Smart park resource computing power scheduling optimization method and device, equipment and medium

Abstract

The invention provides a smart park resource computing power scheduling optimization method and device, equipment and a medium, and belongs to the technical field of resource scheduling. Determining a target computing power node group from the plurality of computing power node groups based on the computing power demand and the task priority of the computing power request task and the predicted load data of each computing power node; screening an optimal execution node based on the predicted load data, the task adaptation degree and the node health degree of each computing power node in the target computing power node group, and distributing the computing power request task to the optimal execution node for execution; and in response to detection that the energy consumption of the optimal execution node exceeds a preset energy consumption threshold value of the node type corresponding to the optimal execution node, or the task execution delay of the computing power request task is greater than a preset delay duration, migrating the computing power request task based on the task priority. The resource utilization rate can be effectively improved, and idle or excessive occupation of resources is avoided.

Description

Technical Field

[0001] This application belongs to the field of resource scheduling technology, and more specifically, it relates to a method, device, equipment, and medium for optimizing resource computing power scheduling in smart parks. Background Technology

[0002] With the deep integration of the digital economy and smart city construction, smart parks have become a new type of carrier integrating industrial development, technological innovation, and green and low-carbon development. As the construction of smart parks deepens, a large number of computing power demand scenarios have emerged in the parks, including data collection and analysis of IoT devices, real-time processing of security videos, and data computing of park operation and management platforms.

[0003] Currently, computing power scheduling in smart parks mostly adopts traditional static allocation or simple dynamic adjustment schemes based on the load of computing nodes. These computing power scheduling methods easily lead to some computing nodes being in a high-load state for a long time, while some nodes are idle, resulting in low utilization of computing resources. Summary of the Invention

[0004] The purpose of this application is to provide a method, device, equipment, and medium for optimizing resource computing power scheduling in smart parks, so as to improve the utilization rate of computing power resources and reduce the energy consumption of computing power nodes through dynamic task migration and computing power configuration optimization mechanisms.

[0005] A first aspect of this application provides a method for optimizing resource computing power scheduling in a smart park, comprising: Based on the operational status data and node attributes of each computing node in the smart park, the predicted load data of each computing node is determined. The operational status data includes computing power data and task execution data. The target computing node group is determined from multiple computing node groups based on the computing power requirements of the computing power request task, the task priority of the computing power request task, and the predicted load data of each computing node. The multiple computing node groups are pre-divided based on the node deployment location and node type. Based on the predicted load data, task suitability and node health of each computing node in the target computing power node group, the optimal execution node is selected and the computing power request task is assigned to the optimal execution node for execution. In response to the detection that the energy consumption of the optimal execution node exceeds the preset energy consumption threshold for the node type corresponding to the optimal execution node, or the task execution delay of the computing power request task is greater than the preset delay duration, the task migration assessment process is initiated and the computing power configuration of the smart park is updated, and the computing power request task is migrated based on the task priority.

[0006] A second aspect of this application provides a smart park resource computing power scheduling and optimization device, comprising: The data prediction module is used to determine the predicted load data of each computing node based on the operating status data and node attributes of each computing node in the smart park. The operating status data includes computing power data and task execution data. The first matching module is used to determine the target computing node group from multiple computing node groups based on the computing power requirements of the computing power request task, the task priority of the computing power request task, and the predicted load data of each computing node. The multiple computing node groups are pre-divided based on the node deployment location and node type. The second matching module is used to filter the optimal execution node based on the predicted load data, task suitability and node health of each computing node in the target computing power node group, and to allocate the computing power request task to the optimal execution node for execution. The optimization scheduling module is used to respond to the detection that the energy consumption of the optimal execution node exceeds the preset energy consumption threshold of the node type corresponding to the optimal execution node, or the task execution delay of the computing power request task is greater than the preset delay duration. In this case, the task migration evaluation process is initiated and the computing power configuration of the smart park is updated, and the computing power request task is migrated based on the task priority.

[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described smart park resource computing power scheduling optimization method.

[0008] In a fourth aspect of this application, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described smart park resource computing power scheduling optimization method.

[0009] The beneficial effects of the smart park resource computing power scheduling optimization method, device, equipment, and medium provided in this application embodiment are as follows: This application embodiment first determines the target computing node group by considering the computing power requirements of the computing power request task, the task priority of the computing power request task, and the predicted load data of each computing power node. Then, it selects the optimal execution node from the target computing power node group, achieving efficient and reasonable allocation of computing power resources, effectively improving resource utilization, and avoiding resource idleness or excessive occupation. In addition, this application embodiment also sets up an energy consumption and task execution delay monitoring mechanism. When the energy consumption of the optimal execution node exceeds the threshold or the task execution delay is too long, the task migration evaluation process is promptly initiated and the computing power configuration is updated, and migration is carried out according to task priority. This approach can not only effectively control node energy consumption and reduce operating costs, but also respond promptly to abnormal task execution situations, ensuring that tasks are completed on time, enhancing the stability and reliability of computing power scheduling in smart parks, and providing strong support for the stable operation of various business operations in the park. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating a smart park resource computing power scheduling optimization method provided in an embodiment of this application; Figure 2 A structural block diagram of a smart park resource computing power scheduling and optimization device provided in an embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0012] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0013] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0014] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a smart park resource computing power scheduling optimization method provided in an embodiment of this application. The method includes steps S101 to S104.

[0015] S101: Determine the predicted load data of each computing node based on the operating status data and node attributes of each computing node in the smart park. The operating status data includes computing power data and task execution data.

[0016] In this embodiment, the operational status data of each computing node in the smart park includes computing power data and task execution data. The computing power data is a set of quantitative indicators characterizing the hardware computing power performance output level and resource utilization of each computing node within the smart park. It directly reflects the current availability and load pressure of computing power resources of the nodes. The computing power data includes computing power performance indicators, computing power supply indicators, and computing power consumption indicators. Computing power performance indicators include, for example, the utilization rate of the Central Processing Unit (CPU), the peak utilization rate of the Graphics Processing Unit (GPU) floating-point operations, memory utilization, and network bandwidth utilization. Computing power supply indicators include, for example, the remaining available computing power value of the node, remaining memory capacity, remaining storage space, and number of idle threads, used to measure the node's capacity to handle new computing power tasks. Computing power consumption indicators include, for example, the computing power consumption value per unit time of the node and the duration of computing power resource occupation, used to assess the computing power consumption trend of the node under continuous operation.

[0017] Task execution data is a set of parameters recording the operational status and results of computing power nodes in a smart park throughout the entire lifecycle of a task as they accept and process computing power requests. Task execution data can include task access data, task processing data, and task output data. Task access data includes the number of tasks accepted by the node and the task submission timestamps. Task processing data includes the execution duration of each task, peak computing resource usage during task execution, task queue waiting time, and the number of tasks executed in parallel. Task output data includes task execution success rate, task execution error rate, and latency fluctuation values ​​of task output results, used to evaluate the stability and reliability of node task processing.

[0018] In one embodiment, the predicted load data for each computing node is determined based on the operational status data and node attributes of each computing node in the smart park, including: After aligning the computing power data and task execution data according to the time series, a multi-dimensional feature vector is constructed. Time series analysis is then performed on the multi-dimensional feature vector to obtain the first predicted load data. The second predicted load data is obtained by inputting the multidimensional feature vector and node attributes into the regression model. The first and second predicted load data are weighted and fused to obtain the predicted load data.

[0019] In this embodiment, because there is a strong correlation between computing power data and task execution data, such as a surge in the number of tasks leading to increased CPU utilization, aligning computing power data and task execution data according to a time series can ensure that computing power data and task execution data at the same point in time are related, avoiding the break in the causal relationship between features and load due to time misalignment.

[0020] First, the time granularity and time window are determined based on the load fluctuation frequency of the computing nodes and the predicted demand. For example, assuming the load data for the next 10 minutes is predicted, computing power data and task execution data with a time granularity of 1 minute and a time window of the past 24 hours are selected as samples.

[0021] Secondly, the original collection timestamps of computing power data and task execution data are extracted separately, and all timestamps are converted to the same time format, i.e., standard timestamps. The timestamps of the two types of data are then calibrated. If there is a collection time deviation (for example, the collection time of computing power data is t, and the collection time of task execution data is t+2s), the deviation data is corrected to the standard time point with a preset time granularity through linear interpolation.

[0022] Finally, using the standard timestamp as an index, the computing power data and task execution data are traversed, and the sub-indicators of computing power data and task execution data under the same standard timestamp are associated to form a three-dimensional data set of "timestamp - computing power data - task execution data". Based on the associated three-dimensional data set, the sub-indicators of computing power data and task execution data under each timestamp are concatenated according to the preset feature dimension order to form a one-dimensional vector. All one-dimensional vectors within a continuous time window are arranged in chronological order to finally obtain a multi-dimensional feature vector.

[0023] In this embodiment, after obtaining the multidimensional feature vector, time series analysis needs to be performed on the multidimensional feature vector to obtain the first predicted load data. The time series analysis of the multidimensional feature vector can be performed using an Autoregressive Moving Average (ARMA) model, a Long Short-Term Memory (LSTM) network, or a Gated Recurrent Unit (GRU).

[0024] If LSTM is used for time series analysis, the process of obtaining the first predicted load data by inputting the multidimensional feature vector into LSTM is as follows: LSTM includes memory units, forget gates, input gates, and output gates. The aligned multidimensional feature vectors are split into training and test sets according to time steps, and the data is normalized (scaled to 0-1) and then converted into a three-dimensional input format adapted to LSTM [number of samples, time steps, feature dimensions].

[0025] After the feature vectors are input into the LSTM at each time step, the forget gate filters out redundant information in the current input that is irrelevant to the historical load based on the trained weights; the input gate filters out information valuable for load prediction from the current task features and updates the model's memory units; the output gate outputs the intermediate prediction value for the current time step based on the updated memory units (which retain the load time sequence pattern learned during training). This intermediate prediction value is mapped to the actual load value through a fully connected layer, and then denormalized to finally obtain the first predicted load data.

[0026] In this embodiment, the multidimensional feature vector and node attributes also need to be input into the regression model to obtain the second predicted load data. The node attributes include the node's location within the park, node type, and adapted task type. For example, the node type can be a general-purpose computing node, a dedicated AI computing node, an edge computing node, or a cloud server node. The adapted task type can be a security monitoring data analysis task, a park power consumption analysis task, etc.

[0027] A regression model is a mathematical model based on supervised learning, used to establish a mapping relationship between independent and dependent variables. By learning the correlation between features in historical data and the target, it enables continuous value prediction for new samples. The regression model in this embodiment can be a linear regression model.

[0028] The input features are obtained by concatenating multidimensional feature vectors and node attributes. These input features are then standardized and normalized to eliminate dimensional differences. The normalized data is then input into a regression model, which uses the trained model's feature weights and bias terms to obtain preliminary predictions. These preliminary predictions are then denormalized and destandardized to obtain the final second prediction load data.

[0029] In this embodiment, the first predicted load data and the second predicted load data are multiplied by their respective weighting coefficients and then added together to obtain the predicted load data. The weighting coefficients corresponding to the first predicted load data and the second predicted load data can be set empirically, for example, both are 0.5.

[0030] S102: Based on the computing power requirements of the computing power request task, the task priority of the computing power request task, and the predicted load data of each computing power node, the target computing power node group is determined from multiple computing power node groups. The multiple computing power node groups are pre-divided based on the node deployment location and node type.

[0031] In this embodiment, multiple computing node groups are pre-divided based on node deployment location and node type. For example, based on the basic information of all computing nodes in the smart park, the hierarchical features of node deployment location are extracted, and the park is divided into primary location groups according to geographical regions, such as core data center areas, building edge areas, and workshop terminal areas. Secondary location subgroups are then formed according to the network topology hierarchy (such as core layer, aggregation layer, and access layer) within each area. Simultaneously, nodes are classified according to their functional attributes, into general computing nodes, dedicated AI computing nodes, edge computing nodes, and cloud service layer nodes. Then, a two-dimensional cross-division strategy of "location dimension and type dimension" is used to group computing nodes of the same type within the same geographical area into one computing node group. Finally, a unique identifier and group attribute tag are added to each node group, completing the pre-division of computing node groups and forming a node group architecture with clear hierarchy and defined functions.

[0032] In one embodiment, determining a target computing node group from multiple computing node groups based on the computing power requirement of the computing power request task, the task priority of the computing power request task, and the predicted load data of each computing node includes: Calculate the first matching degree between each computing node group and the computing power request task, which includes computing power requirements and task priority; The computing power node group with a first matching degree greater than the preset first matching degree threshold is selected as the candidate node group; The average predicted load rate of each candidate node group is calculated based on the predicted load data of each computing node, and the average predicted load rate is converted into a load evaluation value. The comprehensive score of each candidate node group is calculated based on the first matching degree and the load evaluation value, and the candidate node group with the highest comprehensive score is selected as the target computing power node group.

[0033] In this embodiment, the computing power request task includes computing power requirements, task priority, task type, task execution time limit, task data volume, and hardware architecture adapted to the task. The computing power requirements represent the hardware resource configuration requirements necessary to complete the computing power request task, including: peak CPU computing power requirements, GPU floating-point computing power requirements, memory capacity requirements, network bandwidth requirements, and computing power resource occupancy time requirements. The task priority represents a preset task priority ranking level determined based on the business importance and urgency of the task.

[0034] This embodiment selects two dimensions: computing power requirement and task priority, to match each computing node group with the computing power request task, and calculates the first matching degree. The weight parameters corresponding to the computing power requirement are... Weight parameters corresponding to task priority It can be determined based on experience. =1. Extract the core computing power parameters of the computing node group, such as the total peak computing power. Extract the computing power requirement parameters of the computing power request task, such as the task's computing power requirement threshold. Then the computing power requirement matching score The value can take two cases: like ≥ This indicates that the computing node group meets the task requirements. =1.

[0035] like < This indicates that the computing power of the computing node group is insufficient, and the score is calculated based on the computing power ratio. = If the ratio is less than 0.5, it is considered a mismatch, and the value is taken as... =0.

[0036] In this embodiment, the task priorities of the computing power node group include high priority, medium priority, and low priority, which can be represented by 3, 2, and 1, respectively. If the task priority of the computing power node group is greater than or equal to the task priority of the computing power request task, then the task priority matching score is obtained. If the task priority of the computing power node group is lower than the task priority of the computing power request task, the score is calculated in descending order of the level difference. For example, if the level difference is 1, then... 0.5; grade difference 2, then 0.

[0037] In this embodiment, the computing power demand is matched with a score. Score matching task priority The weighted summation is used to obtain the final first matching degree. The first matching degree of all computing power node groups is calculated, and all computing power node groups that satisfy the first matching degree greater than the preset first matching degree threshold are selected as candidate node groups.

[0038] In this embodiment, the predicted load data of each computing node within the candidate node group, such as the predicted CPU utilization value, is extracted. The predicted load data is compared with the total computing power of the nodes to obtain the predicted load rate of a single computing node. The arithmetic mean of the predicted load rates of all computing nodes in the group is then calculated to obtain the average predicted load rate of the candidate node group. Further, to obtain the load evaluation value, the average predicted load rate of the candidate node group needs to be normalized, as shown in the following formula:

[0039] Since a higher average predicted load rate indicates less resource reserve in the node group, the normalized load rate needs to be reverse-processed to obtain the load assessment value. That is... .

[0040] In one embodiment, a comprehensive score for each candidate node group is calculated based on a first matching degree and a load assessment value, including: Based on the first matching degree quantile distribution of all candidate node groups, the first weight corresponding to the first matching degree and the second weight corresponding to the load evaluation value are dynamically configured. The sum of the first weight and the second weight is 1, and the weight ratio is adaptively adjusted according to the quantile interval of the first matching degree. The core score is obtained based on the coupling coefficient, first matching degree, and load evaluation value. The core score is then corrected based on the task scenario penalty term to obtain the comprehensive score of each candidate node group.

[0041] In this embodiment, the quantile (e.g., the 90th quantile) of the first matching degree of all candidate node groups is calculated. 50th percentile ).

[0042] If the first matching degree of the candidate node group ≥ Matching weight Load assessment value weight Prioritize matching degree, where i represents the i-th candidate node group.

[0043] like Matching weight .

[0044] like Matching weight Load assessment value weight Select the computing node group with better load assessment value, where A and B are constants less than 1.

[0045] Coupling coefficient This is used to represent the synergistic effect between the match degree and the load assessment value, avoiding extreme cases such as "high match degree but excessive load" or "low load but insufficient match degree". Expressed as a formula:

[0046] in, This represents the load assessment value.

[0047] Core score The calculation formula is:

[0048] Final overall score The calculation formula is:

[0049] The method for calculating the comprehensive score in this embodiment differs from the fixed weighted summation method in the prior art. It calculates the comprehensive score of each candidate node group by dynamically coupling the first matching degree with the load evaluation value and introducing a task scenario penalty term, which is more in line with the dynamic and differentiated needs of computing power scheduling in smart parks.

[0050] S103: Based on the predicted load data, task suitability and node health of each computing node in the target computing power node group, the optimal execution node is selected and the computing power request task is assigned to the optimal execution node for execution.

[0051] In one embodiment, selecting the optimal execution node based on the predicted load data, task suitability, and node health of each computing node within the target computing power node group includes: The predicted load rate of each computing node is calculated based on the predicted load data of each computing node in the target computing node group, and the predicted load rate of each computing node is converted into the load evaluation value of each computing node. The target score of each computing node is obtained by weighted summing of the load assessment value, task adaptability and node health of each computing node, and the computing node with the highest target score is selected as the optimal execution node.

[0052] In this embodiment, each candidate node group contains multiple computing power nodes, so it is necessary to determine the optimal execution node in order to allocate the computing power request task to the optimal execution node for execution.

[0053] First, task suitability represents the degree to which the hardware characteristics and functional configuration of a single computing node within the target computing node group match the task requirements of the computing power request task. It is used to measure whether the node can complete the task efficiently, and its value range is usually [0,1]. Node health represents a quantitative evaluation index of the stability and reliability of the operating status of a single computing node within the target computing node group. It reflects the probability that the node will operate without failure during the task execution cycle, and its value range is usually [0,1].

[0054] In this embodiment, the predicted load rate of each computing node is calculated based on the predicted load data of each computing node in the target computing node group. The predicted load rate of each computing node is then converted into the load evaluation value of each computing node. The specific processing steps have been described above and will not be repeated here.

[0055] Secondly, the target score of each computing node is obtained by weighted summation of the load assessment value, task adaptability and node health of each computing node, and the computing node with the highest target score is selected as the optimal execution node.

[0056] The reason for determining the target computing power node group before selecting the optimal execution node in this embodiment is to reduce the scheduling computational complexity of the equipment and improve the decision-making efficiency of the equipment. Because there are a large number of computing power nodes in a smart park, directly traversing all nodes to calculate comprehensive indicators would generate a lot of invalid calculations. First, the target computing power node group is selected by "computing power requirement + task priority + load", which can significantly narrow down the range of candidate nodes. Then, the optimal node is selected within a small range based on task suitability and node health, which can significantly reduce the scheduling computational complexity of the equipment and improve the decision-making efficiency of the equipment.

[0057] S104: In response to the detection that the energy consumption of the optimal execution node exceeds the preset energy consumption threshold of the node type corresponding to the optimal execution node, or the task execution delay of the computing power request task is greater than the preset delay duration, the task migration evaluation process is initiated and the computing power configuration of the smart park is updated, and the computing power request task is migrated based on the task priority.

[0058] In this embodiment, when the energy consumption of the optimal execution node exceeds a preset energy consumption threshold, it means that the node hardware is operating under high load. Prolonged overload can lead to shortened hardware lifespan, increased failure rate, and even node crashes. When the task execution delay exceeds a preset duration, it indicates that the node's current resource supply capacity is insufficient to meet the task requirements, and continued operation will result in task timeout failure. Therefore, when the device detects any of the above situations, it needs to initiate a task migration assessment process and migrate the computing power request task to other computing power nodes.

[0059] In one embodiment, when node energy consumption exceeds limits or task latency exceeds limits, it essentially indicates that the matching relationship between computing power request tasks and computing power nodes is no longer optimal. The computing power configuration of the park may have local resource bottlenecks, such as an imbalance in the load distribution of candidate node groups. By updating the computing power configuration of the smart park, such as adjusting node group load thresholds, optimizing resource scheduling strategies between nodes, and supplementing resources for scarce node types, the resource mismatch problem can be fundamentally solved, achieving global dynamic balance of computing power resources and improving the overall computing power scheduling efficiency and resource utilization of the park.

[0060] Tasks with different priorities have varying requirements for execution stability and latency. Migrating based on task priority ensures resource availability for high-priority tasks. The task migration assessment process first requires identifying the computing power request tasks to be migrated and determining their priorities, then confirming the migration destination. For example, high-priority tasks can be migrated to low-load, high-priority nodes within the same target computing power node group to ensure core business operations are unaffected; while low-priority tasks can be migrated opportunistically or across groups to other computing power nodes depending on resource availability. This differentiated migration strategy avoids a decline in the service quality of core tasks due to indiscriminate migration.

[0061] As can be seen from the above, this embodiment first determines the target computing node group by considering the computing power requirements of the computing power request task, the task priority of the computing power request task, and the predicted load data of each computing power node. Then, it selects the optimal execution node from the target computing power node group, achieving efficient and reasonable allocation of computing power resources, effectively improving resource utilization, and avoiding resource idleness or excessive occupation. In addition, this embodiment also sets up an energy consumption and task execution delay monitoring mechanism. When the energy consumption of the optimal execution node exceeds the threshold or the task execution delay is too long, the task migration evaluation process is promptly initiated and the computing power configuration is updated, and migration is carried out according to task priority. This approach can not only effectively control node energy consumption and reduce operating costs, but also respond promptly to abnormal task execution situations, ensuring that tasks are completed on time, enhancing the stability and reliability of computing power scheduling in smart parks, and providing strong support for the stable operation of various business operations in the park.

[0062] In one embodiment of this application, the smart park resource computing power scheduling optimization method further includes: The second degree of matching between the node type of the computing power node and the computing power request task is calculated; Determine the success rate of computing nodes executing historical tasks of the same type as the computing power request task; The task suitability of computing nodes is determined based on the second matching degree and the execution success rate.

[0063] In this embodiment, the node type can be a general-purpose computing node, a dedicated AI computing node, an edge computing node, or a cloud server node, etc. Different node types are adapted to different task types, and a second matching degree is calculated using preset type matching rules. For example, the matching coefficient between a dedicated AI node and an AI inference task is 1, while the matching coefficient between a general-purpose node and an AI inference task is 0.6. The second matching degree reflects the basic capability of the computing node to "efficiently handle tasks."

[0064] Historical execution success rate is the percentage of times a computing node successfully completes a task without failure when executing historical tasks of the same type as the current task, out of the total number of executions. A high success rate for a computing node executing similar historical tasks indicates that it is better suited for that type of task in terms of hardware performance, software compatibility, and operational status; conversely, a low success rate suggests potential execution risks. This metric reflects the computing node's practical ability to stably complete tasks.

[0065] After determining the second matching degree and historical execution success rate, this embodiment calculates the final task adaptability based on these two indicators using a preset weight allocation strategy. This can quantify the degree of fit between computing power nodes and tasks from two dimensions: the adaptability of inherent node attributes and the reliability of historical execution, effectively avoiding the execution risks caused by matching based solely on hardware parameters.

[0066] In one embodiment of this application, initiating the task migration assessment process includes: If the energy consumption of the optimal execution node exceeds the preset energy consumption threshold for the node type corresponding to the optimal execution node, the energy consumption of the optimal execution node is determined to be abnormal. Calculate the percentage difference between the current energy consumption of the optimal execution node and the preset energy consumption threshold for the corresponding node type. Determine the anomaly impact level based on the comparison result between the percentage difference and the first preset threshold. Determine candidate migration nodes based on the anomaly impact level. If the task execution delay of the computing power request task is greater than the preset delay time, the task execution of the optimal execution node is determined to be abnormal. The percentage difference between the task execution delay of the computing power request task and the preset delay duration is used to determine the anomaly impact level based on the comparison result of the percentage difference with the second preset threshold, and candidate migration nodes are determined based on the anomaly impact level.

[0067] In this embodiment, if the energy consumption of the optimal execution node is abnormally high, the percentage difference between the current energy consumption of the optimal execution node and the preset energy consumption threshold for the corresponding node type is calculated. The impact level of the anomaly is determined based on the comparison between this percentage difference and a first preset threshold. The preset energy consumption threshold can be set based on historical data. If the percentage difference is greater than or equal to the first preset threshold, the impact level is determined to be high, and candidate migration nodes must be selected simultaneously from both within and across the target computing power node group. If the percentage difference is less than the first preset threshold, candidate nodes within the target computing power node group are prioritized as candidate migration nodes.

[0068] In this embodiment, if the task execution of the optimal execution node is abnormal, the percentage difference between the task execution delay of the computing power request task and a preset delay duration is calculated. The impact level of the anomaly is determined based on a comparison of this percentage difference with a second preset threshold. The preset delay duration can be set based on historical data. If the percentage difference is greater than or equal to the second preset threshold, the impact level is determined to be high, and candidate migration nodes must be selected simultaneously from both within and across the target computing power node group. If the percentage difference is less than the second preset threshold, candidate nodes within the target computing power node group are prioritized as candidate migration nodes.

[0069] In summary, this embodiment determines the impact level of anomalies by calculating the percentage difference between energy consumption exceedance and task execution anomalies and comparing it with a preset threshold. Based on different impact levels, candidate migration nodes are flexibly selected. For high impact levels, both intra-group and cross-group nodes are considered to broaden the selection range and quickly find suitable nodes; for low impact levels, intra-group selection is prioritized to improve migration efficiency. This differentiated processing effectively ensures the rationality and timeliness of task migration and enhances the flexibility and stability of computing power scheduling in smart parks.

[0070] In one embodiment of this application, updating the computing power configuration of a smart park includes: For the optimal execution node where energy consumption exceeds a preset energy consumption threshold or task execution latency exceeds a preset latency duration: The adjustment amount of the computing power task assignment quota is determined based on the level of abnormal impact, and the priority of the optimal execution node in the allocation of similar tasks is reduced by a preset step size. The computing power configuration of each computing power node in the smart park is updated based on the optimal execution node after the computing power adjustment.

[0071] In this embodiment, assume that the general computing node B in the smart park is a core data center general node with a preset energy consumption threshold of 1200W and a preset task execution delay of 1 minute. The initial priority of similar data processing tasks is assigned as 4 (medium priority). If the real-time energy consumption of node B is detected to be 1500W, it is determined that the threshold has been exceeded, and the impact level is determined to be medium. At this time, the corresponding task load capacity of this computing node is reduced by 20%, and the priority of similar data processing tasks is adjusted to 2 (low priority). Based on the adjusted computing power of the general computing node B, the computing power configuration of each computing node in the smart park is updated.

[0072] This embodiment updates the computing power configuration of the entire park by adjusting the configuration of abnormal computing power nodes, thereby achieving dynamic optimization of computing power resources. This not only alleviates the pressure on abnormal nodes in a timely manner, but also ensures the efficiency and stability of the overall computing power scheduling of the park, and improves resource utilization efficiency.

[0073] Corresponding to the smart park resource computing power scheduling optimization method in the above embodiment, Figure 2 This is a structural block diagram of a smart park resource computing power scheduling and optimization device provided in one embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The smart park resource computing power scheduling and optimization device 20 includes: a data prediction module 21, a first matching module 22, a second matching module 23, and an optimization scheduling module 24.

[0074] Among them, the data prediction module 21 is used to determine the predicted load data of each computing power node based on the operation status data and node attributes of each computing power node in the smart park. The operation status data includes computing power data and task execution data. The first matching module 22 is used to determine the target computing node group from multiple computing node groups based on the computing power requirements of the computing power request task, the task priority of the computing power request task, and the predicted load data of each computing node. The multiple computing node groups are pre-divided based on the node deployment location and node type. The second matching module 23 is used to filter the optimal execution node based on the predicted load data, task adaptability and node health of each computing node in the target computing node group, and to allocate the computing power request task to the optimal execution node for execution. The optimization scheduling module 24 is used to respond to the detection that the energy consumption of the optimal execution node exceeds the preset energy consumption threshold of the node type corresponding to the optimal execution node, or the task execution delay of the computing power request task is greater than the preset delay duration, by starting the task migration evaluation process and updating the computing power configuration of the smart park, and migrating the computing power request task based on the task priority.

[0075] In one embodiment of this application, when the data prediction module 21 determines the predicted load data of each computing node based on the operating status data and node attributes of each computing node in the smart park, it is specifically used for: After aligning the computing power data and task execution data according to the time series, a multi-dimensional feature vector is constructed. Time series analysis is then performed on the multi-dimensional feature vector to obtain the first predicted load data. The second predicted load data is obtained by inputting the multidimensional feature vector and node attributes into the regression model. The first and second predicted load data are weighted and fused to obtain the predicted load data.

[0076] In one embodiment of this application, when the first matching module 22 determines the target computing node group from multiple computing node groups based on the computing power requirement of the computing power request task, the task priority of the computing power request task, and the predicted load data of each computing node, it is specifically used for: Calculate the first matching degree between each computing node group and the computing power request task, which includes computing power requirements and task priority; The computing power node group with a first matching degree greater than the preset first matching degree threshold is selected as the candidate node group; The average predicted load rate of each candidate node group is calculated based on the predicted load data of each computing node, and the average predicted load rate is converted into a load evaluation value. The comprehensive score of each candidate node group is calculated based on the first matching degree and the load evaluation value, and the candidate node group with the highest comprehensive score is selected as the target computing power node group.

[0077] In one embodiment of this application, the second matching module 23, when selecting the optimal execution node based on the predicted load data, task suitability, and node health of each computing node within the target computing power node group, is specifically used for: The predicted load rate of each computing node is calculated based on the predicted load data of each computing node in the target computing node group, and the predicted load rate of each computing node is converted into the load evaluation value of each computing node. The target score of each computing node is obtained by weighted summing of the load assessment value, task adaptability and node health of each computing node, and the computing node with the highest target score is selected as the optimal execution node.

[0078] In one embodiment of this application, the smart park resource computing power scheduling and optimization device 20 further includes a first calculation module for calculating the second matching degree between the node type of the computing power node and the computing power request task; Determine the success rate of computing nodes executing historical tasks of the same type as the computing power request task; The task suitability of computing nodes is determined based on the second matching degree and the execution success rate.

[0079] In one embodiment of this application, the optimized scheduling module 24 is specifically used for: If the energy consumption of the optimal execution node exceeds the preset energy consumption threshold for the node type corresponding to the optimal execution node, the energy consumption of the optimal execution node is determined to be abnormal. Calculate the percentage difference between the current energy consumption of the optimal execution node and the preset energy consumption threshold for the corresponding node type. Determine the anomaly impact level based on the comparison result between the percentage difference and the first preset threshold. Determine candidate migration nodes based on the anomaly impact level. If the task execution delay of the computing power request task is greater than the preset delay time, the task execution of the optimal execution node is determined to be abnormal. The percentage difference between the task execution delay of the computing power request task and the preset delay duration is used to determine the anomaly impact level based on the comparison result of the percentage difference with the second preset threshold, and candidate migration nodes are determined based on the anomaly impact level.

[0080] In one embodiment of this application, the optimized scheduling module 24 is specifically used for: For the optimal execution node where energy consumption exceeds a preset energy consumption threshold or task execution latency exceeds a preset latency duration: The adjustment amount of the computing power task assignment quota is determined based on the level of abnormal impact, and the priority of the optimal execution node in the allocation of similar tasks is reduced by a preset step size. The computing power configuration of each computing power node in the smart park is updated based on the optimal execution node after the computing power adjustment.

[0081] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of each module / unit in the above-described device embodiments, for example... Figure 2 The functions of the data prediction module 21, the first matching module 22, the second matching module 23, and the optimization scheduling module 24 are shown.

[0082] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), but it may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0083] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0084] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory.

[0085] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation method described in the smart park resource computing power scheduling optimization method provided in the embodiments of this application, or they can execute the implementation method of the electronic device described in the embodiments of this application, which will not be repeated here.

[0086] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0087] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0088] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0089] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0090] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or modules, or it may be an electrical, mechanical, or other form of connection.

[0091] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0092] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for optimizing resource computing power scheduling in a smart park, characterized in that, The method comprises the following steps: determining predicted load data of each computing power node based on operation state data and node attributes of each computing power node in the smart park, wherein the operation state data comprises computing power data and task execution data; determining a target computing power node group from a plurality of computing power node groups based on computing power requirements of a computing power request task, task priorities of the computing power request task, and the predicted load data of each computing power node, wherein the plurality of computing power node groups are pre-divided based on node deployment locations and node types; selecting an optimal execution node based on the predicted load data, task adaptation degree, and node health degree of each computing power node in the target computing power node group, and assigning the computing power request task to the optimal execution node for execution; in response to detecting that the energy consumption of the optimal execution node exceeds a preset energy consumption threshold of the node type corresponding to the optimal execution node, or the task execution delay of the computing power request task is greater than a preset delay duration, starting a task migration evaluation process and updating the computing power configuration of the smart park, and migrating the computing power request task based on the task priority. 2.The method of claim 1, wherein, The method of determining the predicted load data of each computing power node based on the operation state data and the node attributes of each computing power node in the smart park comprises the following steps: aligning the computing power data and the task execution data in time sequence to construct a multi-dimensional feature vector, and performing time series analysis on the multi-dimensional feature vector to obtain first predicted load data; inputting the multi-dimensional feature vector and the node attributes into a regression model to obtain second predicted load data; weighting and fusing the first predicted load data and the second predicted load data to obtain predicted load data. 3.The method of claim 1, wherein, The method of determining a target computing power node group from a plurality of computing power node groups based on computing power requirements of a computing power request task, task priorities of the computing power request task, and predicted load data of each computing power node comprises the following steps: calculating a first matching degree of each computing power node group and the computing power request task, wherein the computing power request task comprises computing power requirements and task priorities; selecting a computing power node group with a first matching degree greater than a preset first matching degree threshold as a candidate node group; calculating an average predicted load rate of each candidate node group based on the predicted load data of each computing power node, and converting the average predicted load rate into a load evaluation value; calculating a comprehensive score of each candidate node group based on the first matching degree and the load evaluation value, and selecting a candidate node group with the highest comprehensive score as the target computing power node group. 4.The method of claim 1, wherein, The method of selecting an optimal execution node based on the predicted load data, task adaptation degree, and node health degree of each computing power node in the target computing power node group comprises the following steps: calculating a predicted load rate of each computing power node based on the predicted load data of each computing power node in the target computing power node group, and converting the predicted load rate of each computing power node into a load evaluation value of each computing power node; weighting and summing the load evaluation value, task adaptation degree, and node health degree of each computing power node to obtain a target score of each computing power node, and selecting a computing power node with the highest target score as the optimal execution node. 5.The method of claim 1, wherein, The method further comprises the following steps: calculating a second matching degree between the node type of the computing power node and the computing power request task; a success rate of execution of the computing power node when performing a historical task of the same type as the task of the computing power request; determining a task adaptation degree of the computing power node based on the second matching degree and the success rate of execution. 6.The method of claim 5, wherein, the starting of the task migration evaluation process comprises: if the energy consumption of the optimal execution node exceeds a preset energy consumption threshold of the node type corresponding to the optimal execution node, it is determined that the energy consumption of the optimal execution node is abnormal; calculating a difference ratio of the current energy consumption of the optimal execution node and the preset energy consumption threshold of the corresponding node type, determining an abnormal influence level based on a comparison result of the difference ratio and a first preset threshold, and determining a candidate migration node based on the abnormal influence level; if the task execution delay of the computing power request task is greater than a preset delay duration, it is determined that the task execution of the optimal execution node is abnormal; calculating a difference ratio of the task execution delay of the computing power request task and the preset delay duration, determining an abnormal influence level based on a comparison result of the difference ratio and a second preset threshold, and determining a candidate migration node based on the abnormal influence level.

7. The smart park resource computing power scheduling optimization method of claim 6, wherein, the updating of the computing power configuration of the smart park comprises: for the optimal execution node whose energy consumption exceeds the preset energy consumption threshold or whose task execution delay is greater than the preset delay duration: determining an adjustment amount of the computing power task undertaking quota based on the abnormal influence level, and simultaneously reducing the priority of the optimal execution node in the same task allocation by a preset step size; updating the computing power configuration of each computing power node of the smart park based on the optimal execution node after the adjustment of the computing power.

8. A smart park resource computing power scheduling optimization device, characterized in that, comprise: a data prediction module configured to determine predicted load data of each computing power node based on operating state data and node attributes of each computing power node of the smart park, the operating state data comprising computing power data and task execution data; a first matching module configured to determine a target computing power node group from a plurality of computing power node groups based on the computing power demand of the computing power request task, the task priority of the computing power request task, and the predicted load data of each computing power node, the plurality of computing power node groups being pre-divided based on node deployment locations and node types; a second matching module configured to filter an optimal execution node based on the predicted load data, the task adaptation degree, and the node health degree of each computing power node in the target computing power node group, and to allocate the computing power request task to the optimal execution node for execution; an optimization scheduling module configured to, in response to detecting that the energy consumption of the optimal execution node exceeds a preset energy consumption threshold of the node type corresponding to the optimal execution node, or that the task execution delay of the computing power request task is greater than a preset delay duration, start a task migration evaluation process and update the computing power configuration of the smart park, and migrate the computing power request task based on the task priority.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 7.

Images