A method and system for remote monitoring of a machine room power distribution equipment

Abstract

The application belongs to the technical field of machine room power distribution, and provides a machine room power distribution equipment remote monitoring method and system, which comprises the following steps: collecting operation data in real time through deploying edge equipment, fusing basic operation data, transient characteristic data and historical abnormal data to construct a state scoring model, and utilizing neural network training weight coefficients to realize dynamic evaluation of equipment health degree; based on the scoring result, the sampling frequency is adaptively adjusted, combined with current mutation intensity, energy mutation amount, temperature gradient and other multi-feature threshold values, abnormal data is screened, and only abnormal period high-frequency data segments are reserved; through edge-cloud collaborative storage mechanism, after abnormal data is subjected to RAID redundancy check, the abnormal data is locally cached, normal data is uploaded to the cloud according to storage time length and access frequency, data full life cycle traceability is realized through labeling processing, abnormal detection sensitivity is improved, and storage efficiency and fault diagnosis reliability are taken into account.

Description

Technical Field

[0001] This invention belongs to the field of computer room power distribution technology, specifically a method and system for remote monitoring of computer room power distribution equipment. Background Technology

[0002] Data center power distribution equipment is a core infrastructure for the stable operation of a data center. Faults such as current fluctuations and abnormal temperatures can easily lead to serious consequences such as power outages and equipment damage, thus requiring efficient remote monitoring. However, existing technologies have significant shortcomings:

[0003] Fixed sampling frequencies are often used. While high-frequency sampling can capture anomalies, it generates a large amount of redundant data, consuming edge and cloud storage and bandwidth resources. Low-frequency sampling, on the other hand, is prone to missing transient faults such as arc discharge and abnormal contact resistance, resulting in delayed fault response.

[0004] Relying solely on basic data such as current and voltage to determine equipment status, without incorporating risk information such as historical abnormal frequency and maintenance cycle, results in low accuracy in assessment and makes it difficult to predict potential faults.

[0005] Data is not classified by value, normal and abnormal data are stored together, high-frequency data occupies high-performance storage for a long time, resulting in excessive costs; and there is a lack of unified metadata tags, so manual investigation of data storage location is required when tracing the source of faults, which is inefficient.

[0006] Edge storage is mostly based on a single-disk architecture. Hard disk failure can easily lead to the loss of critical abnormal data, which cannot provide a complete basis for fault diagnosis. There is an urgent need for a remote monitoring solution that takes into account accurate monitoring, resource optimization, reliable storage and efficient traceability.

[0007] Therefore, the present invention provides a method and system for remote monitoring of power distribution equipment in a computer room. Summary of the Invention

[0008] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0009] The technical solution adopted by this invention to solve its technical problem is:

[0010] In a first aspect, the present invention provides a method for remote monitoring of power distribution equipment in a computer room, comprising:

[0011] S1: Deploy edge devices to collect real-time operating data of data center power distribution equipment, build a status scoring model for data center power distribution equipment, and build a dynamic sampling matrix based on the status scoring model for data center power distribution equipment;

[0012] S2: Perform transient feature extraction processing on high-frequency sampled data in the edge device, set multiple feature thresholds, filter the high-frequency sampled data after transient feature extraction processing, distinguish abnormal data from normal data, and retain abnormal data;

[0013] S3: Stores abnormal data on edge devices and verifies the storage reliability and real-time access of the stored abnormal data. Normal data is uploaded to the cloud in a tiered manner, and all abnormal and normal data are tagged to form traceable data.

[0014] As a further improvement of the present invention, the specific process of constructing the status scoring model of the computer room power distribution equipment includes:

[0015] A status scoring model for computer room power distribution equipment is generated by integrating basic operational data, transient characteristic data, and historical anomaly data.

[0016] Basic operating data includes: load rate, RMS current, RMS voltage, and ambient temperature and humidity, used to analyze the long-term operating trends of the equipment.

[0017] Transient characteristic data includes: current fluctuation variance, voltage fluctuation variance, and temperature change rate, which are used to capture initial abnormal fluctuation characteristics;

[0018] Historical anomaly data includes: the frequency of anomalies in similar equipment and current equipment maintenance cycle data, which are used to predict potential risks;

[0019] The basic operational data, transient characteristic data, and historical anomaly data are processed separately.

[0020] The basic operational data processing procedure is as follows: select load rate RMS value of current RMS voltage Ambient temperature Ambient humidity Basic characteristics;

[0021] For each basic feature, linear normalization is applied, and then a weighted average is taken of the normalized basic features to obtain the comprehensive normalized value of the basic operational data. : ,in, , , , , The weights for each basic feature are obtained by training a neural network using historical fault data, and satisfy the following conditions: , , , , , These are the normalized values ​​of load factor, RMS current, RMS voltage, ambient temperature, and ambient humidity, respectively.

[0022] The transient characteristic data processing procedure involves selecting the current fluctuation variance. Voltage fluctuation variance Rate of temperature change Transient features are identified; after normalization of each transient feature, a weighted average is taken to obtain the comprehensive normalized value of the transient features. : ,in, , , The transient feature weights are obtained by training a neural network using historical fault data and satisfy the following conditions: , , , These are the normalized values ​​for current fluctuation variance, voltage fluctuation variance, and temperature change rate, respectively.

[0023] The historical anomaly data processing procedure involves defining historical anomaly factors. Abnormal frequency of similar equipment in the past 7 days Compared with the current equipment maintenance cycle percentage Weighted sum: ,in , This is the normalized value of the corresponding indicator; , As the weight, satisfying ;

[0024] Again Inverse normalization means that the more anomalies and the higher the risk, the lower the score should be: .

[0025] As a further improvement of the present invention, the specific process of constructing the status scoring model of the computer room power distribution equipment also includes:

[0026] The median state score was obtained by weighted fusion and normalized weighted processing of basic operational data, transient characteristic data, and historical anomaly data. Then, the median value of the status score is linearly mapped to the [0,100] interval to construct the status score model of the computer room power distribution equipment: ,in, , , The weighting coefficients for basic operational data, transient feature data, and historical anomaly data are respectively obtained through training and optimization using machine learning algorithms. For rounding operation, Based on the comprehensive normalized value of the basic operating data, This is the normalized value of the transient features. Abnormal frequency of similar equipment in the past 7 days Compared with the current equipment maintenance cycle percentage The inverse normalized value of the weighted sum.

[0027] As a further improvement of the present invention, the specific process of constructing the dynamic sampling matrix is ​​as follows:

[0028] Based on the status scoring model of computer room power distribution equipment Define three intervals, namely the stable interval. Fluctuation range Abnormal range The interval position of the status score of the computer room power distribution equipment;

[0029] If the status score of the computer room power distribution equipment is in a stable range, then each sensor will sample according to the basic sampling frequency;

[0030] If the status score of the computer room power distribution equipment is in the fluctuation range, each sensor will be sampled at twice the basic sampling frequency;

[0031] If the status score of the power distribution equipment in the computer room is in the abnormal range, each sensor will sample at three times the basic sampling frequency.

[0032] As a further improvement of the present invention, the specific process of performing transient feature extraction processing on high-frequency sampling data is as follows:

[0033] Extracting transient current characteristics:

[0034] Calculate the current signal over the time interval The rate of change within reflects the intensity of the sudden change in current: ,in, , For a moment , Current value, time interval In this invention ;

[0035] Calculate the abrupt change in current energy by integrating the square of the current over the time interval. The results within reflect the degree of energy release from a short circuit or electric arc: ,in, This refers to the sudden change in current energy. This represents the starting time of the integration time interval. This represents the end of the integration time interval. It is a time-domain function of the current;

[0036] Simpson's numerical integration method is used for approximate calculation. ,in, For time step, Time interval The total number of current sampling points within and Even number, The summation index ranges from 1 to... , Within the time interval At the (2k-1)th odd-numbered sampling time step, Time interval The sampling time of the 2kth even-numbered step within;

[0037] Extracting transient temperature features:

[0038] Temperature gradient, calculating the temperature signal over a time interval The rate of change within reflects the rate of temperature increase and decrease: ,in, , They are time points , Temperature value;

[0039] Thermal diffusion delay: Calculates the time difference between the peak busbar temperature and the peak sensor temperature, reflecting the heat transfer delay caused by abnormal contact resistance. ,in, Busbar temperature peak time, For sensor to detect temperature The peak time.

[0040] As a further improvement of the present invention, the specific process of distinguishing abnormal data from normal data is as follows:

[0041] Data is considered abnormal and data retention is triggered when any of the following conditions are met: current surge intensity. Rate of temperature change Thermal diffusion delay Otherwise, it is considered normal data.

[0042] Anomaly thresholds for each physical feature: ,in, The threshold for the intensity of the current surge. The threshold values ​​for the rate of temperature increase and decrease. This represents the heat transfer delay threshold.

[0043] As a further improvement of the present invention, the specific process of retaining abnormal data is as follows:

[0044] Record before the anomaly occurs Until the anomaly ends Complete high-frequency data, data length is: ,in, This is the current sampling frequency.

[0045] As a further improvement of the present invention, the specific process of storing abnormal data at the edge device is as follows:

[0046] When the status score of the computer room power distribution equipment is in an abnormal range, or when it is determined to be abnormal data based on the abnormal threshold of various physical characteristics, the abnormal data caching mechanism is triggered, and the amount of cached data is [not specified]. satisfy: ,in, The cache duration before the exception occurs. The duration of the anomaly. The cache duration after the exception ends. This refers to the number of bytes of data per sampling point. This is the current sampling frequency.

[0047] As a further improvement of the present invention, the storage reliability and access real-time performance of the stored abnormal data are verified.

[0048] Storage reliability verification:

[0049] The fault injection test was used to verify the integrity of the abnormal data in another SSD by simulating a sudden failure of a single SSD. This included comparing the number of data files and content hash values ​​before the failure to confirm that no data was lost.

[0050] If the integrity of the abnormal data in the other SSD is intact, the storage reliability verification passes.

[0051] If the integrity of the abnormal data in the other SSD is incomplete, the storage reliability verification fails, and the SSD is repaired.

[0052] After the fault is repaired, the faulty disk is replaced, and it is verified whether the RAID array automatically rebuilds the image to ensure the reliability of long-term storage.

[0053] Real-time access verification:

[0054] Initiate a retrieval request for abnormal data and record the response time from the request being initiated to the complete presentation of the abnormal data. Through multiple tests, if the response time is consistently within 1 second, the real-time access requirement is met; if the response time is consistently more than 1 second, the real-time access requirement is not met.

[0055] If the response time is consistently within 1 second, the real-time access requirement is met.

[0056] Secondly, the present invention provides a remote monitoring system for computer room power distribution equipment, comprising:

[0057] Data acquisition and sampling control module: Deploy edge devices to collect real-time operating data of the data center power distribution equipment, build a status scoring model for the data center power distribution equipment, and construct a dynamic sampling matrix based on the status scoring model;

[0058] High-frequency data transient processing and filtering module: Performs transient feature extraction processing on high-frequency sampled data in edge devices, sets multiple feature thresholds, filters the high-frequency sampled data after transient feature extraction processing, distinguishes abnormal data from normal data, and retains abnormal data;

[0059] Data tiered storage and traceability module: Abnormal data is stored at the edge device, and the storage reliability and access timeliness of the stored abnormal data are verified. Normal data is uploaded to the cloud in tiers, and all abnormal and normal data are tagged to form traceable data.

[0060] The beneficial effects of this invention are as follows:

[0061] 1. By integrating three types of data—basic operation, transient features, and historical anomalies—linear normalization is used to eliminate dimensional differences. Combined with weighted fusion of neural network and machine learning training, a [0,100] quantified score is generated, which can characterize the device status from multiple perspectives, avoid the evaluation bias of a single data dimension, and provide accurate basis for subsequent sampling and storage strategies. A dynamic sampling matrix is ​​constructed based on the status score. The basic frequency is used to reduce redundancy in the stable range, the frequency of 2 times is used to balance trend capture and cost in the fluctuating range, and the frequency of 3 times is used to fully record transient faults in the abnormal range. Compared with fixed high-frequency sampling, invalid data can be reduced, and the storage and bandwidth pressure on the edge and cloud can be reduced.

[0062] 2. By extracting features such as current surge intensity, temperature gradient, and thermal diffusion delay through lightweight physical algorithms, microsecond-level current anomalies and latent temperature faults can be captured. At the same time, normal data is downsampled or discarded to reduce redundancy at the source, ensuring that edge device computing power is focused on critical anomaly data processing. Anomaly data is stored using RAID1 dual-disk mirroring. Fault injection tests verify that no data is lost when a single disk fails, and the mirror is automatically rebuilt after repair, ensuring the security of core data. Normal data is classified, and cold data and transitional data are stored in low-cost archive storage. Compared with full-volume high-performance storage, this reduces long-term storage costs and improves the intelligent operation and maintenance level of the data center power distribution system. Attached Figure Description

[0063] The invention will now be further described with reference to the accompanying drawings.

[0064] Figure 1 This is a flowchart of the steps of a remote monitoring method for computer room power distribution equipment according to the present invention;

[0065] Figure 2This is a system module diagram of a remote monitoring system for computer room power distribution equipment according to the present invention. Detailed Implementation

[0066] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0067] Example 1

[0068] like Figure 1 As shown in the embodiment of the present invention, a method for remote monitoring of computer room power distribution equipment includes:

[0069] S1: Deploy edge devices to collect real-time operating data of data center power distribution equipment, build a status scoring model for data center power distribution equipment, and build a dynamic sampling matrix based on the status scoring model for data center power distribution equipment;

[0070] Deploy edge devices to collect real-time operating data of data center power distribution equipment and build a status scoring model for data center power distribution equipment;

[0071] Deploy edge devices to collect real-time operational data of data center power distribution equipment to sense its status, and generate a status scoring model for data center power distribution equipment by integrating basic operational data, transient feature data, and historical anomaly data.

[0072] The basic operating data includes: load rate, RMS current, RMS voltage, and ambient temperature and humidity, which are used to analyze the long-term operating trends of the equipment.

[0073] Transient characteristic data includes: current fluctuation variance, voltage fluctuation variance, and temperature change rate, which are used to capture initial abnormal fluctuation characteristics;

[0074] Historical anomaly data includes: the frequency of anomalies in similar equipment and current equipment maintenance cycle data, which are used to predict potential risks;

[0075] The basic operational data, transient characteristic data, and historical anomaly data are processed separately.

[0076] The basic operational data processing procedure involves selecting the load rate. RMS value of current RMS voltage Ambient temperature Ambient humidity Basic characteristics;

[0077] For each basic feature, linear normalization is applied, and then a weighted average is taken of the normalized basic features to obtain the comprehensive normalized value of the basic operational data. : ,in, , , , , The weights for each basic feature are obtained by training a neural network using historical fault data, and satisfy the following conditions: , , , , , These are the normalized values ​​of load factor, RMS current, RMS voltage, ambient temperature, and ambient humidity, respectively.

[0078] The transient feature data processing procedure involves selecting the current fluctuation variance. Voltage fluctuation variance Rate of temperature change Transient features are identified; after normalization of each transient feature, a weighted average is taken to obtain the comprehensive normalized value of the transient features. : ,in, , , The transient feature weights are obtained by training a neural network using historical fault data and satisfy the following conditions: , , , These are the normalized values ​​for current fluctuation variance, voltage fluctuation variance, and temperature change rate, respectively.

[0079] The historical anomaly data processing procedure involves defining historical anomaly factors. Abnormal frequency of similar equipment in the past 7 days Compared with the current equipment maintenance cycle percentage Weighted sum: ,in , This is the normalized value of the corresponding indicator; , As the weight, satisfying ;

[0080] Again Inverse normalization means that the more anomalies and the higher the risk, the lower the score should be: ;

[0081] The median state score was obtained by weighted fusion and normalized weighted processing of basic operational data, transient characteristic data, and historical anomaly data. Then, the median value of the status score is linearly mapped to the [0,100] interval to construct the status score model of the computer room power distribution equipment: ,in, , , The weighting coefficients for basic operational data, transient feature data, and historical anomaly data are respectively obtained through training and optimization using machine learning algorithms. This is a rounding operation;

[0082] By integrating three types of data—basic operational data, transient characteristic data, and historical anomaly data—it breaks through the limitations of a single data dimension and can comprehensively depict the operating status of the computer room power distribution equipment from multiple perspectives, including long-term trends, real-time fluctuations, and historical risks.

[0083] By eliminating the dimensional differences between different data through linear normalization, and then combining the weight coefficients obtained from training neural networks and machine learning algorithms for weighted fusion, the status score can adapt to the data characteristics under different scenarios, thereby improving the accuracy and relevance of equipment status assessment.

[0084] Based on the status scoring model of the computer room power distribution equipment, a dynamic sampling matrix is ​​constructed;

[0085] Based on the status scoring model of computer room power distribution equipment Define three intervals, namely the stable interval. Fluctuation range Abnormal range The interval position of the status score of the computer room power distribution equipment;

[0086] If the status score of the computer room power distribution equipment is in a stable range, then each sensor will sample according to the basic sampling frequency;

[0087] If the status score of the computer room power distribution equipment is in the fluctuation range, each sensor will be sampled at twice the basic sampling frequency;

[0088] If the status score of the computer room power distribution equipment is in the abnormal range, each sensor will be sampled at three times the basic sampling frequency;

[0089] By dividing the device status into three ranges—stable, fluctuating, and abnormal—and corresponding to different multiples of the base sampling frequency, precise coupling between the sampling frequency and the actual operating status of the device is achieved. High-frequency redundant sampling is avoided when the device is stable; trend capture and storage costs are balanced during fluctuations; and transient processes are fully recorded through high-frequency sampling during abnormal situations, making data collection more targeted and rational.

[0090] Using the basic sampling frequency under stable conditions can significantly reduce the generation of redundant high-frequency data, effectively reduce the storage pressure and data transmission bandwidth occupation on the edge and cloud sides, avoid the resource waste caused by fixed high-frequency sampling, provide complete raw data support for subsequent fault diagnosis and source tracing, and improve the detection sensitivity and response time of data center power distribution anomalies.

[0091] S2: Perform transient feature extraction processing on high-frequency sampled data in the edge device, set multiple feature thresholds, filter the high-frequency sampled data after transient feature extraction processing, distinguish abnormal data from normal data, and retain abnormal data;

[0092] Transient feature extraction processing is performed on high-frequency sampled data in edge devices;

[0093] For common transient anomalies in power distribution equipment, including current transient characteristics, voltage transient characteristics, and temperature transient characteristics, a lightweight physical feature extraction algorithm is used to extract transient features.

[0094] Extracting transient current features specifically includes:

[0095] Calculate the current signal over the time interval The rate of change within reflects the intensity of the sudden change in current: ,in, , For a moment , Current value, time interval In this invention ;

[0096] Calculate the abrupt change in current energy by integrating the square of the current over the time interval. The results within reflect the degree of energy release from a short circuit or electric arc: ,in, This refers to the sudden change in current energy. This represents the starting time of the integration time interval. This represents the end of the integration time interval. It is a time-domain function of the current;

[0097] Simpson's numerical integration method is used for approximate calculation. ,in, For time step, Time interval The total number of current sampling points within and Even number, The summation index ranges from 1 to... , Within the time interval At the (2k-1)th odd-numbered sampling time step, Time interval The sampling time of the 2kth even-numbered step within;

[0098] Extracting transient temperature features, specifically including:

[0099] Temperature gradient, calculating the temperature signal over a time interval The rate of change within reflects the rate of temperature increase and decrease: ,in, , They are time points , Temperature value;

[0100] Thermal diffusion delay: Calculates the time difference between the peak busbar temperature and the peak sensor temperature, reflecting the heat transfer delay caused by abnormal contact resistance. ,in, Busbar temperature peak time, For sensor to detect temperature Peak time;

[0101] The high-frequency sampling data after transient feature extraction is filtered to distinguish between abnormal and normal data, and the abnormal data is retained.

[0102] Set anomaly thresholds for each physical feature;

[0103] Data is considered abnormal and data retention is triggered when any of the following conditions are met: current surge intensity. Rate of temperature change Thermal diffusion delay Otherwise, it is considered normal data.

[0104] Anomaly thresholds for each physical feature: ,in, The threshold for the intensity of the current surge. The threshold values ​​for the rate of temperature increase and decrease. This represents the heat transfer delay threshold.

[0105] Retain abnormal data, retaining only transient fragments of abnormal data, and downsample or discard normal data;

[0106] Record before the anomaly occurs Until the anomaly ends Complete high-frequency data, data length is: ,in, This is the current sampling frequency;

[0107] For normal data, a sliding window mean downsampling method is used to take the data within the window. The average of several high-frequency points is used as a representative point, and the frequency after downsampling is the original frequency. ;

[0108] Let the original data be , Downsampling data satisfy: ,in, , The result of rounding down M / N represents the total number of data points after downsampling. This is a summation operation on the (j-1)N+1th to jNth points in the original data;

[0109] If the fluctuation range of M=100 consecutive high-frequency data points satisfies: If the segment is completely stable, the high-frequency data is discarded.

[0110] It should be noted that M=100 represents the number of consecutive high-frequency data points used to determine the fully stable segment. The maximum value among these M consecutive high-frequency data points. The minimum value among these M consecutive high-frequency data points. The pre-set fluctuation range threshold;

[0111] For core fault scenarios in data center power distribution, a targeted transient feature extraction logic was designed. By measuring the rate of change of current and the sudden change in current energy, microsecond-level transient anomalies of current can be captured. By measuring the temperature gradient and the thermal diffusion delay, latent temperature anomalies can be identified. Through differentiated strategies of retaining all abnormal data and simplifying normal data, the amount of invalid data is reduced from the source of data, avoiding a large amount of stable and worthless high-frequency data from occupying resources.

[0112] Abnormal data retains only the key segments from before to after the abnormality, further controlling the amount of data and reducing redundant data on the edge side. This reduces the local storage load on edge devices and the bandwidth consumption of subsequent data transmission to the cloud, alleviating the overall system resource pressure.

[0113] It can perform large-scale deep learning inference or complex matrix operations without the need for edge devices. Feature extraction can be completed using only basic physical formulas, and the processing latency can be controlled in the millisecond level.

[0114] S3: Store abnormal data on edge devices and verify the storage reliability and real-time access of the stored abnormal data. Upload normal data to the cloud in a hierarchical manner and label all abnormal and normal data to form traceable data.

[0115] Store abnormal data on edge devices;

[0116] When the status score of the computer room power distribution equipment is in an abnormal range, or when it is determined to be abnormal data based on the abnormal threshold of various physical characteristics, the abnormal data caching mechanism is triggered, and the amount of cached data is [not specified]. satisfy: ,in, The cache duration before the exception occurs. The duration of the anomaly. The cache duration after the exception ends. This refers to the number of bytes of data per sampling point. This is the current sampling frequency;

[0117] Verify the storage reliability and real-time access of abnormal data.

[0118] Storage reliability verification:

[0119] The fault injection test was used to verify the integrity of the abnormal data in another SSD by simulating a sudden failure of a single SSD. This included comparing the number of data files and content hash values ​​before the failure to confirm that no data was lost.

[0120] If the integrity of the abnormal data in the other SSD is intact, the storage reliability verification passes.

[0121] If the integrity of the abnormal data in the other SSD is incomplete, the storage reliability verification fails, and the SSD is repaired.

[0122] After the fault is repaired, the faulty disk is replaced, and it is verified whether the RAID array automatically rebuilds the image to ensure the reliability of long-term storage.

[0123] Real-time access verification:

[0124] Initiate a retrieval request for abnormal data and record the response time from the request being initiated to the complete presentation of the abnormal data. Through multiple tests, if the response time is consistently within 1 second, the real-time access requirement is met; if the response time is consistently more than 1 second, the real-time access requirement is not met.

[0125] If the response time is consistently within 1 second, the real-time access requirement is met.

[0126] All abnormal and normal data are tagged to create traceable data.

[0127] Normal data obtained from the status scores of the computer room power distribution equipment being in a stable range or passing the abnormal thresholds of various physical characteristics are classified into warm data, intermediate data, and cold data according to storage duration and access frequency and uploaded to the cloud.

[0128] Specific division rules:

[0129] If the normal data storage duration is less than the duration threshold and the normal data access frequency is greater than the access frequency threshold, then the normal data will be classified as warm data.

[0130] If the normal data storage duration is greater than or equal to the duration threshold, and the normal data access frequency is less than or equal to the access frequency threshold, then the normal data will be classified as cold data.

[0131] If the storage duration of normal data is less than the duration threshold and the access frequency is less than or equal to the access frequency threshold, or if the storage duration of normal data is greater than or equal to the duration threshold and the access frequency is greater than the access frequency threshold, then it is classified as intermediate data.

[0132] Upload the temperature data to a high-performance object storage in the cloud, such as AWS S3 Standard or Alibaba Cloud OSS, to support fast query and analysis.

[0133] Migrate cold data and intermediate data, with cold data being stored in a low-cost cloud archive. Examples include AWS S3Glacier and Alibaba Cloud OSS cold archiving.

[0134] For intermediate data, it is first temporarily stored on edge devices or in a cloud transitional storage area. Once the storage period reaches the threshold, it is then migrated to low-cost cloud archiving storage, significantly reducing long-term storage costs.

[0135] All abnormal and normal data are tagged to create traceable data.

[0136] All abnormal and normal data are tagged and uniformly labeled with metadata tags. The metadata tags include: unique device ID, sensor type, data sampling timestamp, reason for sampling frequency adjustment (triggered by status score), and transient filtering results. With these tags, maintenance personnel can quickly locate the power distribution data of any device at any time without manual judgment of data storage location, and achieve full lifecycle data traceability.

[0137] It should be noted that the duration threshold and access frequency threshold mentioned are reference values ​​determined by those skilled in the art based on historical experience; due to differences in equipment, the present invention does not limit their specific values, and the relevant values ​​can be adjusted by those skilled in the art according to the actual equipment.

[0138] For abnormal data, hardware redundancy is built through fault injection testing. In the event of a single disk failure, another SSD can completely retain the abnormal data, preventing the loss of critical data due to hardware failure. After the fault is repaired, the RAID array automatically rebuilds the image to ensure the continuity of long-term storage and provide irreplaceable original evidence for fault tracing and root cause analysis. Through a response time verification mechanism within 1 second, it is ensured that maintenance personnel can quickly retrieve abnormal data after a fault occurs. Based on the hierarchical strategy of access frequency and storage duration, differentiated storage of normal data is achieved, improving the intelligent management level of the data center power distribution system.

[0139] Example 2

[0140] like Figure 2 As shown in Embodiment 1, the present invention provides a remote monitoring system for computer room power distribution equipment, comprising:

[0141] Data acquisition and sampling control module: Deploy edge devices to collect real-time operating data of the data center power distribution equipment, build a status scoring model for the data center power distribution equipment, and construct a dynamic sampling matrix based on the status scoring model;

[0142] High-frequency data transient processing and filtering module: Performs transient feature extraction processing on high-frequency sampled data in edge devices, sets multiple feature thresholds, filters the high-frequency sampled data after transient feature extraction processing, distinguishes abnormal data from normal data, and retains abnormal data;

[0143] Data tiered storage and traceability module: Abnormal data is stored at the edge device, and the storage reliability and access timeliness of the stored abnormal data are verified. Normal data is uploaded to the cloud in tiers, and all abnormal and normal data are tagged to form traceable data.

[0144] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for remote monitoring of power distribution equipment in a computer room, characterized in that: include: S1: Deploy edge devices to collect real-time operating data of data center power distribution equipment, build a status scoring model for data center power distribution equipment, and build a dynamic sampling matrix based on the status scoring model for data center power distribution equipment; The specific process of constructing the status scoring model for the computer room power distribution equipment includes: A status scoring model for computer room power distribution equipment is generated by integrating basic operational data, transient characteristic data, and historical anomaly data. Basic operating data includes: load rate, RMS current, RMS voltage, and ambient temperature and humidity, used to analyze the long-term operating trends of the equipment. Transient characteristic data includes: current fluctuation variance, voltage fluctuation variance, and temperature change rate, which are used to capture initial abnormal fluctuation characteristics; Historical anomaly data includes: the frequency of anomalies in similar equipment and current equipment maintenance cycle data, which are used to predict potential risks; The basic operational data, transient characteristic data, and historical anomaly data are processed separately. The basic operational data processing procedure is as follows: select load rate RMS value of current RMS voltage Ambient temperature Ambient humidity Basic characteristics; For each basic feature, linear normalization is applied, and then a weighted average is taken of the normalized basic features to obtain the comprehensive normalized value of the basic operational data. : ,in, , , , , The weights for each basic feature are obtained by training a neural network using historical fault data, and satisfy the following conditions: , , , , , These are the normalized values ​​of load factor, RMS current, RMS voltage, ambient temperature, and ambient humidity, respectively. The transient characteristic data processing procedure involves selecting the current fluctuation variance. Voltage fluctuation variance Rate of temperature change Transient features are identified; after normalization of each transient feature, a weighted average is taken to obtain the comprehensive normalized value of the transient features. : ,in, , , The transient feature weights are obtained by training a neural network using historical fault data and satisfy the following conditions: , , , These are the normalized values ​​for current fluctuation variance, voltage fluctuation variance, and temperature change rate, respectively. The historical anomaly data processing procedure involves defining historical anomaly factors. The frequency of abnormalities of similar equipment in the past 7 days Compared with the current equipment maintenance cycle percentage Weighted sum: ,in , This is the normalized value of the corresponding indicator; , As weight, satisfying ; Again Inverse normalization means that the more anomalies and the higher the risk, the lower the score should be: ; The specific process of constructing the status scoring model for the computer room power distribution equipment also includes: The median state score was obtained by weighted fusion and normalized weighted processing of basic operational data, transient characteristic data, and historical anomaly data. Then, the median value of the status score is linearly mapped to the [0,100] interval to construct the status score model of the computer room power distribution equipment: ,in, , , The weighting coefficients for basic operational data, transient feature data, and historical anomaly data are respectively obtained through training and optimization using machine learning algorithms. For rounding operation, Based on the comprehensive normalized value of the basic operating data, This is the normalized value of the transient features. The frequency of abnormalities of similar equipment in the past 7 days Compared with the current equipment maintenance cycle percentage The inverse normalized value of the weighted sum; S2: Perform transient feature extraction processing on high-frequency sampled data in the edge device, set multiple feature thresholds, filter the high-frequency sampled data after transient feature extraction processing, distinguish abnormal data from normal data, and retain abnormal data; The specific process for extracting transient features from high-frequency sampled data is as follows: Extracting transient current characteristics: Calculate the current signal over the time interval The rate of change within reflects the intensity of the sudden change in current: ,in, , For a moment , Current value, time interval ; Calculate the abrupt change in current energy by integrating the square of the current over the time interval. The results within reflect the degree of energy release from a short circuit or electric arc: ,in, This refers to the sudden change in current energy. This represents the starting time of the integration time interval. This represents the end of the integration time interval. It is a time-domain function of the current; Simpson's numerical integration method is used for approximate calculation. ,in, For time step, Time interval The total number of current sampling points within and Even number, The summation index ranges from 1 to... , Within the time interval At the (2k-1)th odd-numbered sampling time step, Time interval The sampling time of the 2kth even-numbered step within; Extracting transient temperature features: Temperature gradient, calculating the temperature signal over a time interval The rate of change within reflects the rate of temperature increase and decrease: ,in, , They are time points , Temperature value; Thermal diffusion delay: Calculates the time difference between the peak busbar temperature and the peak sensor temperature, reflecting the heat transfer delay caused by abnormal contact resistance. ,in, Busbar temperature peak time, For sensor to detect temperature Peak time; The specific process for retaining abnormal data is as follows: Record before the anomaly occurs Until the anomaly ends Complete high-frequency data, data length is: ,in, This is the current sampling frequency; S3: Store abnormal data on edge devices and verify the storage reliability and real-time access of the stored abnormal data. Upload normal data to the cloud in a hierarchical manner and label all abnormal and normal data to form traceable data. The storage reliability and real-time access performance of the stored abnormal data are verified. Storage reliability verification: The fault injection test was used to verify the integrity of the abnormal data in another SSD by simulating a sudden failure of a single SSD. This included comparing the number of data files and content hash values ​​before the failure to confirm that no data was lost. If the integrity of the abnormal data in the other SSD is intact, the storage reliability verification passes. If the integrity of the abnormal data in the other SSD is incomplete, the storage reliability verification fails, and the SSD is repaired. After the fault is repaired, the faulty disk is replaced, and it is verified whether the RAID array automatically rebuilds the image to ensure the reliability of long-term storage. Real-time access verification: Initiate a retrieval request for abnormal data and record the response time from the initiation of the request to the complete presentation of the abnormal data. Through multiple tests, if the response time is consistently within 1 second, the real-time access requirement is met; if the response time is consistently more than 1 second, the real-time access requirement is not met. If the response time is consistently within 1 second, the real-time access requirement is met.

2. The method for remote monitoring of computer room power distribution equipment according to claim 1, characterized in that: The specific process for constructing the dynamic sampling matrix is ​​as follows: Based on the status scoring model of computer room power distribution equipment Define three intervals, namely the stable interval. Fluctuation range Abnormal range The interval position of the status score of the computer room power distribution equipment; If the status score of the computer room power distribution equipment is in a stable range, then each sensor will sample according to the basic sampling frequency; If the status score of the computer room power distribution equipment is in the fluctuation range, each sensor will be sampled at twice the basic sampling frequency; If the status score of the power distribution equipment in the computer room is in the abnormal range, each sensor will sample at three times the basic sampling frequency.

3. The method for remote monitoring of computer room power distribution equipment according to claim 1, characterized in that: The specific process for distinguishing between abnormal and normal data is as follows: Data is considered abnormal and data retention is triggered when any of the following conditions are met: current surge intensity. Rate of temperature change Thermal diffusion delay Otherwise, it is considered normal data. Anomaly thresholds for each physical feature: ,in, The threshold for the intensity of the current surge. The threshold values ​​for the rate of temperature increase and decrease. This represents the heat transfer delay threshold.

4. The method for remote monitoring of computer room power distribution equipment according to claim 1, characterized in that: The specific process of storing abnormal data at the edge device is as follows: When the status score of the computer room power distribution equipment is in an abnormal range, or when it is determined to be abnormal data based on the abnormal threshold of various physical characteristics, the abnormal data caching mechanism is triggered, and the amount of cached data is [not specified]. satisfy: ,in, The cache duration before the exception occurs. The duration of the anomaly. The cache duration after the exception ends. This refers to the number of bytes of data per sampling point. This is the current sampling frequency.

5. A remote monitoring system for computer room power distribution equipment, characterized in that: include: Data acquisition and sampling control module: Deploy edge devices to collect real-time operating data of the data center power distribution equipment, build a status scoring model for the data center power distribution equipment, and construct a dynamic sampling matrix based on the status scoring model; The specific process of constructing the status scoring model for the computer room power distribution equipment includes: A status scoring model for computer room power distribution equipment is generated by integrating basic operational data, transient characteristic data, and historical anomaly data. Basic operating data includes: load rate, RMS current, RMS voltage, and ambient temperature and humidity, used to analyze the long-term operating trends of the equipment. Transient characteristic data includes: current fluctuation variance, voltage fluctuation variance, and temperature change rate, which are used to capture initial abnormal fluctuation characteristics; Historical anomaly data includes: the frequency of anomalies in similar equipment and current equipment maintenance cycle data, which are used to predict potential risks; The basic operational data, transient characteristic data, and historical anomaly data are processed separately. The basic operational data processing procedure is as follows: select load rate RMS value of current RMS voltage Ambient temperature Ambient humidity Basic characteristics; For each basic feature, linear normalization is applied, and then a weighted average is taken of the normalized basic features to obtain the comprehensive normalized value of the basic operational data. : ,in, , , , , The weights for each basic feature are obtained by training a neural network using historical fault data, and satisfy the following conditions: , , , , , These are the normalized values ​​of load factor, RMS current, RMS voltage, ambient temperature, and ambient humidity, respectively. The transient characteristic data processing procedure involves selecting the current fluctuation variance. Voltage fluctuation variance Rate of temperature change Transient features are identified; after normalization of each transient feature, a weighted average is taken to obtain the comprehensive normalized value of the transient features. : ,in, , , The transient feature weights are obtained by training a neural network using historical fault data and satisfy the following conditions: , , , These are the normalized values ​​for current fluctuation variance, voltage fluctuation variance, and temperature change rate, respectively. The historical anomaly data processing procedure involves defining historical anomaly factors. The frequency of abnormalities of similar equipment in the past 7 days Compared with the current equipment maintenance cycle percentage Weighted sum: ,in , This is the normalized value of the corresponding indicator; , As weight, satisfying ; Again Inverse normalization means that the more anomalies and the higher the risk, the lower the score should be: ; The specific process of constructing the status scoring model for the computer room power distribution equipment also includes: The median state score was obtained by weighted fusion and normalized weighted processing of basic operational data, transient characteristic data, and historical anomaly data. Then, the median value of the status score is linearly mapped to the [0,100] interval to construct the status score model of the computer room power distribution equipment: ,in, , , The weighting coefficients for basic operational data, transient feature data, and historical anomaly data are respectively obtained through training and optimization using machine learning algorithms. For rounding operation, Based on the comprehensive normalized value of the basic operating data, This is the normalized value of the transient features. The frequency of abnormalities of similar equipment in the past 7 days Compared with the current equipment maintenance cycle percentage The inverse normalized value of the weighted sum; High-frequency data transient processing and filtering module: Performs transient feature extraction processing on high-frequency sampled data in edge devices, sets multiple feature thresholds, filters the high-frequency sampled data after transient feature extraction processing, distinguishes abnormal data from normal data, and retains abnormal data; The specific process for extracting transient features from high-frequency sampled data is as follows: Extracting transient current characteristics: Calculate the current signal over the time interval The rate of change within reflects the intensity of the sudden change in current: ,in, , For a moment , Current value, time interval ; Calculate the abrupt change in current energy by integrating the square of the current over the time interval. The results within reflect the degree of energy release from a short circuit or electric arc: ,in, This refers to the sudden change in current energy. This represents the starting time of the integration time interval. This represents the end of the integration time interval. It is a time-domain function of the current; Simpson's numerical integration method is used for approximate calculation. ,in, For time step, Time interval The total number of current sampling points within and Even number, The summation index ranges from 1 to... , Within the time interval At the (2k-1)th odd-numbered sampling time step, Time interval The sampling time of the 2kth even-numbered step within; Extracting transient temperature features: Temperature gradient, calculating the temperature signal over a time interval The rate of change within reflects the rate of temperature increase and decrease: ,in, , They are time points , Temperature value; Thermal diffusion delay: Calculates the time difference between the peak busbar temperature and the peak sensor temperature, reflecting the heat transfer delay caused by abnormal contact resistance. ,in, Busbar temperature peak time, For sensor to detect temperature Peak time; The specific process for retaining abnormal data is as follows: Record before the anomaly occurs Until the anomaly ends Complete high-frequency data, data length is: ,in, This is the current sampling frequency; Data hierarchical storage and traceability module: Abnormal data is stored at the edge device, and the storage reliability and access time of the stored abnormal data are verified. Normal data is uploaded to the cloud in a hierarchical manner, and all abnormal and normal data are tagged to form traceable data. The storage reliability and real-time access performance of the stored abnormal data are verified. Storage reliability verification: The fault injection test was used to verify the integrity of the abnormal data in another SSD by simulating a sudden failure of a single SSD. This included comparing the number of data files and content hash values ​​before the failure to confirm that no data was lost. If the integrity of the abnormal data in the other SSD is intact, the storage reliability verification passes. If the integrity of the abnormal data in the other SSD is incomplete, the storage reliability verification fails, and the SSD is repaired. After the fault is repaired, the faulty disk is replaced, and it is verified whether the RAID array automatically rebuilds the image to ensure the reliability of long-term storage. Real-time access verification: Initiate a retrieval request for abnormal data and record the response time from the initiation of the request to the complete presentation of the abnormal data. Through multiple tests, if the response time is consistently within 1 second, the real-time access requirement is met; if the response time is consistently more than 1 second, the real-time access requirement is not met. If the response time is consistently within 1 second, the real-time access requirement is met.

Images