optimization.

[0026] Based on the above, task feature genes and task features are not the same concept; rather, they are strongly related through a "origin-abstraction-application" chain. Task features are the concrete representation of a task, describing its observable attributes and required indicators in different scenarios, exhibiting scenario-specific characteristics. Task feature genes are the abstract encoding form of task features, formed by decomposing concrete task features into basic units such as safety factors, reliability factors, and efficiency factors, and then structurally encoding them into universally operable minimum functional units. Logically, task feature genes originate from task features and must undergo a "decomposition-encoding" transformation before being stored, fused, and retrieved in the historical feature gene database. In application, new feature combinations are generated through cross-fusion methods such as biological gene recombination. Combined with cross-scenario transfer learning mechanisms, specific task feature requirements can be quickly mapped through feature gene matching, achieving cross-scenario task adaptation. This relationship is analogous to the correspondence between biological traits and genes, echoing the various technical methods mentioned above.

[0027] By employing transfer learning algorithms, these features are adapted to new scenarios, reducing the cost of repetitive learning. The objective is set as "reducing the average task response time while ensuring safety and reliability." This step is executed in the outer loop module, and its key feature is setting the objective based on cross-scenario task feature transfer learning, ensuring that the system can leverage historical experience to reduce the cost of repetitive learning.

[0028] The aforementioned work objective refers to the system extracting transferable "task feature genes" from historical personnel data. These genes are broken down into multiple quantifiable basic factors, including security factors, encryption algorithm strength, and intrusion detection sensitivity. This process involves transforming the security requirements of edge computing tasks into quantifiable and coded feature units, integrating them into the entire construction process of the historical feature gene library. Firstly, in the task feature decomposition stage, for the security requirements of industrial control and intelligent security tasks in different edge computing scenarios, the abstract "security assurance requirements" are concretely decomposed into the core constituent dimensions of security factors: encryption algorithm strength and intrusion detection sensitivity. These two are clearly defined as the core quantitative support for security factors, jointly defining the security protection level of the task.

[0029] The next step is the gene encoding stage, which standardizes and abstracts parameters such as encryption algorithm strength, algorithm type adaptability, key protection capability, intrusion detection sensitivity, abnormal behavior recognition threshold, and response time. These parameters are then combined with the characteristics of the task scenario to assign corresponding attribute tags and transform them into a unified feature gene encoding form, ensuring that security-related features in different scenarios can be uniformly identified and processed by the system.

[0030] During the gene storage and fusion stage, these encoded basic factors will serve as the core gene units of the security dimension, be stored in a structured database and establish association relationships, and form an association mapping with other basic units such as reliability factors and efficiency factors. During the fusion process, through algorithms such as biological gene recombination, the encryption algorithm strength and intrusion detection sensitivity feature genes under different scenarios are cross-combined to generate new feature combinations that adapt to complex security needs.

[0031] When expanding dynamically in the future, after a new task is completed, the system will automatically extract its actual performance data in terms of security dimensions, such as the stability of the encryption algorithm and the accuracy of intrusion detection; it will also reverse-optimize the feature gene weights corresponding to the encryption algorithm strength and intrusion detection sensitivity and add them to the database; when a task with new security requirements is encountered, the system will quickly adapt and form a targeted security protection scheme by searching for the feature genes of the corresponding basic factors in the feature matching database, so as to achieve a precise match between security requirements and edge computing tasks.

[0032] The parameters in this invention are processed as follows: Reliability factors: task redundancy, node health status. Efficiency factors: such as data processing latency, resource consumption rate. Through transfer learning algorithms, these factors and their weight relationships are adapted to the new smart factory scenario. For example, the initial working objective set in this embodiment is: "To reduce the average response time of image processing tasks by 30% while ensuring a safety score of not less than 0.9 and a reliability score of not less than 0.95." Or, "To reduce the average response time of image processing tasks by 17% while ensuring a safety score of not less than 0.8 and a reliability score of not less than 0.90," etc.

[0033] In this invention, task decomposition, as described above, refers to breaking down the work objective into multiple tasks and distributing them to a group of edge nodes. For example, the tasks can be decomposed as follows: Task A: Image processing task, assigned to edge nodes Node1 and Node2. Task B: Data aggregation task, assigned to edge nodes Node3 and Node4. The decomposition considers the resource capabilities and geographical location of each node. Specifically, it comprehensively considers the computing power, storage resources, network bandwidth, and geographical location of each node.

[0034] Specifically, such as Figure 1As shown, in step S200, the inner loop executes the work tasks and collects execution data. The task execution edge node group executes the work tasks in a distributed autonomous manner. Nodes synchronize state information through a consensus protocol, eliminating the need for centralized control. This step is executed autonomously at the execution layer by the inner loop module. After receiving the task, the edge node group synchronizes its state through the consensus protocol, enabling collaborative work without the intervention of a central controller. While each node is executing its task, each node independently executes its assigned task and collects execution data in real time, gathering multi-dimensional execution data, including: Security performance: such as the number of attack detections, the number of data breaches, the number of attack interceptions, and the frequency of security log alerts.

[0035] Reliability metrics include task success rate, node availability, and node online rate.

[0036] Efficiency data: such as task processing delays and resource utilization.

[0037] This data is transmitted to the feedback analysis module via a lightweight message queue, where it undergoes preprocessing and other operations.

[0038] Specifically, such as Figure 1 As shown, step S300 involves feedback analysis to determine the problem type, performing quantitative analysis based on execution data to determine whether the execution deviation stems from a tactical or strategic problem. The specific judgment logic is as follows: Tactical problem assessment: If tactical execution fails to meet expectations, but the work objectives remain reasonable, it is assessed as a tactical problem. Quantitative data includes feature usage rate, user operation time, and the number of negative feedback reviews.

[0039] Strategic problem assessment: If tactical execution is successful but the work objectives are not achieved, it is assessed as a strategic problem. For example, due to changes in the scenario, the original feature model is no longer applicable.

[0040] The judgment process uses a decision tree algorithm, with thresholds trained based on historical data. In this invention, the core of step S300 is to distinguish between "tactical problems" and "strategic problems" through quantitative indicators and the decision tree algorithm. Essentially, it uses cross-validation of "tactical execution results" and "work goal achievement" to pinpoint the root cause of deviations. The logical framework can be simplified as: Tactical problem = tactical execution not meeting expectations + work goals still reasonable; Strategic problem = tactical execution meeting standards + work goals not achieved. The quantitative formulas for defining core indicators require first converting "tactical execution" and "work goals" into calculable quantitative indicators. Based on user-mentioned "feature usage rate, user operation time duration, and number of negative feedback reviews," strategic-level indicators are supplemented, as shown below: Tactical execution metrics are used to measure whether the tactics have achieved their intended goals. Preset thresholds (obtained through training with historical data) are used to define the expected targets for tactical execution. Actual metrics are compared to these thresholds to determine if the targets have been met. Among these metrics is the function utilization rate. The formula reflecting user acceptance of tactical-related functions is: Expected threshold: (Historical minimum usage rate for tactical execution to meet targets). User operation time. The efficiency of the tactical process is reflected in its duration; the shorter the duration, the better. The formula is: Where n is the number of operating users and the expected threshold The maximum allowable duration for tactical execution to meet targets historically. Feedback negative review rate. This reflects user satisfaction with tactical execution; a lower negative review rate is better. The formula is: Expected threshold: The maximum allowable negative feedback rate for tactical execution in history.

[0041] Work objective-related indicators are used to measure whether "work objectives are reasonable / achievable." Work objectives are the ultimate tactical direction, such as "improving user retention rate" or "expanding market share." Macro-level indicators are needed to measure their achievement, and the expected value of setting work objectives should be set. The actual indicators are The degree of achievement of the work objectives is as follows: The formula is: For example, if the work objective is "monthly retention rate exceeding 30%", and the actual retention rate is 25%, then... Expected threshold: Historically, the minimum degree of achievement required to determine that "work objectives are still reasonable".

[0042] The above formula is used to comprehensively judge whether the tactical execution meets the standards. A single indicator may have biases; for example, the usage rate may meet the standard, but the negative review rate may be too high. Therefore, a weighted comprehensive score is needed to determine whether the overall tactical execution meets the standards. In this invention, weighting refers to a collaborative weighting mechanism, which is the core resource allocation and cooperation strategy for the efficient collaborative execution of tasks by edge node groups. Its core essence is the deep integration of multiple technical features. This mechanism dynamically assigns weights based on the real-time security performance, reliability indicators, and efficiency data of edge nodes. Through weight adjustment, it optimizes resource scheduling, task distribution, and data transmission to achieve optimal overall performance of the node group. Technically, the weight adjustment uses the pheromone concentration of the ant colony heuristic algorithm as input, possessing the dynamic optimization characteristics of internal loop feedback iteration. Nodes with better performance have higher pheromone concentrations, resulting in increased weights; nodes with greater risk fluctuations have their weight ratio reduced through a pheromone evaporation mechanism. Simultaneously, relying on a distributed autonomous mode, nodes synchronize their weight states through a consensus protocol, ensuring consistent collaborative cognition without centralized control. This effectively solves the collaboration efficiency problems caused by the performance heterogeneity of edge computing nodes and the dynamic nature of scenarios, while balancing the security and reliability of task execution. Tactical Execution Comprehensive Score The formula is: ,in The weights are trained using historical data to reflect the degree of influence of each indicator on tactical execution, such as higher feature utilization. The percentage of usage that meets the target; Operation time as a percentage of the target; The percentage of negative reviews meeting the target. Tactical execution judgment, if... The comprehensive score threshold is obtained through training on historical data. If the threshold is met, the tactical execution is deemed "acceptable"; otherwise, the tactical execution is deemed "not as expected". Decision tree model design and threshold training.

[0043] Based on the above weight description, the weight state is directly related to the collaborative weight and is the real-time dynamic representation of the collaborative weight. Its core definition is a collection of collaborative weight-related information for the edge node at the current moment, encompassing core collaborative weight data, its effective scope, update timeliness, adjustment basis, status identifier, and other additional information. The two are inextricably linked; the weight state is a "dynamic snapshot" of the collaborative weight, and the collaborative weight is the core data of the weight state. Functionally, weight state synchronization is crucial for distributed autonomous collaboration—nodes synchronize weight states through a consensus protocol, ensuring consistent understanding of collaborative weights in a distributed environment and providing a foundation for efficient task allocation. Simultaneously, real-time updates of the weight state support internal loop feedback iteration. When changes in node execution data trigger adjustments to the collaborative weight, the updated weight state is synchronized to all nodes through the consensus protocol, ensuring consistent weight adjustments across all nodes and aligning with the internal loop iteration logic.

[0044] Based on the above calculations, an automated decision tree algorithm is used to achieve the judgment, the core of which is to train each threshold using historical data. This ensures that the decision-making logic aligns with historical cases. The decision tree's structure tree is a binary tree. Root node: Whether the tactical execution met the criteria, based on... To make a judgment, branch 1: the tactical execution did not meet expectations. Then proceed to child node 1, branch 2: tactical execution achieved ( If so, proceed to child node 2; whether the working objective of child node 1 is still reasonable, based on... To make a judgment, if If the work objective is still achievable, then output the tactical problem. If it might be a "strategic + tactical dual problem," then additional handling is required. Whether the working objective of child node 2 is achieved depends on... The judgment is that 100% corresponds to complete compliance. If work objectives are not achieved, output strategic problems. Then both tactics and strategy meet the requirements, and there are no problems.

[0045] The collected historical case data did not include known "tactical issues" or "strategic issues" tags, and their corresponding... The thresholds are trained using a decision tree algorithm as described above. The goal is to find the optimal combination of thresholds that distinguishes between "tactical problems" and "strategic problems" by maximizing information increment. If in historical cases of "tactical problems", If it is generally below 60%, then It may be trained to 60%. In the case of "strategic issues" If it is generally below 80%, then It may be trained to 80%.

[0046] Based on the above processing procedure, the real-time judgment process is as follows: assuming the historical training threshold for a certain tactic is... , , , , Enter actual data. Calculate the overall score for tactical execution. , , , Negative values ​​are treated as zero. Therefore, the tactical execution did not meet expectations. Calculate the achievement value of the work objective. , If the tactical execution did not meet expectations but the work objective remains reasonable, then the problem is classified as a tactical issue. This process, through the logic of "quantitative indicator definition → comprehensive score calculation → decision tree branch judgment," transforms the abstract "tactical / strategic problem" into a computable and trainable model. The core is to calibrate thresholds using historical data to ensure the objectivity and accuracy of the judgment. The formula design takes into account the physical meaning of each indicator (e.g., shorter durations result in higher scores) and the weight differences highlighting the impact of key indicators, ultimately achieving precise identification of the root cause of the deviation.

[0047] If the feedback analysis determines it to be a tactical problem, the process enters an inner loop iteration. An inner loop ant colony heuristic weight adjustment mechanism is used to optimize the collaborative weights between edge nodes. When the feedback analysis determines it to be a "tactical problem," meaning the work objective is reasonable but the tactical execution has not met expectations, the collaborative weights between edge nodes need to be optimized through inner loop iteration; this is a tactical-level execution optimization. This mechanism borrows the core idea of ​​the ant colony algorithm: nodes transmit performance signals through "pheromone concentration," autonomously adjusting task allocation weights to ultimately achieve dynamic optimization of node collaboration efficiency. Specific implementation: The core mapping relationships are as follows: edge nodes → "paths" in the ant colony algorithm; node security, reliability, and efficiency → "advantages and disadvantages" of the path; pheromone concentration → "attractiveness" of the node to the task; collaborative weight → "probability percentage" of the task assigned to the node.

[0048] Pheromone concentration calculation is a quantitative mapping of node performance. Pheromone concentration is a core indicator for measuring the overall performance of a node and directly determines its attractiveness to tasks. Its calculation is based on three core performance dimensions of the node: security, reliability, and efficiency, and is quantified through weighted summation. The core performance indicator is defined as the security score. This measures a node's ability to withstand risks. The calculation logic is as follows: The threshold is the "maximum acceptable number of security events" based on historical statistics. If the number of events exceeds the threshold, S = 0. For example, if a node experiences 2 security events within 1 hour, and the threshold is 5, then S = 1 − 2 / 5 = 0.6.

[0049] Reliability rating This measures a node's ability to stably complete tasks, including fault-free operation and task success rate. The calculation logic is as follows: Example: If a node is assigned 100 tasks and 95 are successfully completed, then R = 95 / 100 = 0.95. Efficiency rating. This measures the speed at which a node processes tasks. The calculation logic is as follows: The threshold is the "acceptable maximum latency" based on historical statistics; if the latency exceeds the threshold, E=0. For example, if the average latency for a node to process a task is 200ms and the threshold is 500ms, then E=1−200 / 500=0.6.

[0050] The formula and calculation example for pheromone concentration: Pheromone concentration is obtained by weighted summation of three major indicators, as shown in the following formula: ,in These are weighting coefficients, reflecting the importance of each indicator, and satisfying... The pheromone concentration ranges from 0 to 1. Since S, R, and E are all 0 to 1, the weighted average remains within the 0-1 range. A higher value indicates better overall node performance. The calculation example assumes the performance data of an edge node Node1 is: S = 0.8 (low security events), R = 0.9 (high task success rate), E = 0.7 (low latency), with weighting coefficients... .

[0051] but 0.3×0.8+0.4×0.9+0.3×0.7=0.24+0.36+0.21=0.81. The pheromone update mechanism dynamically adjusts node attractiveness. The pheromone concentration is not a fixed value and needs to be dynamically updated through "evaporation" and "enhancement" mechanisms to ensure it reflects the node's recent performance, rather than its historical static performance, thereby achieving adaptive adjustment of weights. Pheromone evaporation weakens the influence of historical performance. To prevent nodes from maintaining high weights due to past excellent performance while ignoring recent performance decline, pheromones will naturally decay over time, as shown in the following formula. Where t is the current iteration period, The volatile coefficient represents the pheromone retention rate after each iteration. ×100%, attenuation 1− Function: To penalize nodes with consistently declining performance. If Node2 experiences frequent security incidents, its pheromones will continuously evaporate, reducing its attractiveness. Example of evaporation: Node2's initial pheromone level is 0.7. =0.9, after 1 round of iterations If Node 2 still does not improve, after the second round The attractiveness continues to decrease. Pheromone enhancement amplifies recent superior performance; when a node demonstrates better performance in a task, pheromones are additionally enhanced, as shown in the following formula: in, The pheromone increment is positively correlated with the node's task performance, and the calculation logic is as follows: k is the enhancement coefficient. If k=0.1, it ensures the increment does not exceed 1. (S+R+E) / 3 is the average performance score, reflecting the overall performance. For example, if Node1 completes its task in an iteration with S=0.8, R=0.9, E=0.7, and the average performance score is (0.8+0.9+0.7) / 3=0.8, k=0.1, then ΔPheromone=0.1×0.8=0.08. If Node1's pheromone level in the previous round was 0.81, then ρ=0.9.

[0052] After this update, Pheromone(t) = 0.81 × 0.9 + 0.08 = 0.729 + 0.08 = 0.809. Due to its excellent performance, the pheromone level remains relatively high. The collaborative weight adjustment is based on the pheromone-based task allocation strategy. The collaborative weight between nodes, i.e., the proportion of tasks allocated to each node, is directly determined by the pheromone concentration. Nodes with higher pheromone levels have greater weights and receive more task allocation. The collaborative weight calculation formula is as follows: Assume there are n edge nodes in the system, and the pheromone concentration of node i is... Then its collaborative weight for The weights satisfy This indicates that the total distribution of tasks across all nodes is 100%. Example of weight adjustment: The system has two nodes, Node1 has a pheromone level of 0.81, and Node2 has a pheromone level of 0.567. Therefore, the total pheromone level is 0.81 + 0.567 = 1.377. The weight of Node1 is... The weights of Node2 are 412. Subsequent image tasks will be preferentially assigned to Node1, which conforms to the logic of "the more efficient node gets more tasks".

[0053] Based on the above processing procedure, distributed self-optimization is performed. Decentralized coordination edge nodes based on the Gossip protocol are typically deployed in a distributed manner, without a central node. The Gossip protocol is needed to achieve network-wide synchronization of pheromone concentrations, ensuring consistency in weight adjustments. The core function of the Gossip protocol is information broadcasting: each node periodically and randomly selects a few neighboring nodes to send its own pheromone concentration and the pheromones observed from other nodes. Information fusion: the receiver compares the received pheromone with its locally stored pheromone, takes the latest value to update its local data, and ensures eventual consistency of the pheromone state across the entire network. Distributed process example: Node1 calculates its own pheromone level to be 0.81 and broadcasts it to its neighbors Node3 and Node4. Node2 calculates its own pheromone level to be 0.567 and broadcasts it to its neighbors Node3 and Node5. After receiving pheromones from Node1 and Node2, Node3 updates its local storage to Node1=0.81 and Node2=0.567, and broadcasts this information to its other neighbors. After several iterations, all nodes in the network are synchronized with the latest pheromones of Node1 and Node2, ensuring consistent collaborative weights calculated by each node. For example, it is assumed that the weight of Node1 is approximately 58.8%. The complete iterative process optimizes tactical issues in a closed loop. Initial state: Each node initializes its pheromone concentration based on historical data, calculates its initial collaborative weight, and assigns tasks. Task execution and data collection: After executing tasks, nodes record the number of security events, task success rate, processing latency, and calculate S, R, and E. Pheromone update: Each node's pheromone is updated according to the "evaporation + enhancement" formula. Distributed synchronization: The entire network's pheromones are broadcast and synchronized via the Gossip protocol. Weight adjustment: New collaborative weights are calculated based on the synchronized pheromones, and tasks in the next round are assigned according to the new weights. Iterative loop: The steps are repeated until tactical execution is achieved. This mechanism simulates the "pheromone communication" behavior of ant colonies to achieve decentralized self-optimization of collaborative weights for edge nodes. Nodes with superior performance automatically receive more tasks, improving overall execution efficiency; nodes with declining performance automatically have their weights reduced, reducing resource waste. It adapts to the dynamic adjustment needs of tactical issues without manual intervention, quickly responding to execution deviations.

[0054] Specifically, such as Figure 1 As shown, in step S400, the outer loop iterates the feature gene recombination strategy. If the feedback analysis determines it to be a strategic problem, the outer loop iteration begins. The outer loop feature gene recombination strategy is used to calibrate the work objectives and re-decompose the work tasks. Specific implementation includes: Feature gene decomposition: This involves decomposing task features from different scenarios into basic units, including security factors (encryption algorithm type), reliability factors (redundancy backup level), and efficiency factors (such as compression ratio). Gene recombination: This involves cross-fusion, similar to biological gene recombination, to generate new feature combinations. For example, using a genetic algorithm: selecting high-fitness feature combinations from the historical feature gene library, or selecting features for efficient image processing from the smart factory scenario. Cross-fertilization: This involves cross-fertilizing features from different scenarios, such as the real-time features of the autonomous driving scenario and the reliability features of the smart factory, to generate new feature models. Mutation: This involves randomly adjusting certain factors, such as changing the encryption algorithm, to enhance diversity. Work objective calibration: When new task types or scenario changes occur in the edge computing environment, and new VR / AR tasks are added, the system automatically retrieves the historical feature gene library to form a feature model adapted to the new requirements, and feeds back into the inner loop's weight adjustment logic. For example, the work objective is reset to "improving task throughput while ensuring real-time performance."

[0055] The task characteristics of different scenarios are broken down into basic gene units, each unit corresponding to a quantifiable core factor, which facilitates recombination. The quantification method is as follows: safety factor. Reliability factor Efficiency factor Real-time factor By drawing upon the principles of biological inheritance from the basic elements in the aforementioned basic units, new trait combinations are generated through "selection-crossover-mutation." Combinations with high fitness are selected from the historical trait gene pool as the fitness function of the "parent generation." The formula used to measure the suitability of feature combinations to the current work objective is as follows: ,in, , , , Weights are assigned based on work objectives. The selection rule uses a roulette wheel selection method, with fitness as a criterion. The higher the feature combination, the higher the probability of being selected. The larger, Where m is the number of candidate combinations. For the first The fitness of each combination is assessed. After the above processing, crossover is performed to achieve cross-scenario specific combinations by crossover of gene fragments to generate offspring combinations. The crossover operator randomly selects the crossover point and exchanges gene fragments from the parent generation. Parent Generation 1 Smart Factory , , , Parent 2 autonomous driving , 6, 8, Crossover offspring , , 8, This example combines the high reliability of a factory with the high real-time performance / efficiency of an autonomous driver.

[0056] After the above cross-processing, the mutation process begins, introducing random diversity. To avoid feature combinations getting trapped in local optima, the values ​​of some gene factors are randomly adjusted to simulate gene invariance. The mutation probability is... The value is usually set to 0.01-0.1. The mutation method is to randomly add or subtract a small value from the selected gene factor to ensure that it is still within the range after mutation. , The perturbation is random. The new feature combination needs to match the adjusted working objective. The adjustment of the working objective is based on the objective reconstruction of the new feature combination, and the objective is quantified into actionable metrics. The calibration logic extracts key metrics from the recombined feature combination according to the requirements of the new scenario and sets the working objective threshold. The real-time threshold is... Throughput target is Feedback to the inner loop is an ant colony algorithm that synchronizes the weights of the work objectives to the inner loop, adjusting the pheromone calculation values. .

[0057] Specifically, such as Figure 1 As shown, step S500, the iteration effect mirror evaluation mechanism, constructs a virtual twin scenario for each iteration to simulate the optimization effect under extreme conditions. To ensure the effectiveness of iterative optimization, a virtual scenario is constructed using digital twins to simulate extreme conditions and verify the optimization effect, avoiding the risk of trial and error directly in the real environment. Specifically, this is achieved through a closed loop of virtual twin scenario modeling, extreme condition simulation, virtual-real data comparison, and strategy correction, which allows for early detection of defects in the optimization scheme and dynamic adjustment of the iteration direction. Virtual twin scenario construction uses digital twin technology to create a virtual model consistent with the actual edge environment, including node topology, task flow, and network conditions. Extreme condition simulation simulates high loads without affecting actual operation. Network latency, the formula for increasing latency by 100ms is: Or, a security attack such as a DDoS attack formula is: Virtual-real scene comparison correction evaluates optimization effectiveness by comparing execution data, task latency, and error rates between virtual and real-world scenarios. If optimization is ineffective in the virtual scene and latency is not reduced, the iteration direction is corrected, adjusting the weight coefficients of the ant colony algorithm or the feature reorganization strategy. For example, iteration parameters can be adjusted using reinforcement learning algorithms. Key metrics between the virtual and real-world scenarios are compared, and the deviation is calculated. If the deviation is significant or the optimization effect is unsatisfactory, the iteration strategy is adjusted.

[0058] In this invention, high load is not defined within a quantified range, which aligns with the diverse characteristics of edge computing scenarios. Different edge nodes exhibit significant differences in hardware configuration and task types, making a unified quantification standard insufficient for all scenarios. From a technical perspective, high load is a relative threshold concept, characterized by the utilization rate of core resources of an edge node or node cluster approaching or exceeding its rated capacity, leading to increased task execution latency, decreased response speed, and even the risk of data loss and service interruption—a state consistent with the critical state of "extreme working conditions." The specific range needs to be dynamically calibrated based on node hardware specifications, task type, and service quality requirements. This design echoes the technical means of simulating extreme working conditions using virtual twin scenarios, allowing for further optimization of load threshold settings through virtual-real comparison, ensuring the flexibility of the technical solution in adapting to various scenarios.

[0059] Based on the processing described above, the next step is to calculate the evaluation indicators and deviation rate. The core evaluation indicators include delay. Error rate Throughput The formula for the deviation rate is: ,like If the optimization effect in the virtual scene is satisfactory, then the virtual scenario is effective; if the optimization effect in the virtual scenario does not meet the target, then policy correction is triggered. Policy correction is based on parameter adjustment using reinforcement learning. Key parameters are adjusted using reinforcement learning, such as... and genetic algorithms The goal is to minimize the bias in metrics within the virtual environment.

[0060] Continuous iterative optimization, repeating the above steps, will achieve continuous iterative optimization of the edge computing environment. Each iteration cycle can be set to hours or days, adjusted according to business needs. The iteration cycle setting and dynamic adjustment require an initial cycle set based on business sensitivity; the cycle adjustment formula is as follows: ,in For the initial period, To adjust the coefficient, This represents the fitness improvement amount for this iteration. The iteration can be paused when the iteration termination conditions are met. These conditions are: the achievement rate of the work objective exceeds 100% in three consecutive iterations, the deviation between the virtual and actual scenarios does not exceed 5%, and the core indicators remain stable within the target threshold.

[0061] Through the implementation of this embodiment, the edge computing environment can adapt to dynamic changes, improving task processing efficiency, security, and reliability. Experimental data shows that in smart factory scenarios, the average task response time decreases, node resource utilization increases, but security incidents also increase.

[0062] The difference between this invention and traditional models lies in the dual-loop feedback model, which combines strategic and tactical levels, utilizes transfer learning and gene recombination, and employs virtual twin evaluation. This invention uses an inner-loop ant colony heuristic weight adjustment mechanism to convert performance data into pheromone concentration, enabling nodes to autonomously optimize collaborative weights, and employing a pheromone evaporation mechanism. While traditional ant colony algorithms may be used for path optimization or task scheduling, they may not be directly applied to edge node collaborative weight adjustment, especially when combined with security, reliability, and efficiency data. The pheromone evaporation mechanism is a standard part of ant colony algorithms, but in this invention, it is used for edge computing optimization. In this invention, the outer-loop feature gene recombination strategy decomposes task features into basic units such as safety factors, reliability factors, and efficiency factors, generating new feature combinations through gene recombination to adapt to new tasks. Compared to traditional methods that use feature engineering or transfer learning, this invention incorporates a biological gene recombination approach, similar to the crossover operation in genetic algorithms, applied to task feature transfer learning.

[0063] This invention employs an iterative effect mirror evaluation mechanism, constructing a virtual twin scenario to simulate extreme working conditions and correcting the iteration direction through virtual-real comparison. While traditional digital twin technology is used in the Industrial Internet of Things (IIoT), it is applied here for edge computing optimization, particularly for evaluating iterative effects. Feedback analysis steps: Tactical or strategic issues are determined based on quantitative data. Unlike traditional feedback control, which may be based on thresholds or simple rules, this invention explicitly distinguishes between tactical and strategic issues and uses quantitative data, employing a more refined processing method. The outer loop in this invention sets work objectives based on cross-scenario task feature transfer learning. While traditional transfer learning is common in machine learning, its application to edge computing work objective setting is less common. The inner loop in this invention executes work tasks using a distributed autonomous approach. Edge nodes synchronize pheromone concentration and weight states through a consensus protocol, eliminating the need for centralized control. Compared to traditional edge computing, which may use centralized or partially distributed control, this paper uses a consensus protocol to achieve fully distributed autonomy, similar to blockchain or distributed consensus, and applies it to pheromone synchronization, which differs from traditional processing methods.

[0064] The system modules and models used in this invention include: an outer loop module, an inner loop module, a feedback analysis module, and an iterative evaluation module. Traditional systems lack this module division and do not integrate dual-loop feedback and model libraries. This invention's dual-loop feedback iterative model distinguishes between strategic and tactical levels, optimizing them separately. Cross-scenario task feature transfer learning is used for setting work objectives, reducing repetitive learning. The inner loop uses ant colony heuristic weight adjustment, based on pheromone concentration and evaporation mechanisms, to achieve distributed self-optimization. The outer loop feature gene recombination strategy decomposes task features into factors, adapting them to new scenarios through recombination. An iterative effect mirror evaluation mechanism uses virtual twins to simulate extreme conditions and correct the iteration direction. Feedback analysis, based on quantitative data, clearly distinguishes between tactical and strategic issues. Distributed autonomy, through a consensus protocol, synchronizes pheromones between nodes, eliminating the need for centralized control.

[0065] The differences between this invention and traditional methods are as follows: Traditional edge computing optimization methods only focus on resource allocation or task scheduling, lacking this hierarchical feedback iteration; traditional ant colony algorithms are only used for task scheduling, not for node collaborative weight adjustment; traditional transfer learning is only used for model training, not for setting work objectives; and traditional digital twins are only used for monitoring or simulation, not for iterative evaluation.

[0066] The innovative points and significant differences in this invention include the following specific details: Dual-loop structure: This solution introduces a hierarchical iteration of an outer loop (strategy) and an inner loop (tactics), while traditional solutions are mostly single-layer optimizations and cannot distinguish between strategic and tactical issues.

[0067] Cross-scenario transfer learning: The outer loop of this solution sets work goals based on transfer learning, while traditional solutions usually design goals for specific scenarios and lack cross-scenario knowledge transfer.

[0068] Feedback analysis mechanism: This solution clearly determines the problem type (tactical or strategic) based on the execution data, while traditional solutions usually only perform overall deviation correction without classification judgment.

[0069] Mirror evaluation: This solution uses virtual twins to simulate extreme working conditions, while traditional solutions use simulation testing, which is not integrated into iterative feedback for correcting direction.

[0070] Multidimensional indicator conversion: The pheromone concentration in this scheme is based on comprehensive data such as safety, reliability, and efficiency, while traditional schemes usually only consider a single indicator.

[0071] Distributed self-optimization: This scheme achieves autonomous weight adjustment among nodes through the pheromone evaporation mechanism, without the need for central control. Traditional schemes often rely on a central controller for weight updates.

[0072] Collaboration weight optimization: This solution directly optimizes the collaboration weights between nodes, while traditional solutions focus more on task allocation or path selection. Factor-level decomposition: This approach decomposes task features into basic factors (safety, reliability, efficiency). Traditional approaches typically transfer the overall feature vector, lacking fine-grained reorganization.

[0073] Gene recombination mechanism: This scheme uses biological gene recombination to generate new feature combinations, enhancing cross-scenario adaptability. Traditional schemes often use linear combination or simple weighting.

[0074] Dynamic adaptation: This solution automatically retrieves historical gene pools to feed back into the internal circulation, whereas traditional solutions often require manual adjustment or retraining for transfer learning.

[0075] Integrated Iterative Evaluation: This solution embeds the virtual twin directly into the iterative process for forward-looking evaluation, while traditional solutions are mostly used for post-analysis or real-time monitoring.

[0076] Extreme operating condition simulation: This solution is specifically designed to simulate extreme conditions such as high load and security attacks, while traditional solutions often only cover normal operating conditions.

[0077] Virtual-Real Comparison Correction: This scheme dynamically adjusts the iteration direction by comparing virtual and real data, a continuous correction mechanism lacked in traditional schemes.

[0078] Problem classification and judgment: This solution clearly distinguishes between tactical and strategic problems, while traditional solutions usually handle all deviations uniformly.

[0079] Multi-dimensional quantitative data: This solution uses business-level metrics such as feature usage rate and user operation time, while traditional solutions focus more on technical metrics.

[0080] Intelligent decision-making: This solution automatically selects the iteration path based on data, while traditional solutions require manual intervention or the use of fixed rules.

[0081] Strategic-level transfer: This solution applies transfer learning to the definition of work objectives, while traditional solutions are usually used for model or task-level adaptation.

[0082] Feature gene library: This scheme establishes a historical feature gene library to support rapid adaptation, a knowledge accumulation mechanism that traditional schemes lack.

[0083] Consensus Protocol Synchronization: This scheme uses a consensus protocol to synchronize pheromone and weight states, achieving true distributed autonomy. Traditional schemes typically rely on central node coordination.

[0084] No centralized control: This solution is completely decentralized, while traditional solutions often retain some centralized control functions.

[0085] Module architecture: The module design of this solution is tightly coupled with the dual-loop model, supporting hierarchical iteration, while the modules of traditional solutions are relatively linear.

[0086] Integrated evaluation module: The iterative evaluation module of this solution includes a virtual twin, while traditional solutions do not have a dedicated evaluation module.

[0087] This solution introduces a dual-loop feedback iterative model into edge computing optimization, combining cross-scenario transfer learning, ant colony-inspired distributed self-optimization, feature gene recombination, and virtual twin mirror evaluation to achieve separation of strategic and tactical optimization and continuous adaptation. Compared with traditional solutions, the main innovations are reflected in: architectural innovation, with the dual-loop model providing hierarchical optimization and avoiding the limitations of traditional single-loop models; algorithmic innovation, combining ant colony algorithms with multi-dimensional indicators, and gene recombination strategies to enhance cross-scenario capabilities; evaluation innovation, with virtual twins integrating the iterative process to improve system robustness and foresight; and autonomous innovation, achieving fully distributed control through a consensus protocol to reduce the bottleneck of central nodes. These innovations enable your solution to have better adaptability, efficiency, and reliability in dynamic and complex edge environments, significantly differentiating it from traditional methods.

[0088] Example 2: An edge computing optimization system based on dual-loop feedback iteration, such as Figure 2 As shown, it includes a target decomposition module 100, a data collection module 200, an inner loop module 300, an outer loop module 400, and an iterative evaluation module 500; The target decomposition module 100 sets the work target through an outer loop, decomposes the work target into multiple work tasks, and assigns them to the nodes corresponding to the edge node group. The data collection module 200 executes the work task through an internal loop and collects the execution data of the edge node group in real time. The execution data includes security performance, reliability indicators and efficiency data. The inner loop module 300 performs feedback analysis on the execution data, obtains and judges the execution deviation. If the execution deviation is caused by a tactical problem, it enters the inner loop iteration. The iteration result triggers the inner loop ant colony heuristic weight adjustment to optimize the collaborative weight between pairs of edge nodes. If the execution deviation of the outer loop module 400 originates from a strategic problem, it will enter the outer loop iteration, adopt the outer loop reorganization strategy to calibrate the work objectives and re-decompose the work tasks. The iterative evaluation module 500 constructs a virtual twin scenario to simulate the optimization effect under extreme working conditions, corrects the cyclic iteration direction through virtual-real comparison, and realizes continuous iterative optimization of the edge computing environment through the cyclic iteration direction.

[0089] The various embodiments in this specification are described in a progressive manner. Each embodiment is an important part of the invention. The above description is only a preferred embodiment of the invention and does not limit the patent scope of the invention. Any equivalent structural or procedural transformations made based on the content of this specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this invention.

[0090] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0091] The descriptions focus on the differences from other embodiments. For the same or similar parts between the various embodiments, please refer to each other.

[0092] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0093] This invention is described with reference to flowchart illustrations and / or block diagrams of the method, terminal device (system), and computer program product according to the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0094] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0095] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] The above description of the embodiments is provided to enable those skilled in the art to understand and apply the present invention. It will be apparent to those skilled in the art that various modifications can be made to the above embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. An edge computing optimization method based on dual-loop feedback iteration, characterized in that, Includes the following steps: The work objectives are set through the outer loop, and then broken down into multiple work tasks and assigned to the corresponding nodes of the edge node group. The task is executed through an inner loop, and execution data of the edge node group is collected in real time. The execution data includes security performance, reliability indicators and efficiency data. The execution data is analyzed to obtain and judge the execution deviation. If the execution deviation is due to tactical problems, the inner loop iteration is entered. The iteration results trigger the inner loop ant colony heuristic weight adjustment to optimize the collaborative weight between pairs of edge nodes. If the deviation in execution stems from strategic issues, then an external loop iteration is initiated, employing an external loop restructuring strategy to calibrate work objectives and re-decompose work tasks. A virtual twin scenario is constructed to simulate the optimization effect under extreme working conditions. The direction of the iterative cycle is corrected by comparing the virtual and real environments, and the continuous iterative optimization of the edge computing environment is achieved through the direction of the iterative cycle.

2. The edge computing optimization method based on dual-loop feedback iteration according to claim 1, characterized in that, The inner-loop ant colony heuristic weight adjustment mechanism includes the following steps: Based on the security performance, reliability indicators, and efficiency data of edge nodes, the corresponding pheromone concentration is obtained. By autonomously optimizing the collaborative weights between pairs of edge nodes through pheromone concentration, if an edge node exhibits better performance in a specific task, the pheromone concentration is increased, resulting in task-appropriate bias. For edge nodes with high risk fluctuation frequency, the weight ratio is automatically reduced through the pheromone evaporation mechanism to achieve distributed self-optimization.

3. The edge computing optimization method based on dual-loop feedback iteration according to claim 1, characterized in that, The external circulation restructuring strategy includes the following steps: The task characteristics in different scenarios are broken down into basic units, wherein the basic units include safety factors, reliability factors and efficiency factors; New feature combinations are generated based on basic units and through cross-fusion, wherein the cross-fusion method includes biological gene recombination; A historical feature gene library is constructed. When new task types or scenario changes occur in the edge computing environment, the historical feature gene library is retrieved through new feature combinations to obtain feature models that adapt to the new requirements, and then fed back to the weight adjustment mechanism of the inner loop.

4. The edge computing optimization method based on dual-loop feedback iteration according to claim 1, characterized in that, The construction of a virtual twin scenario to simulate optimization effects under extreme working conditions, and the correction of the direction of iterative iterations through virtual-real comparison, includes the following steps: A virtual twin scenario is constructed for each iteration, and extreme working conditions are simulated in the virtual twin scenario, including high load, network latency, and security attacks; The optimization effect is evaluated by comparing the simulation results with the real data, and the iteration direction is continuously corrected to ensure that the collaborative weighting maintains forward-looking adaptability to complex edge environments. The collaborative weighting is dynamically weighted by the execution data of the edge nodes, and the optimal performance solution is achieved through optimization.

5. The edge computing optimization method based on dual-loop feedback iteration according to claim 1, characterized in that, The process of performing feedback analysis based on execution data to obtain and determine execution deviations includes the following steps: Judgment is made based on execution deviation. If the tactical execution does not meet expectations but the work objectives are still reasonable, it is judged as a tactical problem and enters the inner loop iteration. If tactical execution is satisfactory but work objectives are not achieved, it is judged to be a strategic problem and enters the external loop iteration; The execution deviations include feature usage rate, user operation time, and number of negative feedback reviews.

6. The edge computing optimization method based on dual-loop feedback iteration according to claim 1, characterized in that, The work objective is set through an outer loop based on cross-scenario task feature transfer learning, including extracting task feature genes from historical scenarios and adapting them to new scenarios through transfer learning, thereby reducing the cost of repeated learning.

7. The edge computing optimization method based on dual-loop feedback iteration according to claim 1, characterized in that, The inner loop executes tasks in a distributed autonomous manner, obtains weight status through collaborative weighting, and synchronizes pheromone concentration and weight status among edge nodes through a consensus protocol.

8. An edge computing optimization system based on dual-loop feedback iteration, characterized in that, It includes a target decomposition module, a data collection module, an inner loop module, an outer loop module, and an iterative evaluation module; The target decomposition module sets the work target through an outer loop, decomposes the work target into multiple work tasks, and assigns them to the corresponding nodes of the edge node group. The data collection module executes the work task through an internal loop and collects the execution data of the edge node group in real time. The execution data includes security performance, reliability indicators and efficiency data. The inner loop module analyzes the execution data to obtain and judge the execution deviation. If the execution deviation is caused by tactical problems, it enters the inner loop iteration. The iteration result triggers the inner loop ant colony heuristic weight adjustment to optimize the collaborative weight between pairs of edge nodes. If the deviation in the execution of the outer loop module originates from a strategic problem, it will enter the outer loop iteration, adopt the outer loop reorganization strategy to calibrate the work objectives and re-decompose the work tasks. The iterative evaluation module constructs a virtual twin scenario to simulate the optimization effect under extreme working conditions, corrects the direction of iterative iteration through virtual-real comparison, and achieves continuous iterative optimization of the edge computing environment through the direction of iterative iteration.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

10. An edge computing optimization device based on dual-loop feedback iteration, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

Images