A computing power center energy efficiency intelligent monitoring and optimization management method

Abstract

The application discloses a kind of computing power center energy efficiency intelligent monitoring and optimization management method, it relates to computing power center energy efficiency management technical field, the method includes following steps: S1, multi-source sensor network deployment and calibration;S2, multi-source data fusion acquisition and pre-processing;S3, multi-dimensional energy efficiency evaluation index system construction;S4, energy consumption anomaly identification and root location;S5, early warning information push and intelligent emergency disposal;S6, energy efficiency optimization suggestion generation and iterative update;The computing power center energy efficiency intelligent monitoring and optimization management method is through multi-dimensional data fusion acquisition, intelligent analysis and accurate early warning, realizes the rapid identification and root location of energy consumption anomaly, simultaneously adapts national green data center evaluation requirement, provides reliable data support for subsequent energy efficiency optimization, improves computing power center green low-carbon operation level.

Description

Technical Field

[0001] This invention belongs to the field of energy efficiency management technology for computing centers, and particularly relates to a method for intelligent monitoring and optimization management of energy efficiency in computing centers. Background Technology

[0002] With the rapid development of the digital economy, computing centers, as the core carriers of digital infrastructure, are continuously expanding in scale and their total energy consumption is constantly rising. Green and low-carbon operation has become a core requirement for the industry's development. The national green data center evaluation index system has also put forward clear requirements for power utilization efficiency and digital energy and carbon management level. Currently, the field of energy efficiency monitoring in computing centers faces numerous technical challenges: First, sensor deployment is scattered and lacks systematic planning, resulting in limited data acquisition due to its single-dimensional nature. Traditional monitoring methods can only collect single parameters such as temperature or power consumption, failing to achieve multi-parameter collaborative analysis and leading to high rates of missed or false alarms in energy consumption anomalies. Second, anomaly identification suffers from significant lag. Existing technologies largely rely on manual inspections or fixed threshold triggering for early warnings, failing to achieve real-time dynamic monitoring and struggling to quickly pinpoint the root cause of anomalies, thus impacting energy efficiency optimization and increasing operating costs and carbon emissions. Third, sensor data is susceptible to electromagnetic interference and environmental fluctuations in the data center, resulting in insufficient data accuracy. The limitations of single-sensor data further reduce the reliability of energy efficiency assessment and anomaly identification. Fourth, existing energy efficiency assessments often rely on a single PUE index, failing to reflect the true energy efficiency level of computing centers under different operating conditions and making it difficult to meet the multi-dimensional assessment requirements of national green data center evaluations. Fifth, multi-source data fusion often employs fixed-weight algorithms, which cannot address data conflicts under different operating conditions, resulting in limited fusion accuracy.

[0003] In existing technologies, some solutions attempt to introduce multi-sensor data collection, but no systematic fusion strategy has been formed. Moreover, the anomaly detection algorithm is singular and easily affected by the "curse of dimensionality" of high-dimensional data, resulting in insufficient stability of anomaly identification. At the same time, the emergency response mechanism lacks intelligent adaptability and cannot achieve precise handling based on the anomaly level and root cause. It is difficult to meet the high-end needs of refined and green energy efficiency management of computing centers, and it cannot fully adapt to the national green data center evaluation standards. Summary of the Invention

[0004] The purpose of this invention is to provide a method for intelligent monitoring and optimization management of energy efficiency in computing centers. This method achieves rapid identification and root cause location of energy consumption anomalies through multi-dimensional data fusion collection, intelligent analysis, and precise early warning. At the same time, it is compatible with the national green data center evaluation requirements, providing reliable data support for subsequent energy efficiency optimization and improving the green and low-carbon operation level of computing centers.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for intelligent monitoring and optimization management of energy efficiency in computing centers includes the following steps:

[0007] S1. Deployment and calibration of multi-source sensor networks: Deploy sensors of various specifications and scenarios to the computing center's computer room and outside, and perform unified calibration on all sensors to improve the accuracy of data acquisition.

[0008] S2. Multi-source data fusion acquisition and preprocessing: Through real-time sensor parameters, the raw sensor parameter data is cleaned and standardized. A dynamic weighted fusion algorithm based on evidence distance is used to fuse multi-sensor data in the same scene to generate a standardized fusion dataset.

[0009] S3. Construction of a multi-dimensional energy efficiency evaluation index system: Based on the fusion dataset and combined with the national green data center evaluation index system, a multi-dimensional energy efficiency evaluation index system including basic energy efficiency indicators, equipment operation energy efficiency indicators, cooling system energy efficiency indicators and green and low-carbon indicators is constructed using a big data analysis model. At the same time, the LSTM time series prediction algorithm is used to make short-term predictions for each index.

[0010] S4. Energy Consumption Anomaly Identification and Root Cause Location: Combining historical operating data of the computing center, industry standards and equipment parameter thresholds, a dynamic threshold setting method is used to adjust the normal threshold range of each indicator. An improved anomaly detection algorithm is used to perform real-time analysis on the fused dataset to identify abnormal data and corresponding abnormal indicators. Through correlation analysis model, the correlation between abnormal indicators and sensor data is mined to locate the root cause of the anomaly and generate early warning information including the anomaly level and the location of the root cause.

[0011] S5. Early Warning Information Push and Intelligent Emergency Response: Pushes early warning information to the management platform and personnel terminals in real time, and triggers differentiated emergency response mechanisms based on the level of abnormality and the type of root cause, realizing a combination of automatic, semi-automatic and manual response.

[0012] S6. Energy efficiency optimization suggestion generation and iterative update: Based on the fused dataset, anomaly identification results and emergency response data, generate personalized energy efficiency optimization suggestions; establish a model iterative update mechanism to regularly optimize the data fusion algorithm, anomaly detection algorithm and energy efficiency evaluation index system.

[0013] Furthermore, in step S1, the sensor deployment locations include: server racks in the computing center computer room, air outlets / return vents of the cooling system, the vicinity of power supply equipment, different areas of the computer room, and renewable energy power supply interfaces, where temperature sensors, power consumption sensors, humidity sensors, airflow sensors, cooling system operating parameter sensors, and renewable energy utilization sensors are deployed respectively.

[0014] Furthermore, each server in the server rack is equipped with one power consumption sensor and two temperature sensors to monitor the temperature at the server's air inlet and outlet, respectively, and one humidity sensor is deployed on the top of the rack; each device in the cooling system is equipped with one operating parameter sensor and one airflow sensor, and an additional temperature sensor is deployed at the cooling system's return air vent; at least one set of environmental temperature and humidity sensors is deployed every 50 square meters in the computer room, with denser deployment in the core area and a spacing of no more than 10 meters.

[0015] Furthermore, in step S2, data cleaning includes using an interpolation completion algorithm to handle missing values, using the 3σ principle to identify and correct outliers by combining historical data, using the Z-score method for standardization, and using a dynamic weight fusion algorithm based on evidence distance to dynamically generate weight coefficients for each sensor by calculating the evidence distance of different sensor data, thus handling the problem of multi-source data conflict.

[0016] Furthermore, in step S3, the basic energy efficiency indicators include energy consumption per unit area of ​​the computer room, PUE value, and energy consumption uniformity of the computer room; the equipment operation energy efficiency indicators include server power consumption per unit computing power, server load-power consumption ratio, server energy efficiency level, and server idle rate; the cooling system energy efficiency indicators include cooling power utilization rate, cooling system energy consumption ratio, cooling system COP value, and cooling system adjustment response speed; and the green and low-carbon indicators include renewable energy utilization rate, waste heat and waste cooling utilization level, and water resource utilization level.

[0017] Furthermore, in step S4, the improved anomaly detection algorithm is a sliding window + isolated forest + LOF algorithm. The sliding window is used to capture the temporal features of the data, the isolated forest algorithm is used for high-dimensional data anomaly identification, and the LOF algorithm is used for secondary verification to reduce the false positive rate. The association analysis model adopts a Bayesian network, and the anomaly level is divided into three levels: general, severe, and urgent.

[0018] Furthermore, the warning information, which includes the anomaly level and root cause location, is generated through a formulaic algorithm, specifically as follows:

[0019]

[0020] Di represents the degree of abnormal deviation;

[0021] Comprehensive abnormal deviation ,in As the indicator weight, The anomaly level is determined based on the D value; the root cause location of the anomaly is calculated using the Bayesian network posterior probability formula.

[0022]

[0023] Among them, the position with the highest posterior probability is taken as the root position.

[0024] Furthermore, in step S5, the tiered emergency response mechanism includes: general anomalies pushing adjustment suggestions for management personnel to implement adjustments; severe anomalies triggering semi-automatic handling, simultaneously pushing the handling process; emergency anomalies automatically cutting off power to non-core equipment, activating the emergency cooling system, and isolating the abnormal equipment; and a differentiated emergency response mechanism triggered by a formulaic algorithm, specifically:

[0025]

[0026] Where α is the anomaly level weight, β is the root cause type weight, α+β=1, and K is the handling trigger coefficient, which triggers three handling modes: manual, semi-automatic, and automatic, based on the value of K.

[0027] Furthermore, in step S6, the energy efficiency optimization suggestions include server load balancing adjustment, cooling system operating parameter optimization, renewable energy access ratio increase, and data center layout optimization, with an iterative update cycle of once per quarter.

[0028] In summary, the beneficial technical effects of the present invention are as follows:

[0029] 1. By systematically deploying multiple types of sensors to cover the entire computing center scenario, new data collection dimensions such as renewable energy and power supply parameters are added. Combined with sensor calibration and error correction models, the accuracy of raw data is improved. An improved dynamic weight fusion algorithm is adopted to effectively handle multi-source data conflicts. The fusion accuracy is improved by more than 15% compared with traditional algorithms, providing reliable data support for energy efficiency assessment and anomaly identification. At the same time, it meets the refined data collection requirements of the National Green Data Center.

[0030] 2. Construct a multi-dimensional indicator system covering four major categories: basic energy efficiency, equipment operation, cooling system, and green and low-carbon. Combined with the national green data center evaluation indicator requirements, introduce time-series prediction algorithms to not only comprehensively reflect the real energy efficiency level of the computing center under different operating conditions, but also provide data support for the application of green data centers, thus solving the problem of one-sided energy efficiency assessment in existing technologies.

[0031] 3. An improved anomaly detection algorithm combining sliding window, isolated forest, and LOF is adopted, along with a dynamic threshold setting method, to solve the problem of misjudgment by a single algorithm in high-dimensional data and uneven data distribution scenarios. The anomaly identification response time is shortened to the second level, and the false negative rate and false positive rate are reduced by more than 20%. Through the Bayesian network association analysis model, the root cause of the anomaly is accurately located, solving the pain point that existing technologies are difficult to locate the root cause of anomalies.

[0032] 4. Establish a tiered early warning and differentiated emergency response mechanism, combining automatic, semi-automatic, and manual response based on the anomaly level and root cause type, effectively avoiding energy waste and equipment damage, and improving the operational stability of the computing center; at the same time, establish a response effect evaluation mechanism to continuously optimize response strategies and adapt to the refined management needs of the computing center. Attached Figure Description

[0033] The accompanying drawings are provided to further illustrate the invention and form part of the specification, but do not constitute a limitation thereof. In the drawings:

[0034] Figure 1 This is a flowchart illustrating a method for intelligent monitoring and optimization management of energy efficiency in a computing center, as described in this embodiment. Detailed Implementation

[0035] The present invention will be further described in detail below with reference to the accompanying drawings.

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Please see Figure 1 This invention provides a technical solution: a method for intelligent monitoring and optimization management of energy efficiency in computing centers, comprising the following steps:

[0038] S1. Deployment and calibration of multi-source sensor networks: Deploy sensors of various specifications and scenarios to the computing center's computer room and outside, and perform unified calibration on all sensors to improve the accuracy of data acquisition.

[0039] S2. Multi-source data fusion acquisition and preprocessing: Through real-time sensor parameters, the raw sensor parameter data is cleaned and standardized. A dynamic weighted fusion algorithm based on evidence distance is used to fuse multi-sensor data in the same scene to generate a standardized fusion dataset.

[0040] S3. Construction of a multi-dimensional energy efficiency evaluation index system: Based on the fusion dataset and combined with the national green data center evaluation index system, a multi-dimensional energy efficiency evaluation index system including basic energy efficiency indicators, equipment operation energy efficiency indicators, cooling system energy efficiency indicators and green and low-carbon indicators is constructed using a big data analysis model. At the same time, the LSTM time series prediction algorithm is used to make short-term predictions for each index.

[0041] S4. Energy Consumption Anomaly Identification and Root Cause Location: Combining historical operating data of the computing center, industry standards and equipment parameter thresholds, a dynamic threshold setting method is used to adjust the normal threshold range of each indicator. An improved anomaly detection algorithm is used to perform real-time analysis on the fused dataset to identify abnormal data and corresponding abnormal indicators. Through correlation analysis model, the correlation between abnormal indicators and sensor data is mined to locate the root cause of the anomaly and generate early warning information including the anomaly level and the location of the root cause.

[0042] S5. Early Warning Information Push and Intelligent Emergency Response: Pushes early warning information to the management platform and personnel terminals in real time, and triggers differentiated emergency response mechanisms based on the level of abnormality and the type of root cause, realizing a combination of automatic, semi-automatic and manual response.

[0043] S6. Energy efficiency optimization suggestion generation and iterative update: Based on the fused dataset, anomaly identification results and emergency response data, generate personalized energy efficiency optimization suggestions; establish a model iterative update mechanism to regularly optimize the data fusion algorithm, anomaly detection algorithm and energy efficiency evaluation index system.

[0044] In step S1, the sensor deployment locations include: server racks in the computing center computer room, air outlets / return vents of the cooling system, the vicinity of power supply equipment, different areas of the computer room, and renewable energy power supply interfaces, where temperature sensors, power consumption sensors, humidity sensors, airflow sensors, cooling system operating parameter sensors, and renewable energy utilization sensors are deployed respectively.

[0045] Each server in the server rack is equipped with one power consumption sensor and two temperature sensors to monitor the air inlet and outlet temperatures of the server, respectively. A humidity sensor is also deployed on the top of the rack. Each device in the cooling system is equipped with one operating parameter sensor and one airflow sensor. An additional temperature sensor is deployed at the return air vent of the cooling system. At least one set of environmental temperature and humidity sensors is deployed for every 50 square meters of the server room, with denser deployment in core areas and a spacing of no more than 10 meters.

[0046] In step S2, data cleaning includes using an interpolation completion algorithm to handle missing values, using the 3σ principle to identify and correct outliers by combining historical data, using the Z-score method for standardization, and using a dynamic weight fusion algorithm based on evidence distance to dynamically generate weight coefficients for each sensor by calculating the evidence distance of different sensor data, thus handling the problem of multi-source data conflict.

[0047] In step S3, the basic energy efficiency indicators include energy consumption per unit area of ​​the data center, PUE value, and energy consumption uniformity of the data center; the equipment operation energy efficiency indicators include power consumption per unit computing power of the server, server load-power consumption ratio, server energy efficiency level, and server idle rate; the cooling system energy efficiency indicators include cooling power utilization rate, cooling system energy consumption ratio, cooling system COP value, and cooling system adjustment response speed; and the green and low-carbon indicators include renewable energy utilization rate, waste heat and waste cooling utilization level, and water resource utilization level.

[0048] In step S4, the improved anomaly detection algorithm is a sliding window + isolated forest + LOF algorithm. The sliding window is used to capture the temporal features of the data, the isolated forest algorithm is used for high-dimensional data anomaly identification, and the LOF algorithm is used for secondary verification to reduce the false positive rate. The association analysis model adopts a Bayesian network, and the anomaly level is divided into three levels: general, severe, and urgent.

[0049] Warning information, including anomaly level and root cause location, is generated using a formulaic algorithm, specifically:

[0050]

[0051] Di represents the degree of abnormal deviation;

[0052] Comprehensive abnormal deviation ,in As the indicator weight, The anomaly level is determined based on the D value; the root cause location of the anomaly is calculated using the Bayesian network posterior probability formula.

[0053]

[0054] Among them, the position with the highest posterior probability is taken as the root position.

[0055] In step S5, the tiered emergency response mechanism includes: general anomalies push adjustment suggestions for management personnel to implement adjustments; severe anomalies trigger semi-automatic handling, simultaneously pushing the handling process; emergency anomalies automatically cut off power to non-core equipment, activate the emergency cooling system, and isolate the abnormal equipment; and the differentiated emergency response mechanism is triggered by a formulaic algorithm, specifically:

[0056]

[0057] Where α is the anomaly level weight, β is the root cause type weight, α+β=1, and K is the handling trigger coefficient, which triggers three handling modes: manual, semi-automatic, and automatic, based on the value of K.

[0058] In step S6, the energy efficiency optimization suggestions include server load balancing adjustment, cooling system operating parameter optimization, increasing the proportion of renewable energy access, and data center layout optimization. The iteration update cycle is once per quarter.

[0059] Example

[0060] This technology is applied to a large-scale computing center with 1,000 standard racks, powered by renewable photovoltaic energy, aiming to apply for national green data center status. The specific steps are as follows:

[0061] S1. Deployment and Calibration of Multi-Source Sensor Networks

[0062] Multiple types of sensors are deployed in the computing center's computer room, including 1,000 server racks, 15 chiller units, 20 power distribution cabinets, different areas of the computer room, and photovoltaic access interfaces, forming a fully covered sensor network.

[0063] Server racks: Each server is equipped with one power consumption sensor with a measurement range of 0-5000W and an accuracy of ±1%, and two temperature sensors with a measurement range of -20℃ to 80℃ and an accuracy of ±0.5℃, which monitor the air inlet and outlet temperatures of the server respectively; each rack is also equipped with one humidity sensor on the top, with a measurement range of 0-100%RH and an accuracy of ±3%RH.

[0064] Chiller units: Each chiller unit is equipped with one operating parameter sensor to monitor cooling power 0-1000kW, operating frequency 50-60Hz, outlet water temperature 5-20℃ and one air volume sensor with a measurement range of 0-10000m³ / h. A temperature sensor is also installed at the return air vent of the chiller unit.

[0065] Power distribution cabinet: The total area of ​​the computer room is 5000㎡, divided into 50㎡ areas, with a total of 100 sets of environmental temperature and humidity sensors deployed. The core server cluster area is densely deployed with a spacing of 8m.

[0066] Each distribution cabinet is equipped with one voltage sensor with a measurement range of 0-400V and one current sensor with a measurement range of 0-1000A;

[0067] Renewable energy interface: Two renewable energy sensors are deployed at the photovoltaic access interface to monitor power generation and access power respectively.

[0068] After all sensors are deployed, they are uniformly calibrated using standard calibration equipment, a sensor error correction model is established, and real-time error compensation is performed on the collected data to ensure the accuracy of the original data acquisition.

[0069] S2. Multi-source data fusion acquisition and preprocessing

[0070] Multi-dimensional raw data is collected in real time at a frequency of 10 seconds per acquisition via a sensor network. This includes ambient temperature and humidity in the computer room, server power consumption and inlet / outlet temperatures, cooling system operating parameters, power supply voltage and current, photovoltaic power generation and access power. The raw data is processed using data cleaning algorithms: missing values ​​are filled with linear interpolation, outliers are identified using the 3σ principle, and corrections are made in conjunction with historical data from the same period. All data are standardized to the [0,1] interval through Z-score standardization. A dynamic weighted fusion algorithm based on evidence distance is used to calculate the evidence distance of different sensor data, dynamically allocate weight coefficients, and fuse multi-sensor data in the same scene to generate a standardized fusion dataset with a fusion accuracy of over 98%.

[0071] S3. Construction of a Multi-Dimensional Energy Efficiency Evaluation Index System

[0072] Based on the fused dataset and combined with the national green data center evaluation index system, a multi-dimensional energy efficiency assessment index system is constructed. The specific indicators and thresholds are as follows:

[0073] Basic energy efficiency indicators: Energy consumption per unit area of ​​the computer room ≤200kW·h / ㎡·year, PUE value ≤1.30 (meets the national green data center power utilization efficiency rating requirements), and energy consumption uniformity of the computer room ≤0.2;

[0074] Equipment operating energy efficiency indicators: server power consumption per unit computing power ≤ 0.5kW / TOPS, server load-power ratio ≥ 0.8, server energy efficiency level reaches level 2 or above as specified in GB 43630—2023, and server idle rate ≤ 5%;

[0075] Cooling system energy efficiency indicators: cooling power utilization rate ≥70%, cooling system energy consumption ratio ≤30%, cooling system COP value ≥4.0, cooling system adjustment response speed ≤30s;

[0076] Green and low-carbon indicators: renewable energy utilization rate ≥15% (reaching the renewable energy power consumption responsibility weight of the province), waste heat and waste cooling utilization level ≥20%, and water resource utilization level reaching the advanced level of the industry.

[0077] Using the LSTM time series prediction algorithm, short-term predictions of various indicators are made for 1 hour to identify abnormal trends in advance.

[0078] S4. Energy consumption anomaly identification and root cause location: Combine the historical operation data of the computing power center, industry standards, and equipment parameter thresholds, and use the dynamic threshold setting method to adjust the normal threshold ranges of each indicator; use the improved anomaly detection algorithm to perform real-time analysis on the fusion dataset to identify abnormal data and corresponding abnormal indicators; through the association analysis model, mine the association relationship between abnormal indicators and sensor data, locate the root cause of the anomaly, and generate warning information including the anomaly level and root cause location; among them, the generation of warning information is implemented using a formula-based algorithm, specifically as follows:

[0079] Anomaly level determination algorithm: Let the actual monitored value of the i-th energy efficiency indicator be x i , the upper limit of the dynamic threshold be T i,max , the lower limit of the threshold be T i,min , the indicator weight be w i (i = 1nwi = 1, n is the total number of energy efficiency indicators), the anomaly deviation degree D i The calculation formula is:

[0080]

[0081] The calculation formula for the comprehensive anomaly deviation degree D is: , and the anomaly level is defined according to the D value: when 0 < D ≤ 0.2, it is a general anomaly; when 0.2 < D ≤ 0.5, it is a serious anomaly; when D > 0.5, it is an emergency anomaly.

[0082] Anomaly root cause location algorithm: Based on the Bayesian network, let the abnormal indicator be A, and the data collected by the sensor be B j , j is the number of sensors, the root cause type is C k , k is the number of root cause types, the root cause location is L, and the calculation formula for the posterior probability of the root cause location is:

[0083]

[0084] Take the location with the largest posterior probability as the anomaly root cause location, and combine the root cause type Ck to generate complete warning information.

[0085] S5. Warning information push and intelligent emergency response: Push the warning information to the management platform and the terminals of management personnel in real time, and trigger a differentiated emergency response mechanism according to the anomaly level and root cause type to achieve the combination of automatic response, semi-automatic response and manual response; among them, the differentiated emergency response mechanism is triggered by a formula-based algorithm, specifically as follows:

[0086] Let the anomaly level be Grade and the root cause type be C k , the calculation formula for the response trigger coefficient K is:

[0087]

[0088] Where α is the weight of the anomaly level (0.6≤α≤0.8), β is the weight of the root cause type (0.2≤β≤0.4), and α+β=1; when K≤0.2 (corresponding to general anomaly), the manual handling mode is triggered, and adjustment suggestions are pushed; when 0.2<K≤0.5, corresponding to severe anomaly, the semi-automatic handling mode is triggered, the basic adjustment is automatically executed and the detailed process is pushed; when K>0.5, corresponding to emergency anomaly, the automatic handling mode is triggered, and emergency operations such as power outage and isolation are immediately executed.

[0089] Based on the fused dataset and anomaly handling data, the system generates energy efficiency optimization suggestions: optimize the cooling system outlet water temperature setting range, specifically adjusting it from 22℃±1℃ to 24℃±0.5℃, which is expected to reduce cooling system energy consumption by 15%; increase the photovoltaic grid connection ratio to 20% to further improve the utilization rate of renewable energy; and adopt a server load balancing strategy to reduce server idle rate to below 3%. Operational data is collected regularly, and the data fusion algorithm and anomaly detection algorithm are iteratively optimized quarterly to continuously improve monitoring accuracy and early warning accuracy, helping the computing center meet the national green data center evaluation requirements.

[0090] In this embodiment, after running for 6 months, the PUE value of the computing center decreased from 1.48 to 1.28, the renewable energy utilization rate increased to 18%, the false negative rate and false positive rate of energy consumption anomalies both decreased to below 5%, and the emergency response time was shortened to within 3 minutes. This effectively reduced operating costs and carbon emissions, met the core evaluation requirements of the national green data center, and verified the effectiveness and practicality of the method of this invention.

[0091] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0092] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent monitoring and optimization management of energy efficiency in computing centers, characterized in that, Includes the following steps: S1. Deployment and calibration of multi-source sensor networks: Deploy sensors of various specifications and scenarios to the computing center's computer room and outside, and perform unified calibration on all sensors to improve the accuracy of data acquisition. S2. Multi-source data fusion acquisition and preprocessing: Through real-time sensor parameters, the raw sensor parameter data is cleaned and standardized. A dynamic weighted fusion algorithm based on evidence distance is used to fuse multi-sensor data in the same scene to generate a standardized fusion dataset. S3. Construction of a multi-dimensional energy efficiency evaluation index system: Based on the fusion dataset and combined with the national green data center evaluation index system, a multi-dimensional energy efficiency evaluation index system including basic energy efficiency indicators, equipment operation energy efficiency indicators, cooling system energy efficiency indicators and green and low-carbon indicators is constructed using a big data analysis model. At the same time, the LSTM time series prediction algorithm is used to make short-term predictions for each index. S4. Energy Consumption Anomaly Identification and Root Cause Location: Combining historical operating data of the computing center, industry standards and equipment parameter thresholds, a dynamic threshold setting method is used to adjust the normal threshold range of each indicator. An improved anomaly detection algorithm is used to perform real-time analysis on the fused dataset to identify abnormal data and corresponding abnormal indicators. Through correlation analysis model, the correlation between abnormal indicators and sensor data is mined to locate the root cause of the anomaly and generate early warning information including the anomaly level and the location of the root cause. S5. Early Warning Information Push and Intelligent Emergency Response: Pushes early warning information to the management platform and personnel terminals in real time, and triggers differentiated emergency response mechanisms based on the level of abnormality and the type of root cause, realizing a combination of automatic, semi-automatic and manual response. S6. Generation and Iterative Update of Energy Efficiency Optimization Suggestions: Based on the fused dataset, anomaly identification results, and emergency response data, personalized energy efficiency optimization suggestions are generated. Establish a model iteration and update mechanism to regularly optimize data fusion algorithms, anomaly detection algorithms, and energy efficiency assessment index systems.

2. The intelligent monitoring and optimization management method for energy efficiency of computing centers according to claim 1, characterized in that, In step S1, the sensor deployment locations include: server racks in the computing center computer room, air outlets / return vents of the cooling system, the vicinity of power supply equipment, different areas of the computer room, and renewable energy power supply interfaces, where temperature sensors, power consumption sensors, humidity sensors, airflow sensors, cooling system operating parameter sensors, and renewable energy utilization sensors are deployed respectively.

3. The intelligent monitoring and optimization management method for energy efficiency of computing centers according to claim 2, characterized in that, Each server in the server rack is equipped with one power consumption sensor and two temperature sensors to monitor the temperature at the server's air inlet and outlet, respectively. A humidity sensor is also deployed on the top of the rack. Each device in the cooling system is equipped with one operating parameter sensor and one airflow sensor. An additional temperature sensor is deployed at the return air vent of the cooling system. At least one set of environmental temperature and humidity sensors is deployed every 50 square meters in the computer room, with denser deployment in core areas and a spacing of no more than 10 meters.

4. The intelligent monitoring and optimization management method for energy efficiency of computing centers according to claim 1, characterized in that, In step S2, data cleaning includes using an interpolation completion algorithm to handle missing values, using the 3σ principle to identify and correct outliers by combining historical data, using the Z-score method for standardization, and using a dynamic weight fusion algorithm based on evidence distance to dynamically generate weight coefficients for each sensor by calculating the evidence distance of different sensor data, thus handling the problem of multi-source data conflict.

5. The intelligent monitoring and optimization management method for energy efficiency of computing centers according to claim 1, characterized in that, In step S3, the basic energy efficiency indicators include energy consumption per unit area of ​​the computer room, PUE value, and energy consumption uniformity of the computer room; the equipment operation energy efficiency indicators include power consumption per unit computing power of the server, server load-power consumption ratio, server energy efficiency level, and server idle rate; the cooling system energy efficiency indicators include cooling power utilization rate, cooling system energy consumption ratio, cooling system COP value, and cooling system adjustment response speed; and the green and low-carbon indicators include renewable energy utilization rate, waste heat and waste cooling utilization level, and water resource utilization level.

6. The intelligent monitoring and optimization management method for energy efficiency of computing centers according to claim 1, characterized in that, In step S4, the improved anomaly detection algorithm is a sliding window + isolated forest + LOF algorithm. The sliding window is used to capture the temporal features of the data, the isolated forest algorithm is used for high-dimensional data anomaly identification, and the LOF algorithm is used for secondary verification to reduce the false judgment rate. The association analysis model adopts a Bayesian network, and the anomaly level is divided into three levels: general, severe, and urgent.

7. The intelligent monitoring and optimization management method for energy efficiency of computing centers according to claim 6, characterized in that, The warning information, which includes the anomaly level and the root cause location, is generated through a formulaic algorithm, specifically as follows: Di represents the degree of abnormal deviation; Comprehensive abnormal deviation ,in As the indicator weight, The anomaly level is determined based on the D value; the root cause location of the anomaly is calculated using the Bayesian network posterior probability formula. Among them, the position with the highest posterior probability is taken as the root position.

8. The intelligent monitoring and optimization management method for energy efficiency of computing centers according to claim 1, characterized in that, In step S5, the tiered emergency response mechanism includes: general anomalies push adjustment suggestions for management personnel to implement adjustments; severe anomalies trigger semi-automatic handling, simultaneously pushing the handling process; emergency anomalies automatically cut off power to non-core equipment, activate the emergency cooling system, and isolate the abnormal equipment; and the differentiated emergency response mechanism is triggered by a formulaic algorithm, specifically: Where α is the anomaly level weight, β is the root cause type weight, α+β=1, and K is the handling trigger coefficient, which triggers three handling modes: manual, semi-automatic, and automatic, based on the value of K.

9. The intelligent monitoring and optimization management method for energy efficiency of computing centers according to claim 1, characterized in that, In step S6, the energy efficiency optimization suggestions include server load balancing adjustment, cooling system operating parameter optimization, increasing the proportion of renewable energy access, and data center layout optimization, with an iterative update cycle of once per quarter.

Images