An edge-computing-based high-concurrency controller real-time scheduling method

Abstract

The application discloses a kind of high concurrency controller real-time scheduling methods based on edge computing, comprising the following steps: S1, in target area deployment master-slave edge node architecture;S2, execute task flow control and preprocessing operation;S3, construct priority scheduling strategy S4, issue scheduling result and collect control feedback information, record to local scheduling log;S5, construct global scheduling collaborative mechanism based on node state vector;S6, construct scheduling coding individual and define multi-objective optimization function group, combine non-dominated sorting, crowdedness elite guide and greedy improvement differential evolution algorithm for scheduling optimization, generate optimal scheduling scheme and issue to regional edge node execution.The application has the advantages of low delay, high reliability, strong scheduling balance, suitable for intelligent building, wisdom factory and other edge computing scenarios with high requirements for control task real-time and concurrent processing capability.

Description

Technical Field

[0001] This invention relates to the field of edge computing and intelligent control scheduling technology, and in particular to a real-time scheduling method for high-concurrency controllers based on edge computing. Background Technology

[0002] In IoT scenarios such as smart agriculture, smart buildings, and distributed energy, controllers, as crucial carriers of sensing and execution, typically require high concurrency, diversity, and real-time performance. To meet the demands for rapid task response and efficient resource utilization, the industry has gradually introduced edge computing architectures. By deploying edge nodes with computing capabilities close to the controller, localized processing and intelligent scheduling of control tasks can be achieved. Traditional centralized scheduling methods usually rely on cloud center coordination, facing problems such as high network latency, slow task response, and unstable scheduling results, making it difficult to meet the low latency and high reliability requirements for task execution in complex environments. In contrast, edge computing architectures offer shorter response paths and stronger system autonomy, becoming a key direction for solving the real-time scheduling bottleneck of controllers.

[0003] While existing edge computing-based control task scheduling methods have made some progress in distributed deployment and local task response, they still suffer from shortcomings when facing large-scale concurrent controller requests. These include imprecise scheduling strategies, low resource utilization, and a lack of adaptability in the scheduling mechanism. Some scheduling methods employ static priority or simple round-robin strategies, failing to dynamically adjust the scheduling order based on task urgency, controller operating status, and node load, easily leading to resource bottlenecks and response delays. Secondly, most methods lack precise evaluation mechanisms for handling abnormal nodes and task migration, and migration decisions often lack data support, affecting the overall stability and effectiveness of scheduling. Furthermore, current scheduling optimization algorithms generally focus on single-objective performance, ignoring the trade-offs between latency, load balancing, and migration overhead that are common in control task scheduling, making it difficult to obtain robust and efficient scheduling schemes under multi-dimensional optimization objectives.

[0004] Against this backdrop, there is an urgent need to propose a scheduling method for high-concurrency control scenarios that possesses real-time performance, adaptability, and global optimization capabilities. This method should integrate edge computing architecture, utilize intelligent algorithms to construct a mapping relationship between tasks and resources, dynamically optimize scheduling strategies, and form a feedback loop mechanism for task execution, thereby comprehensively improving the scheduling efficiency and system stability of the control system. Summary of the Invention

[0005] One objective of this invention is to propose a real-time scheduling method for high-concurrency controllers based on edge computing. This invention fully integrates edge node architecture, multi-objective differential evolutionary optimization algorithm, and scheduling feedback closed-loop mechanism, and details the entire process of achieving real-time scheduling and adaptive optimization of control tasks in a high-concurrency environment. By constructing a scheduling urgency scoring function, an autonomous task flow control preprocessing mechanism, a node state vector collaborative strategy, and a greedy-guided multi-objective evolutionary scheduling algorithm, it effectively solves the bottleneck problems of traditional methods in terms of sudden high concurrency of tasks, uneven node load, and scheduling response lag. It possesses significant advantages such as strong real-time scheduling performance, high algorithm adaptability, and excellent system robustness.

[0006] A real-time scheduling method for a high-concurrency controller based on edge computing according to an embodiment of the present invention includes the following steps: S1. Deploy a master-slave edge node architecture in the target area. The regional edge nodes connect to and manage the high-concurrency controller of the environment and collect control task requests. S2, Perform task flow control and preprocessing operations; S3. Construct a priority scheduling strategy based on the scheduling urgency scoring function, and perform scheduling for regional edge nodes according to the priority of the scoring value; S4. Issue scheduling results and collect control feedback information, and record them to the local scheduling log; S5. Construct a global scheduling and coordination mechanism based on node state vectors. When abnormal load and scheduling failure are detected at the edge nodes of the region, execute task migration across nodes. S6. The main edge node constructs a scheduling code individual based on the aggregated task unit and the set of regional edge nodes in the current scheduling cycle and defines a multi-objective optimization function group. It combines non-dominated sorting, crowding elite guidance and greedy improved differential evolution algorithm to optimize the scheduling, generate the optimal scheduling scheme and send it to the regional edge node for execution.

[0007] Optionally, the node state vector includes computational resource utilization, task queue length, control task success rate, and migration cost.

[0008] Optionally, S1 specifically includes: S11. Deploy an edge scheduling architecture in the target area, including a main edge node and multiple regional edge nodes. The main edge node has global task coordination and scheduling optimization functions, and the regional edge nodes have access management and scheduling task processing capabilities for high-concurrency environment controllers. S12. The regional edge node completes the registration and access of each environmental high-concurrency controller through wireless communication, and establishes the access mapping relationship between the high-concurrency controller and the accessed regional edge node; S13. Each environment's high-concurrency controller periodically reports control task requests to the edge node of its region when a trigger event occurs. The control task request includes the task type, priority label, deadline, control parameter vector, and the current operating load information of the high-concurrency controller. S14. The region edge node performs data structure standardization processing on the received control task requests, extracts task identifiers, task features, and resource status, and constructs a structured control task request set. .

[0009] Optionally, S2 specifically includes: S21. The regional edge nodes apply a token bucket mechanism to control the flow of control task request sets, setting the bucket capacity to [value missing]. The token generation rate is When the number of requests exceeds the number of available tokens within a unit of time, the excess tasks are cached to smooth out the high-concurrency request flow. S22. Received set of control task requests Perform data deduplication to retain the latest valid task request within the same time window for the same high-concurrency controller; S23, Set of Control Task Requests Control parameter vector in Perform parameter normalization processing to make All elements Mapping to interval This is used to eliminate the influence of different units on subsequent scoring functions; S24, Based on task type Based on the principle of time proximity, control task requests that meet the merging conditions are aggregated to form aggregated task units. .

[0010] Optionally, S3 specifically includes: S31, the region edge node aggregation task unit Each control task in the process requests the calculation of a scheduling urgency score function value. ; S32, For aggregation task units All task requests are processed according to the scoring function value. Sort the tasks in descending order to obtain the sorted list of scheduling tasks. ; S33, the region edge nodes, according to the sorted scheduling task list... The control tasks are executed sequentially in the order of sorting, and the scheduling priority is determined by the score function value. The higher the score, the higher the scheduling priority.

[0011] Optionally, S4 specifically includes: S41. Regional edge nodes are sorted according to the scheduling task list. The system will match the control task requests with the highest priority positions to the corresponding high-concurrency controllers in sequence, and issue the scheduling scheme to the corresponding high-concurrency controllers in the form of control commands. S42. After receiving the scheduling scheme, each high-concurrency controller in the environment executes the corresponding control task and sends back control feedback information in real time. The control feedback information includes a response status code. Actual execution delay Successful execution flag and abnormal event identifiers ; S43. The regional edge node will link the control feedback information with the corresponding control task request identifier. Perform binding and construct control feedback log entries, then write all control feedback log entries to the local scheduling log. middle.

[0012] Optionally, S5 specifically includes: S51. The main edge node periodically collects the operating status of all regional edge nodes and constructs a data structure for each regional edge node. Node state vector ,in This indicates the utilization rate of computing resources. Indicates the length of the task queue. Indicates the task success rate. Indicates the cost of task migration; S52, Task Success Rate From the edge nodes of the region This is calculated based on the task execution results recorded in the local scheduling log; S53, The main edge node is based on the node state vector Determine if any abnormal nodes exist. If a node exists and meets any of the following conditions, it is considered an abnormal node: Calculate resource utilization. Exceeding the threshold Task success rate Below the threshold The number of scheduling failures exceeded the threshold. ; S54. The primary edge node, combined with node idle time, success rate, and migration cost, calculates the node scheduling adaptability score. ; S55, if node If a node is identified as an abnormal node, the primary edge node will select a node from the remaining nodes based on its node scheduling adaptability score. Largest node As the migration target, some unscheduled tasks will be migrated to the node. .

[0013] Optionally, S6 specifically includes: S61, The main edge node, based on the aggregated task units within the current scheduling period... and the set of edge nodes in the region Construct scheduling coding individual ,in Indicates control task Assigned to a node And introduce a task-node adaptation matrix. As a constraint, ensure ; S62, For individual scheduling codes Construct a set of multi-objective optimization functions, including: delay synthesis objective Balanced task distribution Total migration cost target The latency comprehensive objective measures the response latency of control tasks during communication, queuing, and execution. The task distribution balance objective reflects the variance of task load at edge nodes in each region. The total migration cost objective evaluates the bandwidth, network, and resource reconfiguration overhead of tasks in cross-node scheduling. S63. Generate the population using a uniform initialization and minimum load priority strategy. For each individual scheduling code Computational delay integrated objective Balanced task distribution Total migration cost target Then perform a non-dominated sort to obtain the front set. And calculate the crowding distance for elite selection; S64. The primary edge node introduces a greedy guidance mechanism based on the current non-dominated front set in the mutation operation of the differential evolution algorithm; for each scheduling coded individual in the current population... From the non-dominated frontier set Select the most crowded elite individuals And construct the guiding mutation vector Controlling elite individuals The strength of the impact on the current mutation search direction; The item is a leading item, utilizing elite individuals. Approaching the current scheduling code individual Direction, guiding the population Converge towards the better-performing scheduling scheme in the leading solution set; S65, For each scheduling code individual and the corresponding test individuals The main edge nodes calculate the objective function values ​​based on a set of multi-objective optimization functions and execute the selection strategy accordingly. S66, The main edge node sets the maximum number of generations for differential evolution. Once the convergence condition is met, the current non-dominated frontier set... Select the individual with the largest congestion distance and the optimal overall objective value for scheduling. This will serve as the final scheduling scheme.

[0014] S67. The main edge node will schedule the optimal coding individual. The corresponding scheduling scheme is decoded to extract each control task request. Mapped target region edge nodes The scheduling instructions are then sent out one by one to the corresponding regional edge nodes; after receiving the scheduling plan, each regional edge node is responsible for distributing the control tasks to its respective environment high-concurrency controller. It drives the control task to execute corresponding control behaviors, records the high-concurrency controller operation feedback information, and constructs feedback data tuples; Optionally, the aforementioned high-concurrency controller real-time scheduling method based on edge computing is characterized in that the selection strategy: if the guiding mutation vector Non-inferior to scheduling code individuals across all objectives Then retain the guiding mutation vector. If the guiding mutation vector With scheduling coding individual If they are not mutually dominant, then compare their non-dominant rank and crowding distance, selecting the individual with the better rank and greater distance to be retained for the next generation; if the mutation vector is guided... If controlled, the scheduling code individual is retained. .

[0015] The beneficial effects of this invention are: This invention constructs a master-slave edge node architecture, deploying multiple regional edge nodes with computing capabilities near the environment controller. This enables local access and immediate response to control task requests, significantly reducing the scheduling and communication latency issues inherent in traditional centralized architectures. The regional edge nodes, combined with a token bucket mechanism and task preprocessing, can effectively perform flow control and load balancing under high concurrency, improving the system's processing capacity and stability under sudden task pressure.

[0016] This invention introduces a scheduling urgency scoring function, which comprehensively considers task deadlines, priority tags, and the current operating load of the controller, achieving fine-grained sorting and dynamic priority response for task scheduling. This mechanism enables edge nodes to prioritize high-urgency critical control tasks, improving the overall real-time scheduling performance and critical task assurance capabilities of the system.

[0017] The main edge node achieves intelligent optimization of the scheduling scheme by constructing a multi-objective optimization function set based on latency, load balancing, and migration cost, combined with a differential evolution algorithm guided by non-dominated sorting and congestion. An innovative greedy guidance mechanism is introduced to guide the mutation direction towards the current elite solution, improving the algorithm's search efficiency and solution quality in complex scheduling scenarios. Regarding the abnormal node detection and scheduling coordination mechanism, this invention utilizes node state vectors to achieve anomaly perception and task migration, ensuring the continuous and stable operation of the scheduling process.

[0018] This invention effectively solves the problems of response lag, uneven resource allocation, and weak scheduling optimization capability of existing edge scheduling methods when handling high-concurrency control tasks. It significantly improves the real-time performance, robustness, and scheduling efficiency of the control system and has good engineering application prospects and promotion value. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a high-concurrency controller real-time scheduling method based on edge computing proposed in this invention; Figure 2 This is a schematic diagram of a scheduling optimization structure based on a differential evolution algorithm with a multi-objective optimization function set proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figure 1-2 A real-time scheduling method for high-concurrency controllers based on edge computing includes the following steps: S1. Deploy a master-slave edge node architecture in the target area. The regional edge nodes connect to and manage the high-concurrency controller of the environment and collect control task requests. S2, Perform task flow control and preprocessing operations; S3. Construct a priority scheduling strategy based on the scheduling urgency scoring function, and perform scheduling for regional edge nodes according to the priority of the scoring value; S4. Issue scheduling results and collect control feedback information, and record them to the local scheduling log; S5. Construct a global scheduling and coordination mechanism based on node state vectors. When abnormal load and scheduling failure are detected at the edge nodes of the region, execute task migration across nodes. S6. The main edge node constructs a scheduling code individual based on the aggregated task unit and the set of regional edge nodes in the current scheduling cycle and defines a multi-objective optimization function group. It combines non-dominated sorting, crowding elite guidance and greedy improved differential evolution algorithm to optimize the scheduling, generate the optimal scheduling scheme and send it to the regional edge node for execution.

[0022] This invention proposes a real-time scheduling method for high-concurrency controllers based on edge computing. By constructing a master-slave edge node architecture, it achieves localized access to the high-concurrency controller in the environment and the collection of task requests. It combines a token bucket mechanism and a task preprocessing process for flow control management and implements priority scheduling of tasks based on a scheduling urgency scoring function. Furthermore, it introduces node state vectors to construct a global scheduling coordination mechanism to complete task migration when there is an abnormal load or scheduling failure. The master edge node constructs individual scheduling codes and defines a multi-objective optimization function group, integrating non-dominated sorting, congestion-based elite guidance, and a greedy improved differential evolution algorithm to achieve intelligent optimization and real-time distribution of scheduling schemes, thereby improving the real-time performance of scheduling, system robustness, and resource utilization efficiency in high-concurrency control scenarios.

[0023] In this embodiment, the node state vector includes computational resource utilization, task queue length, control task success rate, and migration cost.

[0024] In this embodiment, S1 specifically includes: S11. Deploy an edge scheduling architecture in the target area, including a main edge node and multiple regional edge nodes. The main edge node has global task coordination and scheduling optimization functions, and the regional edge nodes have access management and scheduling task processing capabilities for the environment controller. S12. The regional edge node registers and accesses multiple environmental controllers through wired or wireless communication, and establishes a mapping relationship between the environmental controller identifier and the regional edge node to which it belongs, which is used for subsequent task allocation and scheduling path identification. S13. Each environmental controller periodically or upon the occurrence of a triggered event reports a control task request to the edge node of its respective region. The control task request includes a task type, priority label, deadline, control parameter vector, and the current operating load information of the controller, wherein the priority label is represented as a positive integer. The deadline is a real number. The control parameter vector is represented as The controller load is represented as a normalized real number. ; S14. The region edge node performs data structure standardization processing on the received control task requests, extracts task identifiers, task features, and resource status, and constructs a structured control task request set. This indicates the number of control task requests. Number the controller. For task type, For priority tags, The deadline is [date / time]. For the control parameter vector, This is the controller load information.

[0025] In this invention, an edge scheduling architecture consisting of a main edge node and multiple regional edge nodes is deployed within the target area. The main edge node is responsible for global task coordination and optimization decisions, while the regional edge nodes are responsible for the access and task processing of the environment controllers. The regional edge nodes register and access the environment controllers via wired or wireless means, and establish a mapping relationship between the controller identifier and its respective edge node, providing a path basis for subsequent scheduling. Each environment controller periodically or in an event-driven manner reports control task requests containing task type, priority label, deadline, control parameter vector, and controller load to its respective regional edge node. The regional edge node performs standardized processing, extracts task features and resource status information, and constructs a structured set of control task requests to support subsequent scheduling scoring and optimization processes.

[0026] In this embodiment, S2 specifically includes: S21. The regional edge nodes apply a token bucket mechanism to control the flow of control task request sets, setting the bucket capacity to [value missing]. The token generation rate is When the number of requests exceeds the number of available tokens within a unit of time, the excess tasks are cached or discarded to smooth out the high-concurrency request flow. S22. Perform data deduplication on the received set of control task requests, and retain the latest valid task request of the same controller within the same time window. S23. Perform parameter normalization on the control parameter vector in the set, so that all elements in the set are mapped to the interval. This is used to eliminate the influence of different units on subsequent scoring functions; S24. Based on the principle of task type and time proximity, control task requests that meet the merging conditions are aggregated to form aggregated task units. This serves as the input set for subsequent scheduling, scoring, and sorting.

[0027] In this invention, after receiving a set of control task requests, the regional edge node uses a token bucket mechanism to limit the flow of requests, smoothing out high-concurrency requests by setting the bucket capacity and token generation rate. Subsequently, it performs data deduplication on the task request set, retaining the latest valid requests from the same controller within the same time window. Next, it normalizes the control parameter vectors, mapping elements to intervals to unify dimensions, improving the accuracy and stability of the subsequent scoring function. Finally, based on the principle of task type and time proximity, it aggregates task requests with similar characteristics to generate aggregated task units, which serve as the input for scheduling priority scoring and ranking, enhancing the system's adaptability to high-concurrency scheduling tasks and improving computational efficiency.

[0028] In this embodiment, S3 specifically includes: S31. The regional edge node calculates the scheduling urgency score function value for each control task request in the aggregated task unit; S32, The scheduling urgency scoring function is defined as follows: ; This indicates the deadline for controlling task requests. This is the current scheduling timestamp. For priority tags, The current operating load of the controller. , , These are the adjustable weighting coefficients for the scoring function; S33. Sort all task requests in the aggregated task unit in descending order according to the scoring function value to obtain the sorted scheduling task list. ; S34. The edge nodes of the region execute control task scheduling sequentially according to the sorting order of the list. The scheduling priority is determined by the score function value. The higher the score value, the higher the scheduling priority.

[0029] In this invention, for each control task request in the aggregated task unit, the regional edge node calculates its scheduling urgency score function value. This function comprehensively considers the task deadline, the current scheduling timestamp, the priority label, and the current operating load of the controller, and introduces an adjustable weight coefficient for weighting, thereby quantifying the scheduling urgency of the task. The node sorts all task requests in descending order according to the score value, generates a sorted scheduling task list, and executes scheduling operations sequentially according to the order of the list, realizing an efficient task priority scheduling mechanism driven by urgency, effectively improving the timeliness of task response and the rationality of resource allocation.

[0030] In this embodiment, S4 specifically includes: S41, The region edge nodes are sorted according to the scheduling task list. The system sequentially matches the control task requests with the highest sorting positions to their corresponding controllers and sends the scheduling scheme to the controllers in the form of control commands. S42. After receiving the scheduling plan, each environmental controller executes the corresponding control task and sends back control feedback information in real time. The control feedback information includes a response status code. Actual execution delay Successful execution flag and exception event identifier ,in: Indicates the scheduling response status. This indicates the time delay from the issuance of the schedule to its completion. Indicates whether the execution was successful or failed. Indicates the code for the exception event; S43. The region edge node binds the aforementioned control feedback information with the corresponding control task request identifier and constructs a control feedback record item. All records are written to the local scheduling log, which is used to support subsequent scheduling performance evaluation and scheduling policy updates.

[0031] In this invention, the regional edge nodes match the top-ranked control task requests to the corresponding environmental controllers according to the scheduling task sorting list, and issue the scheduling scheme in the form of control commands. After receiving the scheduling command, each controller executes the corresponding control task and sends back feedback information in real time, including response status code, execution delay, execution success flag, and abnormal event identifier. The regional edge nodes bind each feedback information with the original task identifier to generate a control feedback record item, and write it into the local scheduling log, providing accurate data support for subsequent scheduling performance evaluation and adaptive optimization of scheduling strategies.

[0032] In this embodiment, S5 specifically includes: S51. The main edge node periodically collects the running status of all regional edge nodes and constructs a node state vector for each regional edge node: ; This represents the computational resource utilization rate. Indicates the length of the task queue. Indicates the task success rate. Indicates the cost of task migration; S52. The task success rate is obtained by the regional edge node based on the task execution results recorded in its local scheduling log, reflecting the proportion of control tasks successfully completed in the previous scheduling cycle. S53. The main edge node determines whether there are abnormal nodes based on the node state vector. If a node meets any of the following conditions, it is considered an abnormal node: the computing resource utilization rate exceeds the threshold. The task success rate is below the threshold. Or the number of scheduling failures exceeds the threshold. ; This indicates the maximum acceptable utilization rate of the node's computing resources; This indicates the minimum success rate of task execution; This represents the maximum number of scheduling failures allowed per unit time; the threshold can be set by the main edge node during deployment, or dynamically and adaptively adjusted through runtime history. S54. The primary edge node, combining node idle time, success rate, and migration cost, calculates the node scheduling adaptability score: ; in , , These are adjustable weighting coefficients; S55. If a node is determined to be an abnormal node, the primary edge node selects the node with the highest scheduling score from the remaining nodes as the migration target, and migrates some unscheduled tasks to the node. .

[0033] In this invention, the main edge node periodically collects the operational status of regional edge nodes to construct a node state vector. This vector includes computational resource utilization, task queue length, task success rate, and task migration cost, comprehensively characterizing the node's operational load and scheduling performance. The task success rate is calculated by each node based on the proportion of tasks successfully completed in the previous period according to its local scheduling logs. The main edge node identifies abnormal nodes based on the state vector. When a node's resource utilization exceeds a preset threshold, its success rate falls below a threshold, or its scheduling failure count exceeds a limit, the node is marked as abnormal. A scheduling adaptability score is calculated by combining the node's idle time, success rate, and migration cost. If an abnormal node exists, the node with the highest scheduling score is selected from the remaining nodes as the migration target, and partial task migration operations are performed to achieve global scheduling coordination and task balancing.

[0034] In this embodiment, S6 specifically includes: S61. The main edge node, based on the aggregated task units and regional edge node set within the current scheduling period, Construct scheduling coding individual This indicates that control tasks are assigned to nodes. And introduce a task-node adaptation matrix as a constraint to ensure ; S62. Construct a set of multi-objective optimization functions for each scheduling code, including: Delayed Integrated Objectives: ; in These represent network latency, node queuing latency, and control execution time, respectively. This is the normalized result of the scheduling urgency scoring function value; The goal is to achieve balanced task distribution. ; Where is the number of tasks on the node. This is the average number of tasks across all nodes. Total migration cost target: ; This includes bandwidth requirements. To migrate network overhead, For resource reallocation costs, , , As a weighting factor; S63. Generate the population using a uniform initialization and minimum load priority strategy. Calculate for each individual , , Then perform a non-dominated sort to obtain the front set. And calculate the crowding distance for elite selection; S64. The primary edge node introduces a greedy guidance mechanism based on the current non-dominated front set during the mutation operation of the differential evolution algorithm. Specifically, for each target scheduling encoded individual in the current population... Elite individuals with the highest crowding density are selected from the non-dominated frontier. And construct the guiding mutation vector: ; in: Three distinct individuals are randomly selected from the current population; It is the difference factor; This is the greedy guiding coefficient, which controls the strength of the influence of non-dominant elite individuals on the current mutation search direction. The term serves as a guiding term, using the non-dominated optimal solution to approximate the current individual's direction, guiding the population to converge toward a scheduling scheme with better performance in the leading solution set; S65. For each scheduling coding individual and its corresponding test individual The main edge nodes calculate the objective function values ​​based on a set of multi-objective optimization functions and execute the selection strategy accordingly. If it is not inferior on all targets, then retain ; If with If they do not dominate each other, they are compared according to their non-dominance level and crowding distance, and the better one enters the next generation; If it is controlled, then retain it; Ensure that the next generation of the population maintains balanced evolution in terms of size, non-dominant hierarchy, and solution set diversity. S66, the main edge node sets the maximum number of generations for differential evolution. Once the convergence condition or algebraic termination condition is met, the scheduling code individual with the largest crowding distance and the optimal comprehensive objective value is selected from the current non-dominated frontier set. This will serve as the final scheduling scheme.

[0035] The scheduling code represents the optimal scheduling mapping result of each task in the control task request set on the set of regional edge nodes. The main edge node distributes the scheduling scheme to each regional edge node for execution and initiates the subsequent closed-loop feedback mechanism.

[0036] S67. The main edge node decodes the scheduling scheme corresponding to the optimal scheduling code individual selected in the differential evolution algorithm, and extracts the target region edge node mapped to each control task request. And the scheduling instructions are sent one by one to the corresponding regional edge nodes.

[0037] After receiving the scheduling plan, each edge node in each region is responsible for distributing control tasks to its respective environment controller and driving it to execute corresponding control behaviors. Simultaneously, it records operational feedback information such as controller response status, task completion delay, and resource usage, and constructs feedback data tuples. in: Indicates the actual execution delay of the control task; This indicates the current computing resource utilization rate of the execution node; This sets a status feedback flag for the controller. The feedback data will be synchronously uploaded to the main edge node and used to: update the local scheduling log in real time; through this closed-loop feedback mechanism, the scheduling scheme is converged in a closed loop from optimization, execution to verification, thereby improving the overall system's adaptive scheduling capability and steady-state performance.

[0038] This invention designs a differential evolution algorithm for scheduling optimization based on a multi-objective optimization function set in the main edge node. First, it constructs a scheduling encoding individual to represent the mapping relationship between control tasks and regional edge nodes, and introduces a task-node adaptation matrix as a constraint. Then, it defines a multi-objective optimization function set consisting of a delay synthesis objective, a task distribution balancing objective, and a total migration cost objective, comprehensively measuring the performance of the scheduling scheme in terms of response time, load balancing, and system resource consumption. Population evolution is guided by non-dominated sorting and congestion degree, combined with a greedy mechanism to guide the mutation direction towards the high-quality solution set concentrated in the non-dominated front, accelerating convergence to a better-performing scheduling scheme. Finally, the optimal individual in the non-dominated front is selected as the scheduling scheme. The main edge node decodes it and distributes it to the corresponding regional edge node, completing the mapping and execution of tasks and controllers. Simultaneously, it collects operational feedback information, including task delay, resource consumption, and control status, constructs feedback data tuples, and synchronously transmits them back, forming a closed-loop feedback path. This achieves dynamic integration and performance closed-loop improvement of scheduling optimization, execution, and verification.

[0039] Example 1: To verify the feasibility and practicality of this invention during implementation, it was applied to the automated environmental scheduling system of a large-scale intelligent manufacturing park. The park deploys over 300 distributed environmental controllers across multiple independent production workshops and storage spaces. Control tasks include temperature and humidity regulation, harmful gas emission, equipment heat dissipation control, and cleanliness maintenance. Each controller supports dynamic task assignment and periodic feedback uploads. In actual operation, due to periodic fluctuations and sudden concurrent events in control task triggering, such as batch cleaning, nighttime temperature control, and emergency ventilation, control requests are highly concentrated in specific time periods. Traditional scheduling strategies suffer from response delays and uneven resource allocation, leading to significantly increased execution latency, partial task loss, and control failures, severely impacting the stability of the production environment.

[0040] This embodiment employs a high-concurrency controller real-time scheduling method based on edge computing proposed in this invention. It constructs a master-slave collaborative distributed scheduling system by deploying one master edge node and ten regional edge nodes at the edge of the campus. The regional edge nodes manage and receive tasks from the connected controllers in groups. They capture control task requests in real-time when controllers periodically report or when unexpected events occur, performing token bucket rate limiting, task deduplication, and parameter normalization to construct a standardized set of control tasks. The master edge node optimizes the task-node mapping relationship based on the task load and node status within the current scheduling cycle using a greedy-guided differential evolution algorithm, considering objectives such as latency response, load balancing, and migration costs. This optimized scheduling scheme is then distributed to each regional edge node in real-time as control commands. The regional nodes allocate tasks to their local controllers for execution and collect feedback information such as response latency, execution status, and resource usage, continuously optimizing the scheduling strategy through a closed-loop mechanism.

[0041] Table 1. Performance comparison data of different scheduling methods in high-concurrency control task scenarios.

[0042] During the testing period, we selected a high-load control scenario for 30 consecutive days for comparative experiments, with each scheduling cycle lasting 30 seconds. The comparison algorithms were the commonly used "least recent task node first scheduling algorithm" and the "static priority scheduling algorithm based on a scoring function". In actual operation, the scheduling method proposed in this invention showed significant advantages in terms of latency response, task loss rate, and resource utilization.

[0043] As shown in Table 1, the real-time scheduling method for high-concurrency controllers based on edge computing proposed in this invention outperforms traditional methods in several key performance indicators. Regarding the average latency of control tasks, the method of this invention controls the latency to within 115ms in high-concurrency scenarios, significantly lower than static polling and priority-based scheduling strategies; the task completion rate remains consistently above 98%, demonstrating strong task processing reliability; the number of task migrations and scheduling failure rates are both low, reduced by approximately 32.5% and 67% respectively, indicating that this invention possesses superior load balancing capabilities and anomaly adaptability; simultaneously, the load variance between nodes decreases significantly, further proving that the system scheduling strategy effectively achieves balanced task distribution while maintaining high efficiency. These data fully validate the scheduling optimization effect and practical value of this invention in high-concurrency control environments.

[0044] The method of this invention not only significantly outperforms comparative methods in latency control, but also maintains an extremely low task loss rate, especially under extreme load conditions. Simultaneously, it effectively balances the load of edge nodes in different regions, improving resource utilization and reducing cross-node migration overhead, thereby ensuring system stability and sustainable operation. Furthermore, by constructing a scheduling log through a closed-loop feedback mechanism and dynamically adjusting the scoring weight coefficients, the system possesses high adaptability, continuously optimizing scheduling performance and responding to dynamic environmental changes. The above embodiments fully verify the technical advantages and engineering feasibility of this invention in large-scale practical deployments.

[0045] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-time scheduling method for high-concurrency controllers based on edge computing, characterized in that, Includes the following steps: S1. Deploy a master-slave edge node architecture in the target area. The regional edge nodes connect to and manage the high-concurrency controller of the environment and collect control task requests. S2, Perform task flow control and preprocessing operations; S3. Construct a priority scheduling strategy based on the scheduling urgency scoring function, and perform scheduling for regional edge nodes according to the priority of the scoring value; S4. Issue scheduling results and collect control feedback information, and record them to the local scheduling log; S5. Construct a global scheduling and coordination mechanism based on node state vectors. When abnormal load and scheduling failure are detected at the edge nodes of the region, execute task migration across nodes. S6. The main edge node constructs a scheduling code individual based on the aggregated task unit and the set of regional edge nodes in the current scheduling cycle and defines a multi-objective optimization function group. It combines non-dominated sorting, crowding elite guidance and greedy improvement differential evolution algorithm to optimize the scheduling, generate the optimal scheduling scheme and send it to the regional edge node for execution.

2. The real-time scheduling method for high-concurrency controllers based on edge computing according to claim 1, characterized in that, The node state vector includes computational resource utilization, task queue length, control task success rate, and migration cost.

3. The real-time scheduling method for high-concurrency controllers based on edge computing according to claim 1, characterized in that, S1 specifically includes: S11. Deploy an edge scheduling architecture in the target area, including a main edge node and multiple regional edge nodes. The main edge node has global task coordination and scheduling optimization functions, and the regional edge nodes have access management and scheduling task processing capabilities for high-concurrency environment controllers. S12. The regional edge node completes the registration and access of each environmental high-concurrency controller through wireless communication, and establishes the access mapping relationship between the high-concurrency controller and the accessed regional edge node; S13. Each environment's high-concurrency controller periodically reports control task requests to the edge node of its region when a trigger event occurs. The control task request includes the task type, priority label, deadline, control parameter vector, and the current operating load information of the high-concurrency controller. S14. The region edge node performs data structure standardization processing on the received control task requests, extracts task identifiers, task features, and resource status, and constructs a structured control task request set. .

4. The real-time scheduling method for high-concurrency controllers based on edge computing according to claim 1, characterized in that, S2 specifically includes: S21. The regional edge nodes apply a token bucket mechanism to control the flow of control task request sets, setting the bucket capacity to [value missing]. The token generation rate is When the number of requests exceeds the number of available tokens within a unit of time, the excess tasks are cached to smooth out the high-concurrency request flow. S22. Received set of control task requests Perform data deduplication to retain the latest valid task request within the same time window for the same high-concurrency controller; S23, Set of Control Task Requests Control parameter vector in Perform parameter normalization processing to make All elements Mapping to interval This is used to eliminate the influence of different units on subsequent scoring functions; S24, Based on task type Based on the principle of time proximity, control task requests that meet the merging conditions are aggregated to form aggregated task units. .

5. The real-time scheduling method for high-concurrency controllers based on edge computing according to claim 1, characterized in that, S3 specifically includes: S31, the region edge node aggregation task unit Each control task in the process requests the calculation of a scheduling urgency score function value. ; S32, For aggregation task units All task requests are processed according to the scoring function value. Sort the tasks in descending order to obtain the sorted list of scheduling tasks. ; S33, the region edge nodes, according to the sorted scheduling task list... The control tasks are executed sequentially in the order of sorting, and the scheduling priority is determined by the score function value. The higher the score, the higher the scheduling priority.

6. The real-time scheduling method for high-concurrency controllers based on edge computing according to claim 1, characterized in that, S4 specifically includes: S41. Regional edge nodes are sorted according to the scheduling task list. The system will match the control task requests with the highest priority positions to the corresponding high-concurrency controllers in sequence, and issue the scheduling scheme to the corresponding high-concurrency controllers in the form of control commands. S42. After receiving the scheduling scheme, each high-concurrency controller in the environment executes the corresponding control task and sends back control feedback information in real time. The control feedback information includes a response status code. Actual execution delay Successful execution flag and abnormal event identifiers ; S43. The regional edge node will link the control feedback information with the corresponding control task request identifier. Perform binding and construct control feedback log entries, then write all control feedback log entries to the local scheduling log. middle.

7. The real-time scheduling method for high-concurrency controllers based on edge computing according to claim 1, characterized in that, S5 specifically includes: S51. The main edge node periodically collects the operating status of all regional edge nodes and constructs a data structure for each regional edge node. Node state vector ,in This indicates the utilization rate of computing resources. Indicates the length of the task queue. Indicates the task success rate. Indicates the cost of task migration; S52, Task Success Rate From the edge nodes of the region This is calculated based on the task execution results recorded in the local scheduling log; S53, The main edge node is based on the node state vector Determine if any abnormal nodes exist. If a node exists and meets any of the following conditions, it is considered an abnormal node: Calculate resource utilization. Exceeding the threshold Task success rate Below the threshold The number of scheduling failures exceeded the threshold. ; S54. The primary edge node, combined with node idle time, success rate, and migration cost, calculates the node scheduling adaptability score. ; S55, if node If a node is identified as an abnormal node, the primary edge node will select a node from the remaining nodes based on its node scheduling adaptability score. Largest node As the migration target, some unscheduled tasks will be migrated to the node. .

8. The real-time scheduling method for high-concurrency controllers based on edge computing according to claim 1, characterized in that, S6 specifically includes: S61, The main edge node, based on the aggregated task units within the current scheduling period... and the set of edge nodes in the region Construct scheduling coding individual ,in Indicates control task Assigned to a node And introduce a task-node adaptation matrix. As a constraint, ensure ; S62, For individual scheduling codes Construct a set of multi-objective optimization functions, including: delay synthesis objective Balanced task distribution Total migration cost target The latency comprehensive objective measures the response latency of control tasks during communication, queuing, and execution. The task distribution balance objective reflects the variance of task load at edge nodes in each region. The total migration cost objective evaluates the bandwidth, network, and resource reconfiguration overhead of tasks in cross-node scheduling. S63. Generate the population using a uniform initialization and minimum load priority strategy. For each individual scheduling code Computational delay integrated objective Balanced task distribution Total migration cost target Then perform a non-dominated sort to obtain the front set. And calculate the crowding distance for elite selection; S64. The primary edge node introduces a greedy guidance mechanism based on the current non-dominated front set in the mutation operation of the differential evolution algorithm; for each scheduling coded individual in the current population... From the non-dominated frontier set Select the most crowded elite individuals And construct the guiding mutation vector Controlling elite individuals The strength of the impact on the current mutation search direction; The item is a leading item, utilizing elite individuals. Approaching the current scheduling code individual Direction, guiding the population Converge towards the better-performing scheduling scheme in the leading solution set; S65, For each scheduling code individual and the corresponding test individuals The main edge nodes calculate the objective function values ​​based on a set of multi-objective optimization functions and execute the selection strategy accordingly. S66, The main edge node sets the maximum number of generations for differential evolution. Once the convergence condition is met, the current non-dominated frontier set... Select the individual with the largest congestion distance and the optimal overall objective value for scheduling. This will serve as the final scheduling scheme. S67, The main edge node will schedule the optimal coding individual. The corresponding scheduling scheme is decoded to extract each control task request. Mapped target region edge nodes The scheduling instructions are then sent out one by one to the corresponding regional edge nodes; after receiving the scheduling plan, each regional edge node is responsible for distributing the control tasks to its respective environment high-concurrency controller. It drives the control task to execute corresponding control behaviors, records the high-concurrency controller operation feedback information, and constructs feedback data tuples.

9. A real-time scheduling method for a high-concurrency controller based on edge computing according to claim 8, characterized in that, The selection strategy: if the guiding mutation vector Non-inferior to scheduling code individuals across all objectives Then retain the guiding mutation vector. If the guiding mutation vector With scheduling coding individual If they are not mutually dominant, then compare their non-dominant rank and crowding distance, selecting the individual with the better rank and greater distance to be retained for the next generation; if the mutation vector is guided... If controlled, the scheduling code individual is retained. .

Images