Real-time monitoring and scheduling method and system based on edge cloud cooperation

Abstract

The application discloses a real-time monitoring scheduling method and system based on edge cloud cooperation, belongs to the technical field of edge computing, and comprises the following steps: acquiring a real-time monitoring task; inputting the real-time monitoring task into a global scheduling model to generate a candidate scheduling scheme; simulating the candidate scheduling scheme in a cloud digital twin to generate a scheduling simulation result; determining a target scheduling scheme that meets the processing time delay requirement, a preset resource conflict threshold and a QoE index constraint according to the scheduling simulation result; issuing the target scheduling scheme to a corresponding edge node, starting a containerized service based on resource reservation parameters and accessing a real-time monitoring data stream to execute a task by the edge node; collecting task execution feedback data of the edge node and uploading the task execution feedback data to the cloud for updating the global scheduling model. The application significantly enhances the real-time performance, stability and intelligent level of a monitoring system.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, specifically to a real-time monitoring and scheduling method and system based on edge cloud collaboration. Background Technology

[0002] Edge computing and cloud computing collaborative architecture has become a core technological path supporting real-time monitoring and intelligent scheduling in key scenarios such as the Industrial Internet, smart cities, and the Internet of Vehicles. In recent years, with the maturity of 5G communication, AI inference deployment, and containerized edge runtime, edge-cloud collaborative systems have made significant progress in task offloading, resource scheduling, and low-latency response. The industry generally adopts rule-based or lightweight model-based scheduling strategies, combining edge-side data preprocessing with cloud-based global decision-making to achieve efficient allocation and execution of monitoring tasks. Some advanced systems have introduced technologies such as digital twins and federated learning, initially realizing simulation verification and model self-optimization capabilities, and improving the intelligence level of scheduling.

[0003] However, existing technologies still suffer from the following prominent issues, hindering the reliability and quality of service (QoE) assurance capabilities of real-time monitoring and scheduling systems: Scheduling decisions lack pre-verification mechanisms; candidate solutions are directly issued for execution, making them susceptible to timeouts or QoE failures due to resource contention or network fluctuations; QoE requirements are incompletely expressed, with traditional scheduling focusing only on latency or throughput, failing to incorporate QoE constraints as quantifiable and assessable comprehensive scoring thresholds into the decision-making process, thus hindering consistent user experience; and task feedback data is disconnected from scheduling model updates, lacking a closed-loop mechanism from execution results to model optimization, resulting in scheduling strategies unable to adapt to dynamic environmental changes. These issues lead to resource mismatches, service degradation, and response delays in high-load, multi-task concurrent scenarios, failing to meet the critical monitoring business's scheduling requirements for "predictable, assessable, and evolvable" scheduling. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] To solve the above technical problems, the present invention provides the following technical solution: a real-time monitoring and scheduling method based on edge cloud collaboration, which includes: acquiring real-time monitoring tasks, wherein the real-time monitoring tasks include task type, data source location, processing latency requirements and QoE metric constraints;

[0006] The real-time monitoring task is input into the global scheduling model to generate candidate scheduling schemes, which include target execution nodes, transmission paths and resource reservation parameters.

[0007] The candidate scheduling scheme is simulated in a cloud-based digital twin to generate scheduling simulation results, which include simulation response latency, resource conflict status, and simulation QoE value.

[0008] Based on the scheduling simulation results, a target scheduling scheme that meets the processing latency requirements, the preset resource conflict threshold, and the QoE index constraints is determined.

[0009] The target scheduling scheme is distributed to the corresponding edge nodes, and the edge nodes start containerized services and connect to the real-time monitoring data stream to execute tasks based on the resource reservation parameters.

[0010] The task execution feedback data of the edge nodes is collected and uploaded to the cloud to update the global scheduling model. The task execution feedback data includes task completion time, resource utilization rate and actual QoE value.

[0011] As a preferred embodiment of the real-time monitoring and scheduling method based on edge cloud collaboration described in this invention, the step of acquiring the real-time monitoring task includes:

[0012] Monitor the data acquisition devices in each monitored area and receive monitoring event signals triggered by cameras, sensors, or industrial terminals;

[0013] In response to the monitoring event signal, the event metadata is extracted and an initial monitoring task request is generated. The event metadata includes device identifier, geographic location information and event timestamp.

[0014] Based on a preset task classification rule base, the corresponding task type is determined based on the event type. The task type includes video stream analysis task, real-time alarm detection task, or batch data processing task.

[0015] Based on the service quality configuration strategy of the monitoring system, processing latency requirements and QoE indicator constraints are assigned to the initial monitoring tasks. The processing latency requirements are set to the maximum allowable response time according to the task type, and the QoE indicator constraints are the comprehensive service quality score thresholds corresponding to the tasks, the values ​​of which are configured according to the task type.

[0016] The task type, data source location, processing latency requirements, and QoE metric constraints are encapsulated into a structured real-time monitoring task and uploaded to the cloud scheduling center through a secure communication channel.

[0017] As a preferred embodiment of the real-time monitoring and scheduling method based on edge cloud collaboration described in this invention, the generation of candidate scheduling schemes includes:

[0018] Receive structured real-time monitoring tasks from the edge, and parse the task type, data source location, processing latency requirements, and QoE metric constraints.

[0019] Based on the task type, the corresponding task scheduling strategy sub-model is called from the global scheduling model, wherein the global scheduling model is a deep neural network model generated based on federated learning aggregation;

[0020] The feature vector of the real-time monitoring task is input into the strategy sub-model, and multiple candidate resource allocation combinations are generated by combining the current resource status information of each edge node and the cloud.

[0021] For each candidate resource allocation combination, path planning and bandwidth estimation are performed to determine the corresponding transmission path and the required reserved computing resources, storage capacity and network bandwidth, thus forming a candidate scheduling scheme;

[0022] Output at least one candidate scheduling scheme, which includes the target execution node, transmission path and resource reservation parameters.

[0023] As a preferred embodiment of the real-time monitoring and scheduling method based on edge cloud collaboration described in this invention, the generation of scheduling simulation results includes:

[0024] Based on the physical resource configuration and operating status of the edge nodes, a corresponding digital twin is constructed in the cloud. The digital twin includes a computing resource model, a network connection model, and a task processing queue model.

[0025] Receive the candidate scheduling scheme and parse the target execution node, transmission path, and resource reservation parameters therein;

[0026] The candidate scheduling scheme is injected into the digital twin corresponding to the target execution node to simulate the entire process of task scheduling and execution;

[0027] During the simulation, the time span from data input to result output is recorded to generate simulation response latency; resource contention within the same node is detected to generate resource conflict status; and the simulation QoE value is calculated in reverse based on the task processing quality index.

[0028] The simulation response delay, resource conflict status, and simulation QoE value are summarized into the scheduling simulation results.

[0029] As a preferred embodiment of the real-time monitoring and scheduling method based on edge cloud collaboration described in this invention, the step of determining the target scheduling scheme that satisfies the processing latency requirements, the preset resource conflict threshold, and the QoE metric constraints includes:

[0030] Receive scheduling simulation results, which include simulation response delay, resource conflict status, and simulation QoE value;

[0031] Obtain the processing latency requirements, preset resource conflict thresholds, and QoE constraints corresponding to real-time monitoring tasks;

[0032] Determine whether the simulation response delay is not greater than the processing delay requirement, whether the resource conflict status is lower than the preset resource conflict threshold, and whether the simulation QoE value is not lower than the QoE index constraint; if all are satisfied, then the candidate scheduling scheme is determined as the target scheduling scheme.

[0033] As a preferred embodiment of the real-time monitoring and scheduling method based on edge cloud collaboration described in this invention, the step of starting the containerized service and accessing the real-time monitoring data stream to execute the task includes:

[0034] The target scheduling scheme is encapsulated into a scheduling instruction, which includes the target execution node identifier, transmission path configuration information, and resource reservation parameters.

[0035] The scheduling instruction is sent to the edge node corresponding to the target execution node identifier via an encrypted communication link;

[0036] The edge node parses the resource reservation parameters and calls the local container runtime to create a microservice container instance with the corresponding resource quota;

[0037] The edge node establishes a real-time data stream connection based on the data source location, and inputs the received real-time monitoring data stream into the microservice container instance to perform task processing.

[0038] As a preferred embodiment of the real-time monitoring and scheduling method based on edge cloud collaboration described in this invention, uploading the task execution feedback data to the cloud includes:

[0039] During the execution of the target scheduling scheme, monitor the running status of microservice container instances and collect task completion time, resource utilization, and actual QoE value;

[0040] The task completion time, the resource utilization rate, and the actual QoE value are combined to form task execution feedback data;

[0041] The task execution feedback data is uploaded to the cloud scheduling center via an encrypted communication link.

[0042] This invention provides a real-time monitoring and scheduling system based on edge cloud collaboration.

[0043] To address the aforementioned technical problems, the present invention further provides the following technical solution: a real-time monitoring and scheduling system based on edge cloud collaboration, comprising: an edge-side monitoring agent module, used to acquire real-time monitoring tasks, wherein the real-time monitoring tasks include task type, data source location, processing latency requirements, and QoE metric constraints;

[0044] The candidate scheduling scheme module is used to input the real-time monitoring task into the global scheduling model and generate a candidate scheduling scheme, which includes the target execution node, transmission path and resource reservation parameters.

[0045] The scheduling simulation processing module is used to simulate the candidate scheduling scheme in a cloud-based digital twin and generate scheduling simulation results, which include simulation response delay, resource conflict status, and simulation QoE value.

[0046] The target scheduling processing module is used to determine a target scheduling scheme that meets the processing latency requirements, the preset resource conflict threshold, and the QoE index constraints based on the scheduling simulation results.

[0047] The target scheduling and execution module is used to distribute the target scheduling scheme to the corresponding edge nodes, and the edge nodes start the containerized service and access the real-time monitoring data stream to execute the task based on the resource reservation parameters;

[0048] The task execution feedback module is used to collect task execution feedback data from the edge nodes and upload the task execution feedback data to the cloud to update the global scheduling model. The task execution feedback data includes task completion time, resource utilization rate, and actual QoE value.

[0049] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the real-time monitoring and scheduling method based on edge cloud collaboration.

[0050] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the real-time monitoring and scheduling method based on edge cloud collaboration.

[0051] The beneficial effects of this invention are as follows: By constructing an edge cloud collaborative scheduling mechanism that integrates digital twin simulation verification, QoE index constraint judgment, and closed-loop feedback optimization, this invention significantly improves the reliability, service assurance capability, and system adaptability of real-time monitoring task scheduling. By simulating and verifying candidate scheduling schemes in a cloud-based digital twin, the simulation response latency, resource conflict status, and simulation QoE value are predicted in advance. Execution is only performed when processing latency requirements, preset resource conflict thresholds, and QoE constraints are met, achieving a "verify first, decide later" secure scheduling mode. The QoE constraint is defined as a comprehensive service quality score threshold based on task type, weighted by multiple dimensions such as frame rate, resolution, and response latency, ensuring that scheduling decisions consider both system performance and user experience. By collecting task completion time, resource utilization, and actual QoE values ​​and feeding them back to the cloud to update the global scheduling model, a closed-loop mechanism of "execution-feedback-optimization" is formed, improving the adaptability of scheduling strategies to dynamic environments. Simultaneously, it supports differentiated scheduling and precise resource allocation for multiple task types, effectively reducing resource mismatch risks and improving edge resource utilization efficiency. Overall, the scheduling process is predictable, assessable, and evolvable, significantly enhancing the real-time performance, stability, and intelligence of the monitoring system. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 The overall flowchart of a real-time monitoring and scheduling method based on edge cloud collaboration provided in one embodiment of the present invention is shown below.

[0054] Figure 2 A computer device diagram illustrating a real-time monitoring and scheduling method based on edge cloud collaboration provided in an embodiment of the present invention. Detailed Implementation

[0055] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0056] Example 1, referring to Figure 1This is the first embodiment of the present invention, which provides a real-time monitoring and scheduling method based on edge cloud collaboration, including:

[0057] S1: Obtain real-time monitoring tasks, which include task type, data source location, processing latency requirements, and QoE metric constraints.

[0058] S2: Input the real-time monitoring task into the global scheduling model to generate candidate scheduling schemes. The candidate scheduling schemes include target execution nodes, transmission paths and resource reservation parameters.

[0059] S3: Simulate candidate scheduling schemes in a cloud-based digital twin and generate scheduling simulation results, including simulation response delay, resource conflict status, and simulation QoE value;

[0060] S4: Based on the scheduling simulation results, determine the target scheduling scheme that meets the processing latency requirements, preset resource conflict thresholds, and QoE index constraints;

[0061] S5: Distribute the target scheduling plan to the corresponding edge nodes, and the edge nodes start the containerized service based on the resource reservation parameters and connect to the real-time monitoring data stream to execute the task;

[0062] S6: Collect task execution feedback data from edge nodes and upload the task execution feedback data to the cloud for updating the global scheduling model. The task execution feedback data includes task completion time, resource utilization, and actual QoE value.

[0063] It should be noted that edge monitoring tasks are characterized by high real-time requirements, limited resources, and large network fluctuations. During their operation, the task type, data source location, processing latency requirements, and Quality of Service (QoE) requirements change dynamically, leading to problems such as uncertain execution node selection, uneven resource allocation, and fluctuating response latency in task scheduling. At the same time, edge nodes have limited computing power, making it difficult to independently complete global optimization decisions for complex tasks. Relying entirely on cloud scheduling introduces additional latency due to network transmission, compromising real-time requirements. Furthermore, existing scheduling methods mostly use static rules or empirical strategies, lacking a pre-verification mechanism for scheduling results, which can easily lead to resource conflicts or substandard service quality. Moreover, performance feedback data during task execution is scattered across various edge nodes, lacking an effective collaborative learning mechanism to achieve continuous optimization of the scheduling model.

[0064] Therefore, to address the aforementioned issues such as delayed scheduling decisions, high risk of resource conflicts, uncontrollable service quality, and difficulties in model updates, steps S1-S6 are employed. A global scheduling model is used to generate candidate scheduling schemes, which are then simulated and verified using digital twins of edge nodes built in the cloud. Based on the simulation results, a target scheduling scheme that meets the original task constraints is selected, achieving pre-emptive security and compliance verification of scheduling decisions. By uploading task execution feedback data to the cloud to update the global scheduling model, a closed-loop learning mechanism of "decision-execution-feedback-optimization" is formed, enhancing the adaptive capability of the scheduling strategy. Simultaneously, QoE constraints are used as one of the core judgment conditions to ensure the service quality of high-priority monitoring tasks. Ultimately, this achieves an overall improvement in the real-time performance, reliability, and intelligence level of monitoring task scheduling in an edge-cloud collaborative environment.

[0065] Example 2, referring to Figure 1, is the second embodiment of the present invention, which provides a real-time monitoring and scheduling method based on edge cloud collaboration.

[0066] S1: Obtain real-time monitoring tasks, which include task type, data source location, processing latency requirements, and QoE metric constraints.

[0067] It should be noted that edge-side monitoring tasks are characterized by diverse types, latency sensitivity, and varying service quality requirements. Incomplete task information or a lack of structured modeling can lead to unfounded scheduling decisions, resulting in resource mismatches or response timeouts. Furthermore, traditional task reporting methods do not explicitly include QoE (Quality of Service) constraints, making it difficult to guarantee the service quality of critical business operations. Therefore, it is necessary to clearly define the task type, data source location, processing latency requirements, and QoE constraints during the task generation phase to create structured monitoring tasks. This provides a complete input basis for subsequent intelligent scheduling, ensuring that scheduling decisions align with business needs.

[0068] S101: Monitors data acquisition devices in each monitored area and receives monitoring event signals triggered by cameras, sensors, or industrial terminals;

[0069] S102: In response to a monitoring event signal, extract event metadata and generate an initial monitoring task request. The event metadata includes device identifier, geographic location information and event timestamp.

[0070] S103: Based on the preset task classification rule library, determine the corresponding task type based on the event type. The task type includes video stream analysis task, real-time alarm detection task, or batch data processing task.

[0071] S104: Combining the service quality configuration strategy of the monitoring system, assign processing latency requirements and QoE index constraints to the initial monitoring task. The processing latency requirement sets the maximum allowable response time according to the task type, and the QoE index constraint is the comprehensive service quality score threshold corresponding to the task, the value of which is configured according to the task type.

[0072] S105: Encapsulate the task type, data source location, processing latency requirements, and QoE metric constraints into a structured real-time monitoring task, and upload it to the cloud scheduling center through a secure communication channel.

[0073] Specifically, acquiring real-time monitoring tasks is the initial triggering step in the scheduling process. The generation of real-time monitoring tasks depends on the perception and structured encapsulation of monitoring events. In this embodiment, monitoring events are generated by data acquisition devices such as cameras, temperature and humidity sensors, and vibration detectors deployed in industrial plants, traffic intersections, or urban public areas. When the devices detect an abnormal state or reach a preset trigger condition, they generate a monitoring event signal and send it to the edge-side monitoring agent module.

[0074] Furthermore, the edge-side monitoring agent module runs on the edge gateway or local server and has event listening and task preprocessing capabilities. In this embodiment, the edge-side monitoring agent module subscribes to the topic channels of each data acquisition device via the MQTT protocol to receive monitoring event signals in real time; upon receiving an event, it parses the JSON format message reported by the device, extracts the device identifier, geographical location information, and event timestamp to form event metadata, and uses this as the basis to generate an initial monitoring task request.

[0075] Furthermore, the task type is determined by matching according to a preset task classification rule base. In this embodiment, the task classification rule base is stored in edge local non-volatile memory and contains multiple sets of "event feature-task type" mapping relationships: if the event originates from a fixed camera and the content is moving target detection, it is classified as a video stream analysis task; if the event originates from a smoke detector and the status is alarm, it is classified as a real-time alarm detection task; if the event is a batch upload of periodically collected environmental data, it is classified as a batch data processing task; the task type serves as one of the core bases for subsequent scheduling strategy selection.

[0076] Furthermore, the processing latency requirement and QoE metric constraints are dynamically loaded from the Quality of Service (QoS) policy table based on the task type. In this embodiment, the QoS policy table is pre-configured in the edge-side monitoring agent module: for real-time alarm detection tasks, the processing latency requirement is set to no more than 200ms, which serves as a hard condition for judging the feasibility of the scheduling scheme; the QoE metric constraint is 75 points, and its corresponding underlying QoS requirements include whether the alarm response latency is less than 300ms, whether the recognition accuracy is not less than 95%, and whether the user interaction feedback is timely, etc., where each metric serves as the input dimension of the QoE scoring model, and a comprehensive score is generated after weighted calculation. For video stream analysis tasks, the processing latency requirement is 300ms, and the QoE metric constraint is 80 points. Its QoE scoring model includes multiple dimensions such as video frame rate, resolution, decoding continuity, and response latency, with the response latency item having a weight of 20%. The processing latency requirement is used for the feasibility screening of scheduling simulation results, and the QoE metric constraint is used for judging the compliance of QoS standards; the two functions are separated and their hierarchical structure is clear. All parameters are consistent with the "processing latency requirements" and "QoE metric constraints", and are ultimately encapsulated into a structured real-time monitoring task, which is then uploaded to the cloud scheduling center via a TLS encrypted channel.

[0077] S2: Input the real-time monitoring task into the global scheduling model to generate candidate scheduling schemes. The candidate scheduling schemes include target execution nodes, transmission paths and resource reservation parameters.

[0078] It should be noted that the scheduling scheme for monitoring tasks needs to balance response speed, resource utilization, and service quality; traditional rule-based scheduling is difficult to adapt to dynamic environments. While global scheduling models possess intelligent decision-making capabilities, they are prone to generating infeasible solutions if they are not finely modeled in conjunction with task characteristics and system status. Therefore, structured real-time monitoring tasks need to be input into a global scheduling model updated based on federated learning. This model, combined with multi-dimensional features, generates candidate scheduling schemes to ensure the technical feasibility and optimization of the schemes in terms of target execution nodes, transmission paths, and resource reservation parameters, providing high-quality input for subsequent simulation verification.

[0079] S201: Receive structured real-time monitoring tasks from the edge side, and parse the task type, data source location, processing latency requirements, and QoE metric constraints.

[0080] S202: Based on the task type, call the corresponding task scheduling strategy sub-model from the global scheduling model. The global scheduling model is a deep neural network model generated based on federated learning aggregation.

[0081] S203: Input the feature vector of the real-time monitoring task into the strategy sub-model, and combine it with the current resource status information of each edge node and the cloud to generate multiple candidate resource allocation combinations;

[0082] S204: Perform path planning and bandwidth estimation for each candidate resource allocation combination, determine the corresponding transmission path and the required reserved computing resources, storage capacity and network bandwidth, and form a candidate scheduling scheme;

[0083] S205: Output at least one candidate scheduling scheme, which includes the target execution node, transmission path and resource reservation parameters.

[0084] Specifically, inputting real-time monitoring tasks into the global scheduling model and generating candidate scheduling schemes is the core step in achieving intelligent resource allocation. In this embodiment, the global scheduling model is deployed in a cloud-based scheduling center and adopts a graph neural network architecture based on federated learning training. This model can model the topological relationships and resource dependencies between edge nodes and support multi-objective optimization solutions for complex monitoring tasks.

[0085] Furthermore, the global scheduling model includes multiple scheduling strategy sub-models corresponding to different task types. Each sub-model shares the underlying feature extraction layer and makes independent decisions at the output layer. In this embodiment, upon receiving a real-time monitoring task, the system loads the corresponding sub-model from the model library based on the "task type" determined in S1: if the task type is a real-time alarm detection task, the low-latency optimization sub-model is invoked; if the task type is a video stream analysis task, the high-throughput sub-model is invoked; if the task type is a batch data processing task, the energy efficiency priority sub-model is invoked, ensuring that the scheduling strategy accurately matches the business requirements.

[0086] Furthermore, the real-time monitoring task is transformed into a feature vector input to the strategy sub-model. In this embodiment, the feature vector includes task-dimensional features (task type, data source location, processing latency requirements, QoE metric constraints) and system-dimensional features (current CPU utilization, memory availability, and uplink bandwidth status of candidate execution nodes). The system-dimensional features are periodically reported by edge nodes and cached in the cloud resource status database. The strategy sub-model processes the feature vector through a multi-layer fully connected network, outputting multiple feasible resource allocation combinations. Each combination includes a recommended target execution node and its preliminary resource configuration suggestions.

[0087] Furthermore, transmission path planning and refined calculation of resource parameters are performed for each resource allocation combination. In this embodiment, when the target execution node is an edge node, the transmission path adopts a local routing method of "data source → local edge gateway → target edge node", and the resource reservation parameters include reserving 2 CPU cores, 4GB memory, and 100Mbps bandwidth; when the target execution node is in the cloud, the transmission path passes through the operator's backbone network, and the system selects the access point with the lowest latency based on historical RTT data and reserves virtual machine instance resources; all candidate scheduling schemes are encapsulated in JSON format, explicitly including the target execution node, transmission path, and resource reservation parameters.

[0088] S3: Simulate candidate scheduling schemes in a cloud-based digital twin and generate scheduling simulation results, including simulation response delay, resource conflict status, and simulation QoE value;

[0089] It should be noted that the execution performance of candidate scheduling schemes is affected by the dynamic nature of edge node resources, network fluctuations, and task coupling, making it difficult for traditional scheduling to accurately predict the actual operating status. Direct execution may result in unsatisfactory QoE due to resource contention or model bias. Therefore, it is necessary to simulate and verify candidate schemes using a cloud-based digital twin to simulate the task execution process and generate reliable simulation response latency, resource conflict states, and simulation QoE values, ensuring the pre-assessmentability and engineering feasibility of scheduling decisions.

[0090] S301: Based on the physical resource configuration and operating status of the edge nodes, construct the corresponding digital twin in the cloud. The digital twin includes a computing resource model, a network connection model, and a task processing queue model.

[0091] S302: Receive candidate scheduling schemes and parse the target execution node, transmission path and resource reservation parameters in them;

[0092] S303: Inject candidate scheduling schemes into the digital twin corresponding to the target execution node to simulate the entire process of task scheduling and execution;

[0093] S304: During the simulation, record the time span from data input to result output of the task, and generate the simulation response delay; detect resource contention within the same node and generate resource conflict status; calculate the simulation QoE value in reverse based on the task processing quality index;

[0094] S305: Summarize the simulation response delay, resource conflict status, and simulation QoE value into scheduling simulation results.

[0095] Specifically, the physical configuration information of the edge node includes the number of CPU cores, memory capacity, storage space, and network interface bandwidth. A virtual runtime environment is established in the cloud with a one-to-one mapping relationship. This environment replicates the resource limitations and task processing behavior of the actual device. In this embodiment, the virtual runtime environment is implemented through lightweight virtual machine or container technology, and is configured with the same operating system kernel and service dependencies as the actual edge node.

[0096] Furthermore, upon receiving a candidate scheduling scheme, the target execution node is parsed, the corresponding virtual runtime environment is located, the transmission path is converted into a routing strategy in the virtual network, and resource reservation parameters are used to set the maximum computing and network resources that the task can occupy.

[0097] Furthermore, tasks are injected into a virtual runtime environment to drive the simulation of data flow access, resource allocation, and processing. The total time taken from task arrival to completion is recorded in time sequence to generate simulation response latency. The competition for the same resource when multiple tasks run concurrently is detected. When the actual usage exceeds the reserved parameter setting value, it is marked as a resource conflict state.

[0098] Furthermore, the simulation QoE value is calculated according to the evaluation criteria corresponding to the task type. Video tasks are scored based on whether the decoding frame rate and resolution meet the standards, while alarm tasks are judged based on whether the response latency exceeds the processing latency requirements. Finally, the simulation response latency, resource conflict status, and simulation QoE value are packaged into a scheduling simulation result and output to the next processing stage.

[0099] S4: Based on the scheduling simulation results, determine the target scheduling scheme that meets the processing latency requirements, preset resource conflict thresholds, and QoE index constraints;

[0100] It should be noted that the performance data generated by candidate scheduling schemes in the simulation environment needs to be compared with the original task requirements for compliance. Making decisions directly based on experience or a single metric can easily lead to substandard service quality or resource overload. Since the simulation results reflect virtual execution performance, they must be mapped to executable decisions through quantitative criteria. Therefore, the simulation response latency must be compared with the processing latency requirements, the resource conflict status with the preset resource conflict threshold, and the simulation QoE value with the QoE constraint. Only when all three criteria are met simultaneously is the scheme confirmed to be feasible for practical deployment, thereby ensuring the accuracy and security of scheduling decisions.

[0101] S401: Receive scheduling simulation results, which include simulation response delay, resource conflict status, and simulation QoE value;

[0102] S402: Obtain the processing latency requirements, preset resource conflict thresholds, and QoE constraints corresponding to the real-time monitoring task;

[0103] S403: Determine whether the simulation response delay is not greater than the processing delay requirement, whether the resource conflict status is lower than the preset resource conflict threshold, and whether the simulation QoE value is not lower than the QoE index constraint; if all are satisfied, then determine the candidate scheduling scheme as the target scheduling scheme.

[0104] Specifically, determining the target scheduling scheme relies on multi-dimensional constraints to assess the performance of candidate scheduling schemes in the simulation environment. In this embodiment, the scheduling simulation results include simulation response latency, resource conflict status, and simulation QoE value. The system compares these three indicators with the processing latency requirements, preset resource conflict thresholds, and QoE constraints defined by the real-time monitoring task. Only when all three are satisfied is the scheme confirmed as executable.

[0105] Furthermore, the simulation response latency is the total simulation time from task access to data source to result output. In this embodiment, the processing latency requirement for the real-time alarm detection task is no more than 200ms. If the simulation response latency is 190ms, it is determined that the processing latency requirement is met; if it is 210ms, it is determined that the requirement is not met, and the corresponding candidate solution is excluded.

[0106] Furthermore, the resource conflict status is defined as the maximum utilization rate of critical resources of the target execution node during the simulation process, including CPU, memory, and network bandwidth. In this embodiment, the preset resource conflict threshold is set to 95%. If the simulation results show that the peak CPU utilization is 90% and the memory utilization is 88%, both of which are below the threshold, it is determined that there is no resource conflict.

[0107] Furthermore, the simulated QoE value is obtained by weighting key service quality indicators according to the task type. In this embodiment, the evaluation dimensions of the video stream analysis task include frame rate, resolution, decoding frame drop rate, and response latency. After each indicator is calculated according to preset weights, a simulated QoE value of 84 points is generated. When this value is not lower than the QoE indicator constraint of 80 points, it is determined that the service quality requirements are met. When all three conditions are met, the candidate scheduling scheme is determined as the target scheduling scheme.

[0108] S5: Distribute the target scheduling plan to the corresponding edge nodes, and the edge nodes start the containerized service based on the resource reservation parameters and connect to the real-time monitoring data stream to execute the task;

[0109] It should be noted that after the target scheduling plan is decided in the cloud, it must be accurately and reliably sent to the edge nodes and translated into actual execution actions. Unreliable instruction transmission or inconsistent resource configuration will lead to task startup delays, service instances occupying resources out of bounds, or data stream access failures. Furthermore, the edge environment is highly heterogeneous; if containerized services are not precisely configured according to resource reservation parameters, resource contention or performance failures are likely to occur. Therefore, a structured instruction issuance mechanism is needed to ensure lossless transmission of scheduling intentions, and service instances should be dynamically created on the edge side based on resource reservation parameters to accurately access the data stream, ensuring the timeliness and stability of task execution.

[0110] S501: Encapsulate the target scheduling scheme into a scheduling instruction, which includes the target execution node identifier, transmission path configuration information, and resource reservation parameters;

[0111] S502: Send scheduling instructions to the edge node corresponding to the target execution node identifier via an encrypted communication link;

[0112] S503: The edge node parses the resource reservation parameters and calls the local container runtime to create a microservice container instance with the corresponding resource quota;

[0113] S504: Edge nodes establish real-time data stream connections based on the data source location and input the received real-time monitoring data streams into the microservice container instance for task processing.

[0114] Specifically, the target scheduling scheme needs to be converted into executable instructions and transmitted to edge nodes to initiate actual task processing. In this embodiment, the scheduling instructions are encapsulated in a lightweight message format, including the target execution node identifier, transmission path configuration information, and resource reservation parameters. They are sent via the MQTTS protocol based on TLS 1.3 encryption to ensure the security and timeliness of instruction transmission.

[0115] Furthermore, upon receiving the scheduling instruction, the edge node verifies its digital signature to confirm its trustworthiness. It then parses the resource reservation parameters and invokes the local container runtime (such as Docker or containerd) to create a microservice container instance. In this embodiment, the system uses the Linux kernel's cgroups and namespaces mechanisms to isolate container resources, precisely limiting its CPU usage to a maximum of 2 cores and its maximum memory usage to 4GB, preventing it from interfering with other tasks.

[0116] Furthermore, the microservice container instance loads a pre-built monitoring and processing service image, which includes task logic such as video decoding, object detection, or anomaly recognition. In this embodiment, the task type is a video stream analysis task, and the service image is based on the ONNX Runtime and integrates a lightweight YOLOv5s model, supporting efficient real-time inference execution on edge devices.

[0117] Furthermore, the edge nodes establish data access connections based on the data source location information in the scheduling instructions.

[0118] S6: Collect task execution feedback data from edge nodes and upload the task execution feedback data to the cloud for updating the global scheduling model. The task execution feedback data includes task completion time, resource utilization, and actual QoE value.

[0119] It should be noted that continuous optimization of scheduling decisions relies on closed-loop feedback from task execution results. Without real-world operational data, the global scheduling model will be unable to adapt to environmental changes, causing subsequent scheduling schemes to deviate from actual needs. Task completion time reflects scheduling responsiveness, resource utilization rate reflects resource efficiency, and the actual QoE value characterizes user experience achievement; these three together constitute the key samples for model training. Therefore, it is necessary to accurately collect the aforementioned feedback data at the edge and reliably report it to the cloud for updating the global scheduling model, enabling dynamic evolution of scheduling strategies and performance improvements.

[0120] S601: During the execution of the target scheduling scheme, monitor the running status of the microservice container instance and collect task completion time, resource utilization and actual QoE value;

[0121] S602: Combine task completion time, resource utilization, and actual QoE value into task execution feedback data;

[0122] S603: Uploads task execution feedback data to the cloud scheduling center via an encrypted communication link.

[0123] Specifically, the collection of task execution feedback data is carried out throughout the entire task execution process. In this embodiment, the edge node periodically collects the operation metrics of the microservice container instance through the container monitoring agent, including the time span from task startup to output result as the task completion time, the average of CPU utilization, memory usage, and network throughput as the resource utilization rate, and the actual QoE value calculated based on video frame rate maintenance rate, recognition accuracy, and user response latency.

[0124] Furthermore, the task completion time is the time difference from receiving the first frame of data to outputting the final processing result. In this embodiment, it is achieved through a timestamp marking mechanism: the start time is recorded when the container starts and the end time is recorded when the task is completed. The difference between the two is the task completion time, with an accuracy of milliseconds.

[0125] Furthermore, the actual QoE value is calculated using the same scoring model as in the simulation phase, based on the task type. In this embodiment, the video stream analysis task is scored based on whether the decoding frame rate is stable above 25fps, whether the resolution is maintained at 1080P, whether the number of consecutive frame drops is less than 3 frames, and whether the response latency exceeds 400ms; the real-time alarm detection task is scored based on whether the alarm response latency exceeds 300ms, and a weighted average of the false alarm rate and the missed alarm rate, to obtain the actual QoE value. This scoring model remains consistent in both the simulation verification and actual execution phases, and the response latency item is only one of the scoring dimensions and does not affect the judgment of processing latency requirements based on task completion time.

[0126] Furthermore, the edge nodes package the collected task completion time, resource utilization, and actual QoE value into structured feedback data, attach a unique task identifier and timestamp, and upload it to the cloud scheduling center via HTTPS protocol. This feedback data is used to update the training sample set of the global scheduling model, enabling the model to learn the task execution patterns in real-world environments and improve the accuracy and adaptability of subsequent scheduling schemes.

[0127] Example 3 is the third embodiment of the present invention. This embodiment provides a real-time monitoring and scheduling system based on edge cloud collaboration, including:

[0128] The edge-side monitoring agent module is used to obtain real-time monitoring tasks, which include task type, data source location, processing latency requirements, and QoE metric constraints.

[0129] The candidate scheduling scheme module is used to input the real-time monitoring task into the global scheduling model and generate a candidate scheduling scheme, which includes the target execution node, transmission path and resource reservation parameters.

[0130] The scheduling simulation processing module is used to simulate the candidate scheduling scheme in a cloud-based digital twin and generate scheduling simulation results, which include simulation response delay, resource conflict status, and simulation QoE value.

[0131] The target scheduling processing module is used to determine a target scheduling scheme that meets the processing latency requirements, the preset resource conflict threshold, and the QoE index constraints based on the scheduling simulation results.

[0132] The target scheduling and execution module is used to distribute the target scheduling scheme to the corresponding edge nodes, and the edge nodes start the containerized service and access the real-time monitoring data stream to execute the task based on the resource reservation parameters;

[0133] The task execution feedback module is used to collect task execution feedback data from the edge nodes and upload the task execution feedback data to the cloud to update the global scheduling model. The task execution feedback data includes task completion time, resource utilization rate, and actual QoE value.

[0134] Example 4, refer to Figure 2 This is the fourth embodiment of the present invention, which differs from the previous three embodiments in that: if the function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0135] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0136] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0137] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0138] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A real-time monitoring and scheduling method based on edge cloud collaboration, characterized in that: include, Obtain real-time monitoring tasks, which include task type, data source location, processing latency requirements, and QoE metric constraints. The real-time monitoring task is input into the global scheduling model to generate candidate scheduling schemes, which include target execution nodes, transmission paths and resource reservation parameters. The candidate scheduling scheme is simulated in a cloud-based digital twin to generate scheduling simulation results, which include simulation response latency, resource conflict status, and simulation QoE value. Based on the scheduling simulation results, a target scheduling scheme that meets the processing latency requirements, the preset resource conflict threshold, and the QoE index constraints is determined. The target scheduling scheme is distributed to the corresponding edge nodes, and the edge nodes start containerized services and connect to the real-time monitoring data stream to execute tasks based on the resource reservation parameters. The task execution feedback data of the edge nodes is collected and uploaded to the cloud to update the global scheduling model. The task execution feedback data includes task completion time, resource utilization rate and actual QoE value. The generated candidate scheduling scheme includes: Receive structured real-time monitoring tasks from the edge, and parse the task type, data source location, processing latency requirements, and QoE metric constraints. Based on the task type, the corresponding task scheduling strategy sub-model is called from the global scheduling model, wherein the global scheduling model is a deep neural network model generated based on federated learning aggregation; The feature vector of the real-time monitoring task is input into the strategy sub-model, and multiple candidate resource allocation combinations are generated by combining the current resource status information of each edge node and the cloud. For each candidate resource allocation combination, path planning and bandwidth estimation are performed to determine the corresponding transmission path and the required reserved computing resources, storage capacity and network bandwidth, thus forming a candidate scheduling scheme; Output at least one candidate scheduling scheme, wherein the candidate scheduling scheme includes the target execution node, transmission path and resource reservation parameters; The generated scheduling simulation results include: Based on the physical resource configuration and operating status of the edge nodes, a corresponding digital twin is constructed in the cloud. The digital twin includes a computing resource model, a network connection model, and a task processing queue model. Receive the candidate scheduling scheme and parse the target execution node, transmission path, and resource reservation parameters therein; The candidate scheduling scheme is injected into the digital twin corresponding to the target execution node to simulate the entire process of task scheduling and execution; During the simulation, the time span from data input to result output is recorded to generate simulation response latency; resource contention within the same node is detected to generate resource conflict status; and the simulation QoE value is calculated in reverse based on the task processing quality index. The simulation response delay, resource conflict status, and simulation QoE value are summarized into the scheduling simulation results.

2. The real-time monitoring and scheduling method based on edge cloud collaboration as described in claim 1, characterized in that: The task of acquiring real-time monitoring includes: Monitor the data acquisition devices in each monitored area and receive monitoring event signals triggered by cameras, sensors, or industrial terminals; In response to the monitoring event signal, the event metadata is extracted and an initial monitoring task request is generated. The event metadata includes device identifier, geographic location information and event timestamp. Based on a preset task classification rule base, the corresponding task type is determined based on the event type. The task type includes video stream analysis task, real-time alarm detection task, or batch data processing task. Based on the service quality configuration strategy of the monitoring system, processing latency requirements and QoE indicator constraints are assigned to the initial monitoring tasks. The processing latency requirements are set to the maximum allowable response time according to the task type, and the QoE indicator constraints are the comprehensive service quality score thresholds corresponding to the tasks, the values ​​of which are configured according to the task type. The task type, data source location, processing latency requirements, and QoE metric constraints are encapsulated into a structured real-time monitoring task and uploaded to the cloud scheduling center through a secure communication channel.

3. The real-time monitoring and scheduling method based on edge cloud collaboration as described in claim 2, characterized in that: The determination of the target scheduling scheme that satisfies the processing latency requirement, the preset resource conflict threshold, and the QoE metric constraint includes: Receive scheduling simulation results, which include simulation response delay, resource conflict status, and simulation QoE value; Obtain the processing latency requirements, preset resource conflict thresholds, and QoE constraints corresponding to real-time monitoring tasks; Determine whether the simulation response delay is not greater than the processing delay requirement, whether the resource conflict status is lower than the preset resource conflict threshold, and whether the simulation QoE value is not lower than the QoE index constraint; if all are satisfied, then the candidate scheduling scheme is determined as the target scheduling scheme.

4. The real-time monitoring and scheduling method based on edge cloud collaboration as described in claim 3, characterized in that: The step of starting the containerized service and connecting to the real-time monitoring data stream to execute the task includes: The target scheduling scheme is encapsulated into a scheduling instruction, which includes the target execution node identifier, transmission path configuration information, and resource reservation parameters. The scheduling instruction is sent to the edge node corresponding to the target execution node identifier via an encrypted communication link; The edge node parses the resource reservation parameters and calls the local container runtime to create a microservice container instance with the corresponding resource quota; The edge node establishes a real-time data stream connection based on the data source location, and inputs the received real-time monitoring data stream into the microservice container instance to perform task processing.

5. The real-time monitoring and scheduling method based on edge cloud collaboration as described in claim 4, characterized in that: Uploading the task execution feedback data to the cloud includes: During the execution of the target scheduling scheme, monitor the running status of microservice container instances and collect task completion time, resource utilization, and actual QoE value; The task completion time, the resource utilization rate, and the actual QoE value are combined to form task execution feedback data; The task execution feedback data is uploaded to the cloud scheduling center via an encrypted communication link.

6. A real-time monitoring and scheduling system based on edge cloud collaboration, employing the real-time monitoring and scheduling method based on edge cloud collaboration as described in any one of claims 1 to 5, characterized in that, include: The edge-side monitoring agent module is used to obtain real-time monitoring tasks, which include task type, data source location, processing latency requirements, and QoE metric constraints. The candidate scheduling scheme module is used to input the real-time monitoring task into the global scheduling model and generate a candidate scheduling scheme, which includes the target execution node, transmission path and resource reservation parameters. The scheduling simulation processing module is used to simulate the candidate scheduling scheme in a cloud-based digital twin and generate scheduling simulation results, which include simulation response latency, resource conflict status, and simulation QoE value. The target scheduling processing module is used to determine a target scheduling scheme that meets the processing latency requirements, the preset resource conflict threshold, and the QoE index constraints based on the scheduling simulation results. The target scheduling and execution module is used to distribute the target scheduling scheme to the corresponding edge nodes, and the edge nodes start the containerized service and access the real-time monitoring data stream to execute the task based on the resource reservation parameters; The task execution feedback module is used to collect task execution feedback data from the edge nodes and upload the task execution feedback data to the cloud to update the global scheduling model. The task execution feedback data includes task completion time, resource utilization rate, and actual QoE value.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the real-time monitoring and scheduling method based on edge cloud collaboration as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the real-time monitoring and scheduling method based on edge cloud collaboration as described in any one of claims 1 to 5.

Images