An intelligent management system for a big data center based on edge computing
By deploying lightweight AI models at the edge of big data centers for real-time data analysis and processing, combined with cloud-based collaborative management, the high latency and high energy consumption problems of traditional centralized cloud management are solved, achieving low-latency, efficient, and reliable data center management.
Patent Information
- Application Number
- CN202611435732.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-16
- Publication Date
- 2026-11-06
AI Technical Summary
Traditional centralized cloud-managed big data centers suffer from problems such as high bandwidth consumption, high transmission latency, slow response, high energy consumption, uneven resource utilization, and lack of edge autonomy, making it difficult to meet the requirements of real-time performance, energy efficiency, and security.
An edge computing-based intelligent management system is adopted. By deploying lightweight AI models at the edge for real-time data collection and analysis, combined with collaborative processing on the cloud platform, millisecond-level latency for fault response and scheduling is achieved, bandwidth requirements are reduced, and global load prediction and cooling collaborative optimization are performed.
It achieves low-latency, high-real-time fault response and scheduling, reduces bandwidth and energy consumption, improves resource utilization, enhances system stability and reliability, supports multi-protocol device access, and is suitable for distributed big data center scenarios.
Smart Images
Figure CN123055587A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of big data center management technology, specifically a big data center intelligent management system based on edge computing. Background Technology
[0002] With the rapid growth of the digital economy and the demand for computing power, big data centers are developing towards high density, high power consumption, and high concurrency. Traditional centralized cloud management and control faces the following problems: Massive amounts of device status, energy consumption, temperature control, traffic and security logs are uploaded to the cloud, resulting in high bandwidth consumption, high transmission latency and cloud computing pressure. Fault detection and resource scheduling rely on centralized cloud processing, which results in a delayed response and makes it difficult to meet the needs of millisecond-level fault handling and real-time scheduling. The load forecasting and resource allocation lack global coordination and edge autonomy capabilities, resulting in high energy consumption and uneven resource utilization. Existing data center management systems mostly adopt a centralized architecture, lacking lightweight intelligent processing at the edge and a collaborative scheduling mechanism between the edge and cloud, thus failing to balance real-time performance, energy efficiency, and security. Therefore, this invention proposes an intelligent data center management system based on edge computing. Summary of the Invention
[0003] To address the problems of the above solutions, this invention provides a big data center intelligent management system based on edge computing.
[0004] The objective of this invention can be achieved through the following technical solutions: A big data center intelligent management system based on edge computing, including a cloud platform and edge terminals; Furthermore, the cloud platform establishes communication connections with each edge device.
[0005] The edge terminal includes a data acquisition module, an edge analysis module, and an interactive analysis module; The acquisition module is used to acquire data from the server in real time and obtain corresponding multi-dimensional acquisition data.
[0006] The edge analysis module is used to perform real-time edge analysis based on multi-dimensional collected data, obtain corresponding edge analysis results, and process the edge analysis results according to a preset processing method.
[0007] Furthermore, real-time edge analysis is performed, including: Identify various edge analytics tasks, including anomaly detection, load prediction, rack-level resource initial scheduling, and fault early warning; Configure a lightweight AI model for each edge analysis task, and analyze the multi-dimensional collected data through the lightweight AI model to obtain the edge analysis results of each edge analysis task.
[0008] Furthermore, based on each edge analysis task, a corresponding lightweight AI model is configured for the edge analysis module, including: Historical operational data from multiple data centers is obtained through a cloud platform, and a standardized training dataset is set up based on the obtained historical operational data. The standardized training dataset is cropped according to the actual data collection dimensions at the edge, and the basic model is trained based on the standardized training dataset. The corresponding basic model is then verified and adjusted according to each edge analysis task. The basic model is distributed to the edge analysis module at the edge, and the basic model is optimized and adjusted based on the preset local edge data. The optimized and adjusted basic model is converted into a format adapted to edge hardware and then stored locally at the edge to obtain a corresponding lightweight AI model.
[0009] The interactive analysis module is used to perform real-time importance assessment of multidimensional collected data based on the cloud interaction requirements of the cloud platform, obtain multidimensional collected data that meets the importance criteria, generate corresponding key summary data based on the multidimensional collected data, and send the key summary data to the cloud platform.
[0010] Furthermore, based on the cloud platform's cloud interaction requirements, real-time importance assessment of multi-dimensional collected data is performed, including: Based on the pre-set interaction judgment method, the corresponding edge interaction judgment requirements are analyzed in real time to obtain the corresponding requirement judgment results, which include whether there is an analysis requirement or not. When the requirement assessment result is that no requirement has been analyzed, no corresponding action is taken; When the demand judgment result indicates that there is an analysis demand, the abnormal cloud analysis demand corresponding to the edge interaction judgment demand is marked as a cloud interaction demand. Receive cloud interaction requests sent by the cloud platform in real time; The multi-dimensional data collected to assess cloud interaction needs meets the importance criteria.
[0011] Furthermore, when the corresponding edge interaction judgment requirements cannot be analyzed according to the pre-set interaction judgment method, collaborative analysis information of the corresponding interaction judgment method is generated and sent to the requirement analysis module of the cloud platform.
[0012] The cloud platform includes a demand analysis module and a cloud analysis module; The demand analysis module is used to perform cloud demand analysis, obtain various cloud analysis requirements of the cloud platform, classify and label each cloud analysis requirement, and obtain daily cloud analysis requirements and non-daily cloud analysis requirements. Mark routine cloud analytics needs as cloud interaction needs; Real-time identification of various non-routine cloud analysis needs that require analysis and judgment; real-time determination of whether the current cloud platform has the aforementioned non-routine cloud analysis needs; when the aforementioned non-routine cloud analysis needs are present, the non-routine cloud analysis needs are marked as cloud interaction needs. Send all cloud-based interaction requests to the edge-side interaction analysis module.
[0013] Furthermore, when there is no non-routine cloud analytics requirement, a request stop message is sent to the edge-side interactive analytics module.
[0014] Furthermore, the various non-routine cloud analytics requirements that need to be analyzed and judged constitute all non-routine cloud analytics requirements.
[0015] Furthermore, it identifies in real time various non-routine cloud analytics needs requiring analysis and judgment, including: Each non-routine cloud analytics requirement was analyzed and simulated separately to obtain comparative analysis simulation data; Based on comparative analysis and simulation data, a comparative evaluation was conducted to identify non-routine cloud analysis needs that meet the edge judgment criteria. Set corresponding edge interaction judgment requirements based on non-routine cloud analysis needs that meet edge judgment criteria; Configure the corresponding interaction judgment methods for the interaction analysis module at the edge based on the various edge interaction judgment requirements; Non-routine cloud analytics requests that do not meet the edge detection criteria will be considered as non-routine cloud analytics requests that require analysis and judgment.
[0016] Furthermore, the collaborative analysis information sent by the edge terminal is acquired in real time, and the corresponding non-routine cloud analysis needs are identified based on the collaborative analysis information. The non-routine cloud analysis needs are then added to the list of non-routine cloud analysis needs that require analysis and judgment.
[0017] The cloud analytics module is used to analyze the received key summary data, obtain cloud analytics results for the corresponding cloud analytics requirements, and manage the big data center accordingly based on the cloud analytics results.
[0018] Compared with the prior art, the beneficial effects of the present invention are: Low latency and high real-time performance: edge-based local processing reduces fault response and scheduling latency to the millisecond level, improving system stability.
[0019] Reduce bandwidth and save costs: Local preprocessing and data summarization reduce invalid data transmission and lower bandwidth and cloud computing costs.
[0020] High intelligence and low energy consumption: Global load prediction and cooling collaborative optimization significantly improve resource utilization, reduce PUE, and achieve green energy saving.
[0021] High reliability and easy operation and maintenance: edge autonomy + cloud coordination, single point of failure does not affect the whole; visualization and closed-loop operation and maintenance reduce labor costs.
[0022] Easy to expand and highly compatible: Supports unified access for multiple protocols, multiple devices, and multiple data centers, adapting to large-scale, distributed big data center scenarios. Attached Figure Description
[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0024] Figure 1 This is a block diagram illustrating the principle of the present invention. Detailed Implementation
[0025] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0026] like Figure 1 As shown, a big data center intelligent management system based on edge computing includes a cloud platform and an edge terminal; The cloud platform communicates with each edge device.
[0027] The edge devices are deployed inside the data center or near the data source to collect and preliminarily process multi-dimensional data such as server operating status, energy consumption, temperature control, network traffic, and security logs locally in real time; they also use lightweight AI models to achieve anomaly detection, load prediction, and initial resource scheduling, and upload key summary data to the central management platform.
[0028] The edge terminal includes a data acquisition module, an edge analysis module, and an interactive analysis module; The acquisition module is used to collect data from the server in real time and obtain corresponding multi-dimensional data, such as operating status, energy consumption, temperature control, network traffic and security logs.
[0029] In one implementation, multidimensional data acquisition undergoes data preprocessing, such as cleaning, noise reduction, and reduction transformation.
[0030] The edge analysis module is used to perform real-time edge analysis based on multi-dimensional collected data, obtain corresponding edge analysis results, and process the edge analysis results according to a preset processing method; that is, to process them according to the preset processing methods based on anomaly detection results, load prediction results, initial scheduling analysis results, fault warnings, etc.
[0031] In one embodiment, performing real-time edge analysis includes: Identify various edge analysis tasks, including anomaly detection, load prediction, rack-level resource initial scheduling and fault early warning. In other embodiments, there may be one or more of the above edge analysis tasks, or other different edge analysis tasks, which are determined according to the user's local analysis needs. Configure a lightweight AI model for each edge analysis task, and analyze the multi-dimensional collected data through the lightweight AI model to obtain the edge analysis results of each edge analysis task.
[0032] In one embodiment, a lightweight AI model is configured for each edge analysis module according to each edge analysis task, and is built manually based on existing AI technology.
[0033] In one embodiment, a lightweight AI model is configured for the edge analysis module according to each edge analysis task, including: Historical operational data (server load, energy consumption, temperature, fault logs, etc.) from multiple data centers are obtained through a cloud platform, and a standardized training dataset is constructed based on the obtained historical operational data. The standardized training dataset is trimmed according to the actual dimensions collected at the edge (e.g., only rack-level aggregated data is retained, and data center-level global data is removed) to ensure that the training data is consistent with the edge input. The basic model was trained using a standardized training set, and its accuracy in tasks such as anomaly detection and load prediction was verified (e.g., anomaly detection recall ≥95%, load prediction MAE ≤3%). The basic model is distributed to the edge analysis module at the edge, and lightweight fine-tuning is performed using 1-2 days of incremental data collected locally at the edge to adapt to the personalized operating characteristics of a single rack / single data center; that is, the basic model is optimized and adjusted based on the preset local edge data. The optimized and adjusted base model is converted into a format adapted to edge hardware (such as ONNX, TensorFlowLite) and solidified to local storage at the edge (such as local flash memory) to obtain the corresponding lightweight AI model.
[0034] For example, the corresponding lightweight AI models are configured for the edge analysis modules according to each edge analysis task, as shown in the table below:
[0035] In one embodiment, the edge device periodically uploads the performance of the lightweight AI model (such as the anomaly false negative rate and prediction error) to the cloud platform in the form of a summary, which can also be regarded as forming a corresponding cloud analysis requirement.
[0036] The cloud platform iteratively optimizes the basic model based on feedback from multiple edge devices and distributes lightweight update packages to the edge devices. The edge device completes local model updates during off-peak business hours (such as early morning), without interrupting data collection and analysis.
[0037] The interactive analysis module is used to perform real-time importance assessment of multidimensional collected data based on the cloud interaction requirements of the cloud platform, obtain multidimensional collected data that meets the importance criteria, generate corresponding key summary data based on the multidimensional collected data, and send the key summary data to the cloud platform.
[0038] In one embodiment, real-time importance assessment of multi-dimensional collected data based on cloud-based platform interaction requirements includes: Based on the pre-set interaction judgment method, the corresponding edge interaction judgment requirements are analyzed in real time to obtain the corresponding requirement judgment results, which include whether there is an analysis requirement or not. When the requirement assessment result is that no requirement has been analyzed, no corresponding action is taken; When the demand judgment result indicates that there is an analysis demand, the abnormal cloud analysis demand corresponding to the edge interaction judgment demand is marked as a cloud interaction demand. Receive cloud interaction requests sent by the cloud platform in real time; The multi-dimensional data collected to assess cloud interaction needs meets the importance criteria.
[0039] In one embodiment, when the corresponding edge interaction judgment requirement cannot be analyzed according to the pre-set interaction judgment method, collaborative analysis information of the corresponding interaction judgment method is generated, mainly including relevant information of non-routine cloud analysis requirements that need to be analyzed by the cloud platform; and the collaborative analysis information is sent to the cloud platform.
[0040] In one embodiment, corresponding key summary data is generated based on the multidimensional collected data. The key summary data range corresponding to each cloud analysis requirement is preset manually, and the corresponding data collection can be carried out subsequently.
[0041] In one embodiment, key summary data is generated based on the multidimensional collected data, or key summary data can be generated based on other existing methods.
[0042] For example, based on a pre-defined value assessment rule base (including weights for four dimensions: anomaly, trend, decision relevance, and transmission cost), the extracted feature data is quantitatively scored (out of 100 points): The comprehensive value score of feature data is calculated based on dimensional weights (e.g., anomaly weight 40% + trend weight 30% + decision relevance weight 20% + transmission cost weight 10%). Set value thresholds (60 points for daily scenarios and 50 points for non-daily scenarios) and filter feature data with scores ≥ the thresholds; The filtered feature data is then further simplified: duplicate features are removed and similar features are merged (e.g., "maximum cabinet temperature" and "hot spot temperature" are merged into "cabinet core temperature control feature").
[0043] The cloud platform includes a demand analysis module and a cloud analysis module; The demand analysis module is used to perform cloud demand analysis, obtain the various cloud analysis requirements of the cloud platform, that is, the various requirements that need to be analyzed by the cloud platform, and determine them based on the established cloud platform. If the analysis requirements of each edge terminal are different, they can be determined according to the analysis requirements of that edge terminal on the cloud platform, that is, determine the cloud analysis requirements corresponding to each edge terminal on the cloud platform. Classify and label the various cloud analytics needs to obtain daily cloud analytics needs and non-daily cloud analytics needs. Eliminate cloud analytics needs that are neither daily nor non-daily cloud analytics needs, that is, eliminate cloud analytics needs that do not require the edge to send corresponding key summary data and do not require interactive analysis. Mark routine cloud analytics needs as cloud interaction needs, and send cloud interaction needs to the edge-side interaction analytics module; The system analyzes in real time whether the cloud platform has non-routine cloud analysis needs, i.e., whether the corresponding cloud analysis needs are non-routine. When there are non-routine cloud analysis needs, the non-routine cloud analysis needs are marked as cloud interaction needs and sent to the edge interaction analysis module. When there are no non-routine cloud analysis needs, a need stop message is sent to the edge interaction analysis module, indicating that the edge interaction analysis module does not need to generate key summary data for the cloud analysis needs.
[0044] Routine cloud analytics requirements: With routine operation and maintenance and steady-state optimization as the core objectives, including daily load balancing, energy consumption monitoring and optimization, routine fault early warning, and equipment health assessment; the requirements are characterized by periodicity, stability, and low real-time (minute-level / hour-level), focusing on the aggregation and analysis of global steady-state data.
[0045] Non-routine cloud analytics needs: With emergency response, special optimization, and anomaly tracing as core objectives, these include fault diagnosis and tracing, analysis of sudden traffic surges, energy efficiency control under extreme weather conditions, security incident review, and system expansion / upgrade assessment. These needs are characterized by suddenness, temporariness, and high real-time requirements (seconds / milliseconds). They emphasize the accurate feedback of local anomaly data and detailed feature data, have high requirements for the "anomaly and detail integrity" of the data, and have increased tolerance for temporary bandwidth occupancy.
[0046] For routine cloud analytics needs, normalized transmission is adopted, that is, key summary data is obtained in real time from the multi-dimensional data collected for routine cloud analytics needs, which is regarded as a routine cloud interaction need; for non-routine cloud analytics needs, cloud interaction needs will only be adopted when the corresponding needs occur.
[0047] Therefore, based on the definitions and requirements of the above-mentioned routine cloud analytics needs and non-routine cloud analytics needs, each cloud analytics need can be classified and identified. Alternatively, it can be classified manually and then matched.
[0048] In one embodiment, certain non-routine cloud analytics needs can be identified and analyzed at the edge. Edge analysis offers higher efficiency, shortens the analysis time for these non-routine cloud analytics needs, and improves timeliness, especially for non-routine cloud analytics needs with high timeliness requirements. The specific process is as follows: Each non-routine cloud analytics request is analyzed and simulated separately to obtain comparative analysis simulation data. This involves simulating the analysis and judgment of each non-routine request by both the cloud platform and the edge, determining whether analysis is necessary, and obtaining relevant data on the accuracy, efficiency, cost, and resource consumption of the edge analysis simulations for both. This data is then integrated into comparative analysis simulation data. If the edge lacks relevant data or cannot perform analysis and judgment due to other reasons, then the data simulated by the edge is unavailable, meaning only the cloud platform can perform the analysis and judgment. A comparative evaluation is then conducted based on the comparative analysis simulation data to obtain satisfactory results. For non-routine cloud analysis needs that meet edge judgment criteria, the edge judgment criteria can be set according to user needs. For example, it can be determined that the simulation accuracy reaches a preset value, the efficiency is greater than the efficiency of the cloud platform analysis, and the impact on the edge is within a preset range (such as cost differences and resource consumption within preset values). It can also be judged by calculating a comprehensive value, or by comparing priorities based on priority algorithms. Multiple methods can be used to determine whether the judgment can be made by the edge. Corresponding edge interaction judgment requirements can be set according to non-routine cloud analysis needs that meet the edge judgment criteria. The above-mentioned non-routine cloud analysis needs are removed from the non-routine cloud analysis needs that are analyzed and judged in real time by the cloud platform. Configure corresponding interaction judgment methods for the interaction analysis module at the edge based on the various edge interaction judgment requirements. That is, determine whether to perform analysis for non-routine cloud analysis requirements based on the data at the edge, such as making judgments based on anomaly detection results. This is equivalent to adapting and migrating the analysis and judgment methods from the cloud platform to the edge.
[0049] In one embodiment, in practical applications, edge devices may be unable to perform analysis based on edge interaction requirements due to resource issues. To avoid delays, feedback can be sent to the cloud platform for analysis and judgment. In other words, the cloud platform can analyze and judge all non-routine cloud analysis requirements.
[0050] The system acquires collaborative analysis information sent from the edge in real time, identifies corresponding non-routine cloud analysis needs based on the collaborative analysis information, and treats the non-routine cloud analysis needs as non-routine cloud analysis needs that need to be analyzed and judged.
[0051] The cloud analytics module is used to analyze the received key summary data, obtain cloud analytics results for the corresponding cloud analytics requirements, and manage the big data center accordingly based on the cloud analytics results.
[0052] It integrates global topology, historical time-series data, edge summary data, and environmental data; employs deep learning and reinforcement learning models to perform long-term load prediction, hotspot location, energy efficiency optimization, cooling strategy matching, security situation analysis, and global resource scheduling; generates integrated optimization strategies for computing, storage, network, and cooling, summarizes the results into cloud analysis, and performs corresponding management based on the cloud analysis results.
[0053] Specifically, the analysis is conducted based on existing analytical techniques.
[0054] For example, the global data aggregation and standardized intelligent analysis engine receives high-value summary data from various edge computing nodes, including server load, rack power consumption, inlet / outlet air temperature, hotspot distribution, fan speed, air conditioning operating status, network traffic, abnormal events, and local initial scheduling results; it also accesses external data such as ambient temperature, humidity, weather forecasts, and electricity price periods. The engine performs unified cleaning, alignment, normalization, and spatiotemporal correlation on this multi-source heterogeneous data, forming a global training dataset and a real-time decision dataset covering the entire data center, all devices, and all time periods, providing a data foundation for subsequent optimization training and dynamic scheduling.
[0055] Model training and construction based on global big data utilizes global historical data as a foundation, and offline training is used to optimize multi-task models. Establish a load forecasting model to learn business patterns, time period characteristics, and load correlation between racks, and achieve accurate prediction of future hourly / daily load trends; Establish an energy consumption and temperature control coupled model to learn the mapping relationship between server load, rack power consumption, supply air temperature, return air temperature, cooling capacity distribution and air conditioning energy consumption, so as to achieve quantitative perception of PUE changes; A reinforcement learning-based optimization decision-making model is established, with the optimization objectives of "minimizing overall energy consumption, maximizing resource utilization, balancing load, and reducing hotspots." Resource allocation and cooling regulation form the action space. Through extensive simulations and trial-and-error iterations, the globally optimal collaborative strategy is output. After model training, it is loaded into the online inference module in real time to support dynamic strategy generation.
[0056] Global load balancing and dynamic resource allocation: Based on real-time global views and load prediction results, dynamic global resource allocation is performed across racks, regions, and clusters. Real-time statistics and hotspot identification of the entire data center load are performed to determine areas of local overload, light load, and idle resources. Based on business priority, reliability requirements, network topology and power supply constraints, computing tasks, storage tasks or network traffic in high-load areas are dynamically migrated, scheduled and diverted to light-load areas. Intelligent power-down, frequency reduction, and power consumption limits are implemented for idle servers, while capacity expansion or power limit increases are appropriately applied to high-load nodes to balance the load of each rack and data center, avoid single-point overload and local hotspots, and improve resource utilization and system stability.
[0057] Based on the rack load, hotspot distribution, and inlet and outlet air temperature difference, dynamically adjust the set temperature, air volume, cooling output, and number of operating precision air conditioners; For hot server racks, prioritize increasing cooling capacity and fan speed, and for low-load server racks, reduce cooling intensity. By combining external ambient temperature and natural cooling source conditions, the set temperature of the computer room is increased within permissible limits to fully utilize free cooling and reduce the energy consumption of the air conditioning system. Through a closed-loop mechanism of load scheduling first and cooling strategy following, the cooling supply and IT heat dissipation needs are matched in real time, avoiding over-cooling and ineffective energy consumption.
[0058] Edge computing nodes execute global scheduling instructions issued from the cloud and combine them with local real-time status to complete rack-level fine-grained execution; During execution, edge nodes will transmit execution results, load changes, temperature feedback, and energy consumption data back in real time in summary form, forming a closed loop of cloud decision-making, edge execution, and effect feedback.
[0059] The above formulas are all numerical calculations after removing dimensions. The formulas are obtained by software simulation based on a large amount of data and are closest to the real situation. The preset parameters and preset thresholds in the formulas are set by those skilled in the art according to the actual situation or obtained by simulation based on a large amount of data.
[0060] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.
Claims
1. A big data center intelligent management system based on edge computing, characterized in that, Including cloud platforms and edge devices; The edge terminal includes a data acquisition module, an edge analysis module, and an interaction analysis module; the cloud platform includes a demand analysis module and a cloud analysis module. The acquisition module is used to acquire data from the server in real time and obtain corresponding multi-dimensional acquisition data; The edge analysis module is used to perform real-time edge analysis based on multi-dimensional collected data, obtain corresponding edge analysis results, and process the edge analysis results according to a preset processing method. The interactive analysis module is used to perform real-time importance assessment of multidimensional collected data based on the cloud interaction requirements of the cloud platform, obtain multidimensional collected data that meets the importance criteria, generate corresponding key summary data based on the multidimensional collected data, and send the key summary data to the cloud platform. The demand analysis module is used to perform cloud demand analysis, obtain various cloud analysis demands of the cloud platform, classify and label each cloud analysis demand, and obtain daily cloud analysis demands and non-daily cloud analysis demands. Mark routine cloud analytics needs as cloud interaction needs; Real-time identification of various non-routine cloud analysis needs that require analysis and judgment; real-time determination of whether the current cloud platform has the aforementioned non-routine cloud analysis needs; when the aforementioned non-routine cloud analysis needs are present, the non-routine cloud analysis needs are marked as cloud interaction needs. Send all cloud-based interaction requests to the edge-side interaction analysis module; The cloud analytics module is used to analyze the received key summary data, obtain cloud analytics results for the corresponding cloud analytics requirements, and manage the big data center accordingly based on the cloud analytics results.
2. The intelligent management system for big data centers based on edge computing according to claim 1, characterized in that, The cloud platform communicates with each edge device.
3. The intelligent management system for big data centers based on edge computing according to claim 1, characterized in that, Perform real-time edge analysis, including: Identify various edge analytics tasks, including anomaly detection, load prediction, initial scheduling of rack-level resources, and fault early warning; Configure a lightweight AI model for each edge analysis task, and analyze the multi-dimensional collected data through the lightweight AI model to obtain the edge analysis results of each edge analysis task.
4. The intelligent management system for big data centers based on edge computing according to claim 3, characterized in that, Configure corresponding lightweight AI models for the edge analysis module according to each edge analysis task, including: Historical operational data from multiple data centers is obtained through a cloud platform, and a standardized training dataset is set up based on the obtained historical operational data. The standardized training dataset is cropped according to the actual data collection dimensions at the edge, and the basic model is trained based on the standardized training dataset. The corresponding basic model is then verified and adjusted according to each edge analysis task. The basic model is distributed to the edge analysis module at the edge, and the basic model is optimized and adjusted based on the preset local edge data. The optimized and adjusted basic model is converted into a format adapted to edge hardware and then stored locally at the edge to obtain a corresponding lightweight AI model.
5. The intelligent management system for big data centers based on edge computing according to claim 1, characterized in that, Based on cloud platform-based cloud interaction requirements, real-time importance assessment of multi-dimensional collected data is performed, including: Based on the pre-set interaction judgment method, the corresponding edge interaction judgment requirements are analyzed in real time to obtain the corresponding judgment results. The judgment results include whether there is an analysis requirement or not. When the requirement assessment result is that no requirements have been analyzed, no corresponding action is taken; When the demand judgment result indicates that there is an analysis demand, the abnormal cloud analysis demand corresponding to the edge interaction judgment demand is marked as a cloud interaction demand. Receive cloud interaction requests sent by the cloud platform in real time; The multi-dimensional data collected to assess cloud interaction needs meets the importance criteria.
6. The intelligent management system for big data centers based on edge computing according to claim 5, characterized in that, When the corresponding edge interaction judgment requirements cannot be analyzed according to the pre-set interaction judgment method, the collaborative analysis information of the corresponding interaction judgment method is generated and sent to the requirement analysis module of the cloud platform.
7. The intelligent management system for big data centers based on edge computing according to claim 1, characterized in that, When there is no non-routine cloud analytics requirement, a request stop message is sent to the interactive analytics module at the edge.
8. The intelligent management system for big data centers based on edge computing according to claim 1, characterized in that, All non-routine cloud analytics needs that require analysis and judgment constitute all non-routine cloud analytics needs.
9. The intelligent management system for big data centers based on edge computing according to claim 1, characterized in that, Real-time identification of various non-routine cloud analytics needs requiring analysis and judgment, including: Each non-routine cloud analytics requirement is analyzed and simulated separately to obtain comparative analysis simulation data; Based on comparative analysis and simulation data, a comparative evaluation was conducted to identify non-routine cloud analysis needs that meet the edge judgment criteria. Set corresponding edge interaction judgment requirements based on non-routine cloud analysis needs that meet edge judgment criteria; Configure the corresponding interaction judgment methods for the interaction analysis module at the edge based on the various edge interaction judgment requirements; Non-routine cloud analytics requests that do not meet the edge detection criteria will be considered as non-routine cloud analytics requests that require analysis and judgment.
10. A big data center intelligent management system based on edge computing according to claim 6, characterized in that, The system acquires collaborative analysis information sent from the edge in real time, identifies corresponding non-routine cloud analysis needs based on the collaborative analysis information, and adds the non-routine cloud analysis needs to the list of non-routine cloud analysis needs that need to be analyzed and judged.