An edge-end computing power optimization scheduling method and system for external wall insulation basket construction

Abstract

The application provides a kind of outer wall thermal insulation basket construction edge end computing power optimization scheduling method and system, it is related to building construction intelligent supervision and edge computing technology field, including the following steps: obtaining video data and computing power resource state information in the process of basket construction, generate video processing task, and extract task attribute information;The priority of video processing task is evaluated, and a multi-level task queue is established according to the task priority;The available computing power resources are partitioned, and the multi-level task queue is partitioned and mapped based on the computing power resource partition result;According to the computing power resource state information, the computing power resource partition result is allocated resource quota;Based on the partition resource quota, the task execution strategy is matched and executed, and the task processing result and feedback information are output;A scheduling loss function is constructed, and the parameters in the priority evaluation and resource quota allocation process are updated and adjusted.The application realizes the efficient, stable and adaptive optimization scheduling of the computing power of basket construction edge.

Description

Technical Field

[0001] This invention relates to the field of intelligent supervision and edge computing technology in building construction, and in particular to an edge computing power optimization scheduling method and system for the construction of external wall insulation suspended scaffolding. Background Technology

[0002] With the continuous improvement of the intelligence level of building exterior wall insulation construction, video monitoring and edge computing technologies based on suspended platform operation scenarios are gradually being applied to construction safety supervision and construction quality control. In existing technologies, cameras and edge computing devices are typically deployed in the suspended platform, floor edges, or construction area to collect and analyze the behavior of construction personnel, the operating status of the suspended platform, and the insulation construction process in real time, enabling functions such as safety helmet recognition, safety belt recognition, dangerous action detection, and construction quality verification. Simultaneously, with the increasing application of deep learning visual models in the building construction field, edge devices are gradually undertaking multiple tasks such as multi-channel video stream parsing, target recognition, risk warning, and process recording, thereby improving the intelligent supervision capabilities of construction sites. Especially in weak network environments, complex construction environments, and scenarios with multiple suspended platforms operating in parallel, utilizing local computing power at the edge to complete real-time inference and rapid response has become an important development direction for current intelligent supervision technology for exterior wall insulation construction.

[0003] However, existing technologies still have significant shortcomings in practical applications. First, most existing edge video processing systems use fixed priority or average resource allocation methods, lacking the ability to dynamically schedule tasks based on construction phases, task risk levels, and timeliness requirements. When multiple types of video analysis tasks are generated simultaneously at the construction site, existing systems struggle to effectively prioritize high-risk, high-real-time tasks, easily leading to delays or even timeouts in critical safety task responses. Second, existing technologies typically do not adequately consider issues such as limited computing resources on edge devices, network bandwidth fluctuations, and changes in device thermal status. Under resource-constrained or weak network conditions, high-priority tasks can easily compete with low-priority tasks for computing resources, resulting in a decline in overall system real-time performance, and even issues such as task backlog, task abandonment, and prolonged high-load operation of edge devices. Finally, most existing technologies lack a closed-loop adaptive optimization mechanism based on operational feedback information. They cannot dynamically adjust scheduling parameters and resource quotas according to operational results such as task waiting delay, execution delay, timeout rate, and task drop rate. Therefore, they are difficult to continuously adapt to changes in concurrent tasks and complex working conditions at the construction site, and cannot achieve the technical effects of prioritizing critical tasks, dynamically optimizing edge resources, and ensuring the overall stable operation of the system.

[0004] How to solve the above-mentioned technical problems is the challenge facing this invention. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for optimizing the scheduling of edge computing power in external wall insulation suspended scaffold construction, achieving efficient, stable, and adaptive optimization scheduling of edge computing power.

[0006] The technical solution adopted by this invention to solve its technical problem is as follows: This invention provides a method for edge computing power optimization scheduling in the construction of external wall insulation suspended scaffolding, comprising the following steps: S1. Obtain video data and computing resource status information during the construction process of the suspended platform, generate video processing tasks based on the video data, and extract task attribute information from the video processing tasks. S2. Based on task attribute information and the current construction stage, prioritize video processing tasks to obtain task priorities; and classify and enqueue each video processing task according to the task priority to establish a multi-level task queue. S3. Divide the available computing resources into partitions to obtain the computing resource partitioning results, and map the multi-level task queues into partitions based on the computing resource partitioning results; allocate resource quotas based on the computing resource status information to obtain the partition resource quotas; S4. Match task execution strategies based on partitioned resource quotas, execute task execution strategies for video processing tasks in multi-level task queues, and output task processing results and feedback information. S5. Construct a scheduling loss function based on feedback information, and perform closed-loop update and adjustment of parameters in the priority evaluation and resource quota allocation process according to the scheduling loss function.

[0007] Preferably, the task attribute information includes at least the task type, generation time, deadline, risk level, associated construction stage, estimated computational load, and execution method requirements; the computing power resource status information includes edge device resource status information and network status information.

[0008] Preferably, the priority evaluation of the video processing task in step S2 includes: Based on task attribute information, the initial value of task importance, task urgency, risk level, stage-related quantitative value, and waiting compensation items can be quantitatively calculated or obtained. The task priority is obtained by weighted fusion calculation of the initial value of task importance, task urgency, risk level, stage correlation quantification value, and waiting compensation item; The multi-level task queue includes level 1 tasks, level 2 tasks, level 3 tasks, and level 4 tasks.

[0009] Preferably, the computing resource partitioning result includes at least a real-time protection zone, an elastic processing zone, and a minimum protection processing zone; The partitioning and mapping of multi-level task queues based on computing resource partitioning results includes: Based on the task priority of the multi-level task queue, first-level and second-level tasks are mapped to the real-time protection zone, third-level tasks are mapped to the elastic processing zone, and fourth-level tasks are mapped to the backup processing zone.

[0010] Preferably, the resource quota allocation based on the computing resource status information in step S3 includes: Extract the processor load utilization, physical memory usage, and current processor chip temperature from the edge device resource status information, and extract the current available physical bandwidth of the network from the network status information; The system computing power index is obtained by weighting processor load utilization, physical memory usage and processor chip current temperature. Initialize the quotas for the real-time protection zone, elastic processing zone, and guaranteed processing zone, and adjust the quotas using a dual threshold method based on the current available physical bandwidth of the network and the system computing power index to obtain the partition resource quotas; The method of adjusting quotas using a dual threshold method includes: Set a first index threshold and a second index threshold for the system computing power index, and set a first bandwidth threshold and a second bandwidth threshold for the current available physical bandwidth of the network. When the system computing power index is less than the first index threshold and the current available physical bandwidth of the network is greater than the first bandwidth threshold, it is determined to be in a normal healthy state, and the initial quotas of the real-time protection zone, elastic processing zone and guaranteed processing zone are maintained. When the system computing power index is between the first index threshold and the second index threshold, and the current available physical bandwidth of the network is greater than the first bandwidth threshold, it is determined to be in a state of resource shortage. The quota of the real-time protection zone is increased to the first preset ratio, and the quota of the backup processing zone is reduced to the second preset ratio. When the system computing power index is less than the second index threshold and the current available physical bandwidth of the network is between the second bandwidth threshold and the first bandwidth threshold, it is determined to be in a weak network restricted state. The quota of the real-time protection zone is increased to the third preset ratio, and the quotas of the elastic processing zone and the guaranteed processing zone are reduced to the fourth preset ratio and the fifth preset ratio, respectively. When the system computing power index is greater than or equal to the second index threshold, or when the current available physical bandwidth of the network is less than the second bandwidth threshold, it is determined to be in a critical network outage state. The quota of the real-time protection zone will be adjusted to the preset maximum value, and the quota allocation of the backup processing zone will be suspended.

[0011] Preferably, the task execution strategy includes an immediate execution strategy, a frequency reduction and merging execution strategy, and a delayed execution and cache-supplemented execution strategy; The task execution strategy based on partitioned resource quota matching includes: When the task is determined to be in a normal healthy state, an immediate execution strategy is adopted for all Level 1, Level 2, Level 3 and Level 4 tasks. When a resource shortage is detected, an immediate execution strategy is adopted for Level 1 tasks, a frequency reduction and merging execution strategy is adopted for Level 2 and Level 3 tasks, and a delayed execution and cache-supplemented execution strategy is adopted for Level 4 tasks. When the network is determined to be in a weak and restricted state, an immediate execution strategy is adopted for Level 1 and Level 2 tasks, a frequency reduction and merging execution strategy is adopted for Level 3 tasks, and a delayed execution and cache-supplemented execution strategy is adopted for Level 4 tasks. When a critical network outage is detected, an immediate execution strategy is adopted for Level 1 tasks, while a delayed execution and cache-based execution strategy is adopted for Level 2, Level 3, and Level 4 tasks.

[0012] Preferably, the feedback information includes at least task waiting delay, task execution delay, timeout rate, and task drop rate; The scheduling loss function is obtained by weighted fusion of task waiting delay, task execution delay, timeout rate and task drop rate; In step S5, the Adam adaptive optimization algorithm is used to perform closed-loop update and adjustment of parameters in the priority evaluation and resource quota allocation process based on the scheduling loss function.

[0013] This invention also provides an edge computing power optimization scheduling system for the construction of external wall insulation suspended scaffolding, comprising: The task generation and attribute extraction module is used to acquire video data and computing resource status information during the construction process of the suspended platform, generate video processing tasks based on the video data, and extract task attribute information from the video processing tasks. The priority assessment and grading module is used to assess the priority of video processing tasks based on task attribute information and the current construction stage to obtain task priorities; and to classify and enqueue each video processing task according to the task priority to establish a multi-level task queue. The resource quota module is used to partition available computing resources, obtain computing resource partitioning results, and perform partition mapping on multi-level task queues based on the computing resource partitioning results; and to allocate resource quotas based on computing resource status information to obtain partition resource quotas. The strategy execution module is used to match task execution strategies based on partition resource quotas, execute task execution strategies for video processing tasks in multi-level task queues, and output task processing results and feedback information. The feedback update module is used to construct a scheduling loss function based on feedback information, and to perform closed-loop updates and adjustments to the parameters in the priority evaluation and resource quota allocation process according to the scheduling loss function.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described edge computing power optimization scheduling method for the construction of external wall insulation suspended scaffolding.

[0015] The present invention also provides a computer storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described edge computing power optimization scheduling method for the construction of external wall insulation suspended scaffolding.

[0016] The beneficial effects of this invention are as follows: It achieves efficient, stable, and adaptive optimization scheduling of computing power at the edge of suspended platform construction. By combining a dynamic priority evaluation mechanism based on construction stage, task risk level, and waiting time, and a priority-based hierarchical queuing strategy, the execution order of safety and quality tasks can be adjusted in real time according to the switching of construction stages, giving priority to high-real-time safety tasks. This effectively solves the problem of delayed response to critical tasks caused by fixed priority or average allocation methods in existing technologies. Based on computing power resource status information, available computing power is partitioned, and a dual-threshold method is used to dynamically allocate resource quotas for real-time protection zones, elastic processing zones, and guaranteed processing zones. This ensures priority supply of computing power for high-priority tasks under resource constraints or weak network conditions, suppresses ineffective contention for computing power by low-priority tasks, and significantly reduces task backlog and task drop rate under high load conditions. Based on feedback information, a scheduling loss function is constructed, and the Adam adaptive optimization algorithm is used to update and adjust the priority evaluation parameters and resource quota allocation parameters in a closed loop. This enables the scheduling strategy to continuously evolve with the dynamic changes in construction task distribution, resource status, and network conditions, overcoming the lack of adaptability caused by the lack of self-correction mechanism in existing technologies, and achieving long-term stability and efficient utilization of edge computing power. Attached Figure Description

[0017] Figure 1 This is a diagram illustrating the method steps of the present invention.

[0018] Figure 2 This is a system module diagram of the present invention.

[0019] Figure 3 This is a diagram of the internal structure of a computer device according to Embodiment 3 of the present invention. Detailed Implementation

[0020] To clearly illustrate the technical features of this solution, the following detailed implementation method will be used to explain the solution.

[0021] Example 1: See Figure 1 As shown, this embodiment is an edge computing power optimization scheduling method for external wall insulation suspended platform construction. This method is applied to the edge devices of the suspended platform in the external wall insulation construction scenario. The edge devices can be deployed on the suspended platform body, the suspended platform control box, or at the edge of the construction floor, and are used for local scheduling and computing power allocation of various video processing tasks generated at the construction site. This method is applicable to scenarios with multiple buildings and multiple suspended platforms in parallel construction, and can prioritize the scheduling and processing of key construction tasks under conditions of network fluctuations, limited edge computing power, and changes in task concurrency. Specifically, it includes the following steps: S1. Obtain video data and computing resource status information during the construction of the suspended platform, generate video processing tasks based on the video data, and extract task attribute information from the video processing tasks; the task attribute information includes at least the task type, generation time, deadline, risk level, associated construction stage, estimated calculation volume, and execution method requirements; the computing resource status information includes edge device resource status information and network status information. It should be noted that the video processing tasks in this embodiment are not limited to a single visual recognition task, but include various types such as safety recognition tasks, construction quality recognition tasks, and auxiliary processing tasks. Safety recognition tasks may include safety helmet recognition, safety belt recognition, anomaly detection in personnel numbers, dangerous action recognition, and collision risk detection; construction quality recognition tasks may include plaster integrity recognition, insulation board laying status recognition, anchor installation sequence recognition, and construction action standardization recognition; auxiliary processing tasks may include video compression, keyframe storage, process recording generation, and report output.

[0022] Furthermore, the method includes step S1, acquiring video data and computing resource status information during the construction process of the suspended platform, generating a video processing task based on the video data, and extracting task attribute information from the video processing task; the task attribute information includes at least the task type, generation time, deadline, risk level, associated construction stage, estimated computational load, and execution method requirements; the computing resource status information includes edge device resource status information and network status information.

[0023] Specifically, at least one video acquisition device is installed at the suspended platform construction site. This device can be one or a combination of visible light cameras, infrared cameras, wide-angle cameras, or pan-tilt cameras. Some cameras face the construction operation area to capture footage of the construction process, including plastering, panel installation, anchoring, and inspection; others face the area where construction workers are active to capture footage related to their safety behaviors. The camera device connects to the edge device via wired or wireless communication and continuously transmits real-time video streams to the edge device.

[0024] It should be noted that the edge device in this embodiment can be an industrial edge computing host, an embedded AI processing terminal, a GPU edge node, or an intelligent control device with local inference capabilities, and its internal components include at least a processor, memory, and a network communication module. The edge device is used to receive video data and perform video parsing, task abstraction, task scheduling, and inference execution locally.

[0025] Furthermore, after receiving video data, the edge device performs frame extraction, timestamp marking, and scene correlation analysis on the video stream, and generates corresponding video processing tasks based on preset task triggering rules. Task triggering rules may include event triggering rules, periodic triggering rules, and hybrid triggering rules.

[0026] Event-triggered rules generate corresponding tasks when specific construction activities or risk events are detected. For example, a safety identification task can be triggered when construction workers enter the suspended platform, the platform undergoes raising or lowering, or people gather in the construction area, as detected in a video. Periodic trigger rules generate tasks periodically at preset time intervals, such as performing a quality verification task on the insulation board laying status every preset time. Hybrid trigger rules combine event conditions and time conditions to generate tasks, balancing real-time performance and stability.

[0027] It should also be noted that after generating video processing tasks, the edge devices further extract corresponding task attribute information for each video processing task. Task type represents the category to which the current task belongs; generation time represents the time the task enters the scheduling system; deadline represents the latest time the task is allowed to complete processing; risk level represents the degree of construction risk corresponding to the task; associated construction stage represents the relationship between the task and the current construction stage; estimated computational load represents the estimated edge computing resources consumed during task execution; and execution method requirements represent the allowed execution methods for the task.

[0028] Furthermore, the associated construction stages can include any one of the following: the suspended platform self-inspection stage, the suspended platform lifting stage, the plastering stage, the insulation board installation stage, the anchoring construction stage, the re-inspection stage, and the completion stage. The edge equipment can identify the current construction stage by combining the current construction time, the suspended platform's operating status, video behavior characteristics, and the construction plan information issued by the management platform, and write the identification results into the task attribute information.

[0029] Furthermore, different construction phases correspond to different task focuses. For example, during the suspended platform lifting phase, the system increases its focus on tasks such as identifying abnormal personnel, detecting collision risks, and recognizing safety belts; during the plastering phase, the system increases its focus on recognizing the integrity of the plastering; and during the anchoring construction phase, the system increases its focus on recognizing the installation sequence of anchors and the standardization of construction actions.

[0030] Furthermore, the computing resource status information is used to characterize the current resource operation status of edge devices and the network operation status. Among them, the edge device resource status information may include one or more of the following: processor load utilization, memory utilization, storage space utilization, processor temperature, and the number of currently queued tasks; the network status information may include one or more of the following: current network bandwidth, network latency, network jitter, packet loss rate, and network connectivity status.

[0031] It should be noted that the edge device can collect the above computing power resource status information in real time according to the preset resource sampling period, and write the collection results into the resource status cache corresponding to the current scheduling period, so as to be called in the subsequent task priority evaluation, resource partitioning and execution strategy adjustment process.

[0032] Furthermore, in some implementations, when network status information indicates that the current network is in a weak network state or a network outage state, the edge device can increase the proportion of local cache resources in advance and limit the immediate execution ratio of non-critical auxiliary tasks, so as to reserve more local computing resources for high-priority task processing.

[0033] Through the above methods, this embodiment can achieve unified abstract modeling of construction site video tasks, and integrate construction stages, risk levels, task timeliness, and edge resource status, providing a unified data foundation and status input for subsequent task priority evaluation, multi-level queue scheduling, and dynamic allocation of edge computing power.

[0034] S2. Based on task attribute information and the current construction stage, prioritize video processing tasks to obtain task priorities; and classify and enqueue each video processing task according to the task priority to establish a multi-level task queue. Furthermore, this step primarily aims to dynamically quantify and evaluate the urgency of different video processing tasks under conditions of limited computing resources on edge devices and concurrent generation of multiple video processing tasks. Based on the evaluation results, a corresponding multi-level task queue is established so that edge devices can subsequently schedule resources and control execution according to task priority. It should be noted that the task priority is not a fixed static priority, but a dynamic priority result that combines changes in the construction phase, task waiting status, and the dynamic changes in edge resource status.

[0035] Specifically, after receiving the video processing tasks generated in step S1, the priority evaluation module in the edge device first reads the task attribute information corresponding to each video processing task, and performs priority evaluation processing on each video processing task in conjunction with the current construction stage. The task attribute information participating in the priority evaluation includes at least the task type, deadline, risk level, associated construction stage, estimated calculation workload, and execution method requirements.

[0036] First, the priority evaluation module obtains the initial value of task importance by directly reading from a static importance mapping table preset in local non-volatile memory, based on task attribute information. In this embodiment, the system pre-defines the following primary tasks: fall risk, collision warning, abnormal number of people, and emergency stop events. A value of 1 indicates a secondary task such as helmet malfunction, seatbelt malfunction, or violation of critical action rules. The value is 0.7; this corresponds to the third-level quality tasks such as plaster integrity identification, insulation board laying status verification, and anchor sequence verification. The value is 0.4; the corresponding level four auxiliary tasks are video compression, report generation, and historical analysis. The value is 0.1 Furthermore, the priority evaluation module first assesses the urgency of each video processing task based on the time difference between the task deadline and the current system time. Perform calculations. Specifically, set up the task. The timestamp generated by the visual event is The longest allowed response deadline for this task is The current real-time physical timestamp obtained by the system clock currently running on the edge device is .but The adaptive segmented time-efficiency calculation formula is as follows:

[0037] For security-related emergency missions, The preset allowed time limit is extremely short (e.g., 200ms), the current time A slight movement is enough to make It is rapidly approaching version 1.0; however, for background auxiliary tasks, the allowed time limit difference is extremely large (e.g., 3600 seconds), therefore its The values ​​have remained at a low level for an extended period. It should be noted that by pre-setting a shorter allowable response time for security identification tasks, the real-time processing capability of these tasks during the overall scheduling process is improved.

[0038] It should also be noted that the priority assessment module further considers the risk level corresponding to the task. This involves adjusting task priorities based on risk levels. This risk level, ranging from [0,1], is used to characterize the potential construction safety or quality risk level of the current task. It is obtained by directly retrieving the internally configured risk level vector based on the input task code. For high-risk tasks, the system increases the priority weight of the corresponding task; for low-risk tasks, it decreases the priority weight. In this embodiment, tasks that may directly affect personnel safety or equipment operational safety are assigned the highest risk. =1.0; Tasks that affect construction quality verification but do not directly trigger immediate hazards are assigned a medium risk. =0.5; assign the lowest risk to auxiliary tasks used solely for recording, archiving, or subsequent statistics. =0.

[0039] Furthermore, the priority evaluation module also combines the matching relationship between the task type of the video processing task and the associated construction stage to obtain the stage association metric value. This allows for phase-based adjustments to task priorities. The system pre-establishes a phase-based mapping relationship between task types and construction phases. Specifically, this involves a phase-based mapping matrix for maintenance within the edge equipment. The stage correlation mapping matrix is ​​used to characterize the degree of correlation between different task types and different construction stages. In the matrix, the row dimension corresponds to different task types, and the column dimension corresponds to different construction stages; the construction stages include at least the start-up self-inspection stage, the lifting and positioning stage, the plastering stage, the paneling stage, and the anchoring stage.

[0040] It should be noted that the variables in the current construction phase The current construction stage can be determined from the suspended platform's operational status information, suspended platform height sensor data, construction plan configuration data, or stage status information issued by the management terminal. After obtaining the current construction stage, the system uses the stage correlation mapping matrix based on the task type corresponding to the current video processing task. Read the corresponding stage-related metric value This serves as a stage-related influencing factor in the task priority evaluation process.

[0041] Furthermore, the stage-related metric value This is used to characterize the degree of correlation between the current task and the current construction phase. Specifically, when the task type is highly correlated with the current construction phase, the phase correlation metric of the corresponding task is increased; when the task type is less correlated with the current construction phase, the phase correlation metric of the corresponding task is decreased.

[0042] For example, when the current construction phase is the lifting and positioning phase, the system increases the phase-related metric values ​​for the collision risk detection task, the abnormal personnel behavior recognition task, and the seat belt recognition task. This also reduces the stage-related quantification values ​​for the tasks of identifying plaster integrity and insulation board laying status, and lowers the corresponding values ​​for these tasks. When the current construction phase switches to the plastering phase, the system increases the phase-related metric value corresponding to the plastering integrity identification task. And correspondingly reduce the stage-related metric values ​​of tasks that are less relevant to the current construction stage. This allows the limited computing resources of edge devices to be dynamically allocated to currently relevant tasks as the construction pace changes.

[0043] It should also be noted that, in some implementations, the stage-related metric value... Configuration can be done using a hierarchical quantification method, such as using a three-level correlation degree division of high correlation, medium correlation and low correlation; or it can be quantified using a continuous numerical method to improve the fine-grained control of priority adjustment between different tasks.

[0044] Through the above methods, the system can dynamically adjust the stage association weights of different task types according to the changes in the construction stage, thereby improving the matching degree between the task priority evaluation results and the actual construction process, and enhancing the dynamic adaptability of edge computing resources under different construction stages.

[0045] Furthermore, the priority evaluation module also incorporates task waiting latency to perform waiting compensation processing on each video processing task. Specifically, as the time a task spends in the waiting queue gradually increases, the system progressively raises the priority of the corresponding task to avoid task starvation caused by low-priority tasks being unable to obtain execution resources for extended periods. In this embodiment, the system's preset maximum backlog aging time threshold is [value missing]. Waiting for compensation The quantification formula is as follows:

[0046] Furthermore, after completing the above evaluations and calculations, the priority evaluation module can use a weighted fusion method to calculate the corresponding task priority results. For example, the priority calculation formula can be expressed as:

[0047] in, This is the initial value for task importance; Based on the urgency of the task; Risk level; This is a stage-related metric. For items awaiting compensation; and The corresponding weighting coefficients can be determined by combining preset rules, on-site configuration, and operational feedback.

[0048] It should also be noted that, in some implementations, to reduce the continuous load impact on the overall system operation caused by high-computational-power-consuming tasks under edge resource constraints, the priority evaluation module further introduces execution benefits and computational cost on the basis of the basic priority, and performs comprehensive priority calculation on video processing tasks to obtain a comprehensive priority. Execution benefits characterize the actual control value or safety response value generated at the construction site after the task is completed, while computational cost characterizes the degree of occupation of edge device computing and storage resources during task execution. Specifically, when a task can directly trigger on-site audible and visual alarm components, safety braking components, or risk linkage control components upon successful execution, the system increases the execution benefits of the corresponding task; when a task is only used to send structured alarm information, write local logs, or generate process records, the execution benefits of the corresponding task are appropriately reduced; for background tasks mainly used for offline statistics, process archiving, or report generation, the execution benefits of the corresponding task are further reduced. Meanwhile, when loading the visual analysis model, the system reads the corresponding theoretical computational load parameters and inference resource usage parameters from the model configuration data. It then combines the current available processing power and storage resources of the edge device to normalize the resource consumption during task execution, thereby obtaining the computational cost of the corresponding task. For tasks with large theoretical computational loads, high inference resource usage, or long-term continuous occupation of edge resources, the system increases the computational cost of the corresponding task. This appropriately suppresses the impact of high resource consumption tasks on the overall real-time scheduling capability of the edge device during the comprehensive priority calculation process, ensuring the priority execution capability of critical real-time tasks under limited edge computing power.

[0049] Furthermore, after obtaining the task priority corresponding to each video processing task, the edge device further performs hierarchical queuing processing on each video processing task according to the task priority to establish a corresponding multi-level task queue. Specifically, the system divides each video processing task into multiple different priority levels according to the task priority and maps them to the corresponding task queues. In this embodiment, the tasks are ultimately divided into four queue levels: Level 1 tasks, Level 2 tasks, Level 3 tasks, and Level 4 tasks. Level 1 tasks correspond to tasks with the highest overall priority, preferably including tasks that require priority execution, such as fall risk, collision warning, abnormal number of people, and emergency stop events; Level 2 tasks preferably include tasks with strong constraints on on-site operations, such as helmet malfunction, safety belt malfunction, and violations of key actions; Level 3 tasks preferably include tasks that allow short-term queuing, such as plaster integrity identification, insulation board laying status verification, and anchor sequence verification; Level 4 tasks preferably include auxiliary tasks that can be executed or delayed during low-load periods, such as video compression, report generation, and historical data organization.

[0050] Furthermore, the multi-level task queue supports mechanisms such as queue insertion, queue rearrangement, time-slice rotation, and waiting aging. When a level-one task arrives, it can trigger the pause, postponement, or degradation of level-three and level-four tasks to prioritize the release of edge computing resources. When high-priority tasks continuously occupy resources for more than a preset duration, the system can allocate a minimum execution window for level-three and level-four tasks to avoid long-term backlog of low-priority tasks. When resources are scarce and there are no level-one tasks, low-priority tasks of the same model, stage, or time can be merged for execution to reduce model reloading and data processing overhead.

[0051] Through the above methods, this embodiment can dynamically prioritize video processing tasks by combining changes in construction phases, task risk levels, task timeliness requirements, and edge device resource status, and establish corresponding multi-level task queues, thereby providing a reliable task organization foundation for subsequent edge computing resource partitioning and task execution strategy scheduling.

[0052] S3. Divide the available computing resources into partitions to obtain the computing resource partitioning results, and map the multi-level task queues into partitions based on the computing resource partitioning results; allocate resource quotas based on the computing resource status information to obtain the partition resource quotas; After completing the task priority assessment and multi-level queue establishment in step S2, this embodiment further differentiates the limited local computing power of edge devices based on the real-time collected computing power resource status information, and dynamically allocates resource quotas to task queues of different levels according to the partitioning results, forming a quantifiable and adjustable task execution strategy.

[0053] It should be noted that the total available computing power resources of edge devices are limited by hardware configuration and will dynamically fluctuate during actual operation due to factors such as processing load, memory usage, chip temperature, and network conditions. The "partitioning" in this step is not physical isolation, but rather a logical allocation of resource quotas, designed to define differentiated resource guarantee boundaries for tasks of different urgency levels. In this embodiment, the available edge resources are divided into a real-time guarantee zone, a flexible processing zone, and a minimum guarantee processing zone. The real-time guarantee zone is prioritized for Level 1 and Level 2 tasks, ensuring the timely execution of tasks such as fall risk, collision warnings, abnormal personnel numbers, emergency stop events, helmet malfunctions, and seatbelt malfunctions. The flexible processing zone is used by medium-priority tasks to dynamically compete for resources based on task priority, construction stage, and current resource status, preferentially serving quality-related tasks such as plaster integrity identification, insulation board laying status verification, and anchor sequence verification. The minimum guarantee processing zone provides a minimum execution quota for low-priority tasks, preferentially serving auxiliary processing tasks such as video compression, report generation, historical analysis, and process record organization, to prevent low-priority tasks from being unable to obtain processing opportunities for extended periods.

[0054] Specifically, the computing resource status information collected in step S1 is read at fixed intervals. The computing resource status information includes edge device resource status information and network status information. The edge device resource status information includes at least the current processing load utilization rate. Physical memory usage Current temperature of the processor chip Network status information includes at least the currently available physical bandwidth of the network. .

[0055] Furthermore, in order to conduct a unified quantitative assessment of the above-mentioned multi-dimensional states, the system calculates the comprehensive system stress index based on resource status information. The calculation formula is as follows:

[0056] in, These are preset system pressure control constant coefficients, used to adjust the contribution of load, memory, and temperature to the overall pressure index, respectively. A high-temperature warning threshold (e.g., 75°C) is set for the processor chip. The temperature limit for chip hardware frequency reduction protection (e.g., 105°C). It should be noted that the temperature item is only applicable to... Exceed It produces a positive contribution when the temperature is within the warning range, and the component is 0 when the warning temperature is not reached.

[0057] It should also be noted that the overall system pressure index It can comprehensively reflect the current computing load pressure, memory pressure, and thermal pressure of edge devices, providing a unified quantitative basis for subsequent partition threshold determination.

[0058] Furthermore, the system internally maintains resource quota vectors. , These correspond to the current available resource quota ratios for the real-time protection zone, the elastic processing zone, and the guaranteed processing zone, respectively, and their sum must be 1. During the initial system operation phase, Pre-set according to the standard initial quota.

[0059] Furthermore, at the beginning of each scheduling cycle, the scheduler calculates the current overall system pressure index. With the current available physical bandwidth of the network Within the given two-dimensional threshold range, a strict branch control procedure is implemented to quantitatively adjust the resource quota vector. The two-threshold control strategy used in this embodiment is as follows: (1) Routine health status: when <70 and When speeds reach ≥5Mbps, the system determines it is in a normal healthy state. The scheduler maintains the standard initial quota, i.e. =0.50、 =0.35、 =0.15. At this point, tasks at all levels are executed according to the normal quota. The real-time guarantee zone meets the computing power requirements of Level 1 and Level 2 tasks, the elastic processing zone supports the normal queuing processing of Level 3 tasks, and the minimum guarantee processing zone provides the minimum execution guarantee for Level 4 tasks.

[0060] (2) Resource shortage: When 70≤ <90 and When speeds reach ≥5Mbps, the system determines that local computing resources are under strain. The scheduler forcibly compresses the minimum processing area, releasing computing power to higher-priority tasks and adjusting the quota. =0.65、 =0.30、 =0.05. At this point, Level 1 tasks continue to execute at full local frequency, Level 2 tasks switch to frequency reduction processing, Level 3 tasks start batch processing with the same model, and Level 4 auxiliary tasks are completely suspended and moved to local cache.

[0061] (3) Weak network restricted state: when <90 and 0.2Mbps≤ When the speed is less than 5Mbps, the system determines that the network is currently in a limited state. The scheduler adjusts the quota to... =0.70、 =0.25、 =0.05. At this point, Level 1 and Level 2 tasks are executed locally on the edge and prioritize the use of real-time protection zone resources. Level 3 tasks are processed by reducing the frequency and skipping frames at large intervals. Level 4 tasks are switched to delayed execution mode and moved to the local cache, prohibiting all network upload overhead.

[0062] (4) Critical network outage state: When ≥90 or When the speed is less than 0.2 Mbps, the system determines that it is in an extreme critical or complete network outage state. The scheduler triggers the ultimate fallback quota and assigns it a value. =0.95、 =0.05、 =0.00. At this point, the system switches to local restricted fallback mode, the real-time protection zone has absolute computing power preemption right, forcibly cuts off the computing power supply of level 3 and level 4 tasks and puts them into the waiting queue for execution, and fully tilts the limited local computing power to level 1 critical security tasks, while the remaining non-critical video streams are all transferred to local non-volatile memory for raw disk storage.

[0063] It should be noted that the above The thresholds (70, 90) and bandwidth thresholds (5Mbps, 0.2Mbps) are configurable parameters and can be adjusted on-site according to the actual performance of edge devices and construction scenario requirements. Furthermore, the specific quota ratios for each partition can also be remotely configured through the management platform or adaptively corrected based on historical operating data. After resource quota allocation is completed, the system generates corresponding task execution strategies based on the current resource partitioning results and the mapping relationship between each task queue. The task execution strategies in this embodiment include immediate execution strategies, frequency reduction and merging execution strategies, and delayed execution and cache-supplemented execution strategies.

[0064] Furthermore, when the first or second level task queues in the real-time protection zone are idle, the system can temporarily schedule some tasks from the elastic processing zone or the backup processing zone to be executed in the real-time protection zone to improve computing power utilization. When a high-priority task suddenly arrives, the system immediately revokes the temporary authorization of the real-time protection zone and restores the original partition boundaries.

[0065] It should also be noted that the resource quota allocation results and task execution strategies are not fixed at once, but are dynamically recalculated and updated based on real-time resource status information in each scheduling cycle. Through the above dynamic partitioning and quota adjustment mechanism, edge devices can achieve differentiated anti-crowding resource guarantees under limited local computing power, ensuring that critical tasks always receive priority computing power support, while avoiding the continuous impact of low-priority tasks on the overall system response capability when resources are scarce.

[0066] S4. Match task execution strategies based on partitioned resource quotas, execute task execution strategies for video processing tasks in multi-level task queues, and output task processing results and feedback information. Furthermore, this step primarily generates and executes task execution strategies, performing corresponding edge-end inference scheduling, resource consumption control, and result output processing for video processing tasks of different priority levels. After task execution, it synchronously collects latency and resource consumption feedback information during task execution to provide operational feedback for dynamic optimization of the scheduling strategy in subsequent step S5. It should be noted that task execution in this step is not performed in a fixed sequence, but rather as a dynamic edge-end collaborative execution process based on multi-level task queues, resource partitioning results, and different task execution strategies.

[0067] Specifically, at the beginning of each scheduling cycle, the task execution control module in the edge device generates and executes a task execution strategy for the video processing tasks in the multi-level task queue. In this embodiment, the task execution strategy is as follows: (a) Immediate Execution Strategy This is triggered when the system determines the task to be in a normal healthy state. At this time, objects entering the first and second-level task queues receive sufficient computing power within the real-time protection zone, without performing any frame extraction or model downgrading. They are directly analyzed and inferred in real-time by the edge devices. For example, the collision risk detection task runs at full resolution and full frame rate during the basket lifting phase, ensuring millisecond-level response. The real-time execution strategy prioritizes the real-time requirements of first-level critical real-time tasks and second-level high-priority tasks.

[0068] (ii) Frequency reduction and merging execution strategies The frequency reduction and merging execution strategy mainly includes two sub-strategies, which are triggered when the system determines that the resource is under pressure or the network is weak and restricted.

[0069] (1) Frequency reduction execution: For the secondary high-priority task queue, the module controls it to perform adaptive frame skipping and frequency reduction processing, reducing the processing frequency to 50% of the original, so as to actively reduce the number of processing frames and reduce the computational load. For example, the safety helmet anomaly recognition task is adjusted from detection every frame to detection every two frames when resources are scarce, so as to release about half of the computing resources for use by the primary task while ensuring the basic recognition capability.

[0070] (2) Merged Execution: Primarily applied to three-level deferred task queues. The module retrieves multiple low-priority quality tasks with the same deep learning visual model requirements, similar construction stages, and adjacent time windows from the multi-level queues to be scheduled, and automatically merges them into the same batch for execution. By loading model weights once, multi-path, multi-frame joint inference is completed, eliminating the additional GPU memory and CPU scheduling overhead caused by frequent model loading. For example, in the plastering stage, the system can aggregate multiple insulation board laying status verification tasks for parallel inference, merging multiple model loadings into one, significantly reducing the average computing power consumption per unit task.

[0071] (III) Delayed execution and cached execution strategies The delayed execution and cache-based execution strategy mainly includes two sub-strategies, which are triggered when the system is determined to be in an extreme critical state or a backup state due to network outage, or when high-priority tasks continuously occupy computing power, causing level 3 and level 4 tasks to be unable to obtain execution resources for a long time.

[0072] (1) Delayed execution: The module switches the Level 3 quality task from immediate execution mode to delayed execution mode. That is, the task continues to be kept in the queue, but the execution cycle is lengthened (for example, from once every 500 milliseconds to once every 2 seconds). Thus, while ensuring the computing power of critical tasks, it can still provide limited processing opportunities for non-critical tasks.

[0073] (2) Cache-based Supplementary Execution: The module controls the complete switching of Level 3 quality tasks and Level 4 auxiliary processing tasks to the local cache management program, ceasing the allocation of any processor computing time slices to the aforementioned low-priority tasks, and first caches the relevant video data or intermediate inference structured results in the local non-volatile memory. The scheduler continuously monitors the system status until it receives a signal that the overall system pressure index has dropped to the healthy range or the network physical bandwidth has recovered. Then, the module reactivates the pending execution queue and uses the remaining computing power quota released from the backup processing area to perform background asynchronous supplementary inference execution on the cached tasks.

[0074] It should also be noted that the three execution strategies mentioned above are not mutually exclusive. The system can use different strategies in combination for task queues of different levels within the same scheduling cycle. For example, under resource pressure, level 1 tasks continue to be executed immediately, level 2 tasks are executed at a reduced frequency, level 3 tasks are executed in a merged manner, and level 4 tasks are executed with a delay or cache.

[0075] Furthermore, during task execution, the execution feedback information for each task is recorded in real time. In this embodiment, for each scheduled task, the recorded feedback information includes: (1) Task waiting delay The time difference between task generation and execution start time. This metric reflects the length of time a task waits in the queue and is used to evaluate the responsiveness of scheduling strategies.

[0076] (2) Task execution delay The time difference between the start and completion times of a task. This metric reflects the computation time of the task itself and is affected by the amount of computation, allocated computing resources, and the current system load.

[0077] (3) Timeout rate This is used to characterize the percentage of tasks that failed to complete execution before the preset deadline within the current scheduling period. Specifically, the system counts the number of tasks whose execution completion time exceeds the corresponding deadline among all executed tasks within the current scheduling period, and calculates the ratio with the total number of executed tasks in the current period to obtain the timeout rate.

[0078] (4) Task drop rate This parameter represents the percentage of tasks that are actively abandoned by the system during the current scheduling period due to reasons such as insufficient resources, timeouts, buffer overflows, or priority downgrades. Specifically, when the cumulative waiting time of a task in the waiting queue exceeds the maximum allowed waiting threshold, or when the system actively terminates a low-priority task under extreme critical conditions, the system marks the corresponding task as a discarded task and records the corresponding reason for discarding. The system obtains the task discard rate by calculating the ratio between the number of discarded tasks and the total number of generated tasks in the current period.

[0079] Furthermore, the above task execution records and feedback information are uniformly written into the local persistent storage area, which can be used for on-site event tracing and construction quality auditing, as well as as input data for the strategy update module in the subsequent step S5, providing an objective basis for the dynamic optimization of the scheduling strategy.

[0080] Through the above S4 steps, the system not only completed the actual execution and result output of video processing tasks in the multi-level task queue, but also simultaneously collected waiting latency, execution latency and resource usage information during task execution, forming a complete closed loop from scheduling decision to execution feedback, laying a solid data foundation for the adaptive update of the strategy in step S5.

[0081] S5. Construct a scheduling loss function based on feedback information, and perform closed-loop update and adjustment of parameters in the priority evaluation and resource quota allocation process according to the scheduling loss function.

[0082] Furthermore, during task execution, the system records execution feedback information for each video processing task in real time, and constructs a scheduling loss function based on this feedback information to quantitatively evaluate the effectiveness of the current edge scheduling strategy. This feedback information includes at least the task waiting latency. Task execution delay Timeout rate and task drop rate Specifically, task waiting latency is used to characterize the queuing time between task generation and start of execution; task execution latency is used to characterize the actual processing time between the start of execution and completion of execution; timeout rate is used to characterize the proportion of tasks that have not been completed before the preset deadline in the current scheduling cycle; and task drop rate is used to characterize the proportion of tasks that are actively abandoned by the system due to insufficient resources, waiting timeout, cache overflow, or priority downgrade.

[0083] Furthermore, based on the above feedback information, the system constructs a scheduling loss function to uniformly characterize the overall scheduling performance state within the current scheduling period. Its expression is:

[0084] in, Let represent the weight coefficients of the corresponding loss terms, and satisfy:

[0085] It should be noted that the weighting coefficients can be dynamically configured according to the needs of the construction scenario. When the system prioritizes ensuring real-time security tasks, the weights corresponding to the timeout rate loss item and the waiting latency loss item are increased; when the system focuses on the overall stability of the edge, the weights corresponding to the execution latency loss item and the task drop rate loss item are increased.

[0086] Furthermore, based on the scheduling loss function, the system employs the Adam adaptive optimization algorithm to dynamically update the task priority evaluation parameters and resource quota allocation parameters. These optimization parameters include at least the weights for task urgency, risk level, stage relevance, waiting compensation, and the resource quota ratios corresponding to the real-time guarantee zone, elastic processing zone, and guaranteed processing zone. At the end of each scheduling cycle, the system calculates the first and second moment estimates of each parameter based on the gradient changes of the current scheduling loss function, and adaptively adjusts the parameter update step size to achieve dynamic optimization and updating of the edge computing power scheduling strategy.

[0087] Furthermore, in some embodiments, when the system detects a continuous increase in the timeout rate, it automatically increases the real-time guarantee resource quota corresponding to high-priority tasks and increases the task urgency weight; when the system detects a continuous increase in the task drop rate, it increases the waiting compensation weight and the proportion of resources in the backup processing area; when the system detects a continuous increase in task execution latency, it increases the computing power cost suppression strength and reduces the scheduling priority of high resource consumption tasks, thereby reducing the risk of continuous high load operation of edge devices.

[0088] Through the above methods, this embodiment can construct a closed-loop adaptive scheduling optimization mechanism based on task execution feedback information, and use the Adam adaptive optimization algorithm to continuously and dynamically update the edge scheduling parameters, thereby improving the real-time performance, stability, and resource utilization efficiency of edge computing resource scheduling in complex construction environments.

[0089] Example 2: See Figure 2 As shown, this embodiment is an edge computing power optimization scheduling system for the construction of external wall insulation suspended scaffolding, including: The task generation and attribute extraction module is used to acquire video data and computing resource status information during the construction process of the suspended platform, generate video processing tasks based on the video data, and extract task attribute information from the video processing tasks. The priority assessment and grading module is used to assess the priority of video processing tasks based on task attribute information and the current construction stage to obtain task priorities; and to classify and enqueue each video processing task according to the task priority to establish a multi-level task queue. The resource quota module is used to partition available computing resources, obtain computing resource partitioning results, and perform partition mapping on multi-level task queues based on the computing resource partitioning results; and to allocate resource quotas based on computing resource status information to obtain partition resource quotas. The strategy execution module is used to match task execution strategies based on partition resource quotas, execute task execution strategies for video processing tasks in multi-level task queues, and output task processing results and feedback information. The feedback update module is used to construct a scheduling loss function based on feedback information, and to perform closed-loop updates and adjustments to the parameters in the priority evaluation and resource quota allocation process according to the scheduling loss function.

[0090] Example 3: This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0091] This computer device can be a server, and its internal structure diagram can be as follows: Figure 3As shown, the computer device includes a processor, memory, and network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores server data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements an edge computing power optimization scheduling method for the construction of external wall insulation suspended scaffolding.

[0092] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0093] Example 4: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0094] If the functions implemented by the method are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art or the current technical solution, can be embodied in the form of a software product. This current computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0095] 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-including 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.

[0096] 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.

[0097] The technical features of this invention not described can be implemented by or using existing technology, and will not be repeated here. Of course, the above description is not a limitation of this invention, and this invention is not limited to the examples above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of this invention should also be within the protection scope of this invention.

Claims

1. A method for edge-end computational power optimization scheduling in the construction of external wall insulation suspended scaffolding, characterized in that, Includes the following steps: S1. Obtain video data and computing resource status information during the construction process of the suspended platform, generate video processing tasks based on the video data, and extract task attribute information from the video processing tasks. S2. Based on task attribute information and the current construction stage, prioritize the video processing tasks to obtain task priorities; Based on task priority, each video processing task is tiered and queued to establish a multi-level task queue. S3. Divide the available computing resources into partitions to obtain the computing resource partitioning results, and perform partition mapping on the multi-level task queues based on the computing resource partitioning results; Based on the computing resource status information, resource quotas are allocated to the computing resource partitioning results to obtain the partition resource quotas. S4. Based on the partition resource quota, match the task execution strategy and execute the task execution strategy for video processing tasks in the multi-level task queue, and output the task processing results and feedback information. S5. Construct a scheduling loss function based on feedback information, and perform closed-loop update and adjustment of parameters in the priority evaluation and resource quota allocation process according to the scheduling loss function.

2. The edge computing power optimization scheduling method for external wall insulation suspended platform construction according to claim 1, characterized in that, The task attribute information includes at least the task type, generation time, deadline, risk level, associated construction stage, estimated computational load, and execution method requirements; the computing power resource status information includes edge device resource status information and network status information.

3. The edge computing power optimization scheduling method for external wall insulation suspended platform construction according to claim 2, characterized in that, The priority evaluation of the video processing task in step S2 includes: Based on task attribute information, the initial value of task importance, task urgency, risk level, stage-related quantitative value, and waiting compensation items can be quantitatively calculated or obtained. The task priority is obtained by weighted fusion calculation of the initial value of task importance, task urgency, risk level, stage correlation quantification value, and waiting compensation item; The multi-level task queue includes level 1 tasks, level 2 tasks, level 3 tasks, and level 4 tasks.

4. The edge computing power optimization scheduling method for external wall insulation suspended platform construction according to claim 3, characterized in that, The computing resource partitioning results include at least a real-time protection zone, an elastic processing zone, and a minimum protection processing zone; The partitioning and mapping of multi-level task queues based on computing resource partitioning results includes: Based on the task priority of the multi-level task queue, first-level and second-level tasks are mapped to the real-time protection zone, third-level tasks are mapped to the elastic processing zone, and fourth-level tasks are mapped to the backup processing zone.

5. The edge computing power optimization scheduling method for external wall insulation suspended platform construction according to claim 4, characterized in that, The step S3, which involves allocating resource quotas based on the computing resource partitioning results according to the computing resource status information, includes: Extract the processor load utilization, physical memory usage, and current processor chip temperature from the edge device resource status information, and extract the current available physical bandwidth of the network from the network status information; The system computing power index is obtained by weighting processor load utilization, physical memory usage and processor chip current temperature. Initialize the quotas for the real-time protection zone, elastic processing zone, and guaranteed processing zone, and adjust the quotas using a dual threshold method based on the current available physical bandwidth of the network and the system computing power index to obtain the partition resource quotas; The method of adjusting quotas using a dual threshold method includes: Set a first index threshold and a second index threshold for the system computing power index, and set a first bandwidth threshold and a second bandwidth threshold for the current available physical bandwidth of the network. When the system computing power index is less than the first index threshold and the current available physical bandwidth of the network is greater than the first bandwidth threshold, it is determined to be in a normal healthy state, and the initial quotas of the real-time protection zone, elastic processing zone and guaranteed processing zone are maintained. When the system computing power index is between the first index threshold and the second index threshold, and the current available physical bandwidth of the network is greater than the first bandwidth threshold, it is determined to be in a state of resource shortage. The quota of the real-time protection zone is increased to the first preset ratio, and the quota of the backup processing zone is reduced to the second preset ratio. When the system computing power index is less than the second index threshold and the current available physical bandwidth of the network is between the second bandwidth threshold and the first bandwidth threshold, it is determined to be in a weak network restricted state. The quota of the real-time protection zone is increased to the third preset ratio, and the quotas of the elastic processing zone and the guaranteed processing zone are reduced to the fourth preset ratio and the fifth preset ratio, respectively. When the system computing power index is greater than or equal to the second index threshold, or when the current available physical bandwidth of the network is less than the second bandwidth threshold, it is determined to be in a critical network outage state. The quota of the real-time protection zone will be adjusted to the preset maximum value, and the quota allocation of the backup processing zone will be suspended.

6. The edge computing power optimization scheduling method for external wall insulation suspended platform construction according to claim 5, characterized in that, The task execution strategies include immediate execution strategy, frequency reduction and merged execution strategy, and delayed execution and cache-supplemented execution strategy. The task execution strategy based on partitioned resource quota matching includes: When the task is determined to be in a normal healthy state, an immediate execution strategy is adopted for all Level 1, Level 2, Level 3 and Level 4 tasks. When a resource shortage is detected, an immediate execution strategy is adopted for Level 1 tasks, a frequency reduction and merging execution strategy is adopted for Level 2 and Level 3 tasks, and a delayed execution and cache-supplemented execution strategy is adopted for Level 4 tasks. When the network is determined to be in a weak and restricted state, an immediate execution strategy is adopted for Level 1 and Level 2 tasks, a frequency reduction and merging execution strategy is adopted for Level 3 tasks, and a delayed execution and cache-supplemented execution strategy is adopted for Level 4 tasks. When a critical network outage is detected, an immediate execution strategy is adopted for Level 1 tasks, while a delayed execution and cache-based execution strategy is adopted for Level 2, Level 3, and Level 4 tasks.

7. The edge computing power optimization scheduling method for external wall insulation suspended platform construction according to claim 6, characterized in that, The feedback information includes at least the task waiting delay, task execution delay, timeout rate, and task drop rate; The scheduling loss function is obtained by weighted fusion of task waiting delay, task execution delay, timeout rate and task drop rate; In step S5, the Adam adaptive optimization algorithm is used to perform closed-loop update and adjustment of parameters in the priority evaluation and resource quota allocation process based on the scheduling loss function.

8. A computational power optimization scheduling system for the construction of external wall insulation suspended scaffolding, characterized in that, The steps of the edge computing power optimization scheduling method for performing the construction of external wall insulation suspended scaffolding as described in any one of claims 1 to 7 include: The task generation and attribute extraction module is used to acquire video data and computing resource status information during the construction process of the suspended platform, generate video processing tasks based on the video data, and extract task attribute information from the video processing tasks. The priority assessment and grading module is used to assess the priority of video processing tasks based on task attribute information and the current construction stage to obtain task priorities; and to classify and enqueue each video processing task according to the task priority to establish a multi-level task queue. The resource quota module is used to partition available computing resources, obtain computing resource partitioning results, and perform partition mapping on multi-level task queues based on the computing resource partitioning results; and to allocate resource quotas based on computing resource status information to obtain partition resource quotas. The strategy execution module is used to match task execution strategies based on partition resource quotas, execute task execution strategies for video processing tasks in multi-level task queues, and output task processing results and feedback information. The feedback update module is used to construct a scheduling loss function based on feedback information, and to perform closed-loop updates and adjustments to the parameters in the priority evaluation and resource quota allocation process according to the scheduling loss function.

9. 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 edge computing power optimization scheduling method for the construction of external wall insulation suspended platform as described in any one of claims 1 to 7.

10. 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 edge computing power optimization scheduling method for the construction of the external wall insulation suspended platform as described in any one of claims 1 to 7.

Images