Machine room power environment heterogeneous data real-time integration and fault diagnosis method and system

Abstract

The embodiment of the invention provides a machine room power environment heterogeneous data real-time integration and fault diagnosis method and system, and the method comprises the following steps: collecting the current sampling data of each branch circuit of a power distribution cabinet in a machine room, and carrying out the classification marking of the current sampling data according to the current sampling precision grade; constructing a dynamic out-of-limit alarm threshold model based on the classified and marked current sampling data; performing real-time comparison on a current current value according to the dynamic out-of-limit alarm threshold model, and identifying whether an out-of-limit alarm event exists or not; and performing multi-source fusion analysis in combination with the preliminary fault positioning result and other power environment monitoring data in the machine room. Through the scheme of the embodiment of the invention, the problem of how to perform hierarchical regulation and control on the out-of-limit alarm criterion according to the current sampling precision level of the branch circuit of the power distribution cabinet so as to solve the problem of inaccurate fault positioning under a high-load working condition can be solved.

Description

Technical Field

[0001] This application relates to the field of intelligent operation and maintenance technology, specifically to a method and system for real-time integration and fault diagnosis of heterogeneous data of data center power and environment. Background Technology

[0002] In critical power scenarios such as data centers and communication equipment rooms, current monitoring of branch circuits in distribution cabinets is a crucial means of ensuring stable system operation and preventing overload faults. However, existing power environment monitoring systems for data centers face challenges in diagnosing faults under high loads and dynamically changing conditions. Traditional over-limit alarm methods typically use fixed current thresholds, failing to fully consider the differences in the sampling accuracy of current sensors in actual deployments (such as high-precision sensors versus ordinary-precision sensors). This leads to varying sensitivity and reliability in sensing the same electrical event, thus affecting the fairness and effectiveness of alarm criteria. Especially under conditions of drastic load fluctuations or instantaneous impacts (such as server cluster start-up and shutdown), fixed thresholds are prone to false alarms or missed alarms, resulting in ambiguous fault location and an inability to accurately distinguish between genuine overload anomalies and normal transient load changes. Furthermore, existing solutions often focus on over-limit judgments from a single data source, lacking deep integration and confidence-weighted analysis with multi-source heterogeneous monitoring data such as temperature, vibration, and partial discharge. This makes it difficult to achieve closed-loop fault management from "abnormal alarm" to "precise location" and then to "tiered handling." Therefore, there is an urgent need for a fault diagnosis method that can adaptively adjust the alarm threshold based on the current sampling accuracy level, intelligently tolerate load impacts, and integrate multi-source data for confidence assessment, so as to improve the accuracy of fault location and the rationality of response strategies under high-load and complex operating conditions. Summary of the Invention

[0003] In view of this, the present disclosure provides a method and system for real-time integration and fault diagnosis of heterogeneous data of computer room power environment, which at least partially solves the problems existing in the prior art.

[0004] This application discloses a method for real-time integration and fault diagnosis of heterogeneous data on the power environment of a computer room, comprising the following steps: The current sampling data of each branch circuit of the power distribution cabinet in the computer room is collected, and the current sampling data is classified and labeled according to the current sampling accuracy level. High-precision sampling data is labeled as Class H, and ordinary-precision sampling data is labeled as Class N. Based on the categorized and labeled current sampling data, a dynamic over-limit alarm threshold model is constructed. This model adaptively adjusts the alarm threshold according to the current sampling accuracy level and the current load change trend, specifically including: For H-type data, a first threshold range is set, and the threshold boundary is dynamically tightened based on recent load volatility. For N types of data, a second threshold range is set, and a load change trend prediction compensation amount is introduced to adapt to load step changes; If the current load change rate exceeds the preset impact tolerance limit, the instantaneous load impact tolerance window will be activated to temporarily suspend the triggering of the over-limit alarm during the window period. Based on the dynamic over-limit alarm threshold model, the current current value is compared in real time to identify whether there is an over-limit alarm event, and preliminary fault location is performed based on the topological connection relationship of the over-limit branch circuit. Combining the preliminary fault location results with other power and environmental monitoring data within the computer room, a multi-source fusion analysis is performed. This other power and environmental monitoring data includes at least temperature and humidity, equipment vibration, partial discharge, and insulation status data. A weighted fusion strategy is used to assign different confidence weights to data of different accuracy levels, enabling accurate fault identification and graded response processing under high-load conditions. If the confidence level of the fault determination is higher than the first threshold, a level 1 response is triggered, and automatic isolation and alarm are executed. If the confidence level of the fault determination is between the first threshold and the second threshold, a level 2 response is triggered, a maintenance work order is pushed and the enhanced monitoring mode is started. If the confidence level for fault determination is lower than the second threshold, only the event is recorded and the data change trend is continued to be tracked.

[0005] In one specific implementation, constructing a dynamic over-limit alarm threshold model based on the current sampling data after classification and labeling further includes: Obtain the average load volatility α of H-type data at the current moment; The dynamic shrinkage coefficient β is calculated using the formula β = (α × K1) / (T0), where K1 is the coefficient of H-class data and T0 is the sampling period. Adjust the upper limit of the first threshold interval U_H_upper = U_H_initial + β; Adjust the lower limit L_H_lower = L_H_initial - β to form an adaptive alarm boundary.

[0006] In one specific implementation, the introduction of load change trend prediction compensation further includes: Obtain the trend compensation coefficient γ of the current N types of data, defined as the ratio of ΔP / Δt, where ΔP is the load increment and Δt is the time window length; Substituting γ into the compensation formula: C_N = γ × K2, where K2 is the compensation ratio factor for N types of data; The compensation amount is added to the original boundary of the second threshold interval; Set the compensated target threshold U_N_target = U_N_initial + C_N for subsequent alarm comparison.

[0007] In one specific implementation, activating the instantaneous load shock tolerance window further includes: The current load change rate dP_dt = ΔP / Δt is detected. When dP_dt > θ, the impact tolerance mechanism is activated, where θ is a preset impact tolerance limit. Set the tolerance window size to W_window = (dP_dt θ) × t0, where t0 is the default tolerance duration constant; During the window period, all over-limit alarms must be disabled. The alarm judgment logic will be restored to normal after the window period ends.

[0008] In one specific implementation, locating a preliminary fault based on topological connectivity includes: Obtain the connection topology information of each branch circuit of the power distribution cabinet; Fault signal matching is performed for current anomalies on each line; The deviation rate is determined using the formula F = (I_real-I_threshold) / I_base × 100%, where I_real is the measured current, I_threshold is the threshold current, and I_base is the reference current. The deviation rate is used as a fault weighting factor and assigned to different branch circuits.

[0009] In one specific implementation, the multi-source fusion analysis combining other dynamic environment monitoring data further includes: Collect the output values ​​T_ambient and H_level of the temperature and humidity sensors; Obtain the equipment vibration frequency F_vib and the partial discharge signal amplitude P_discharge; The weighted calculation is as follows: W_total = w_T × T_ambient + w_H × H_level + w_vib ×F_vib + w_dis × P_discharge, where w_T, w_H, w_vib, and w_dis represent the confidence weights of the temperature, humidity, vibration, and discharge data, respectively. Combine W_total with the load status of the job scenario to determine whether the fault triggering conditions are met.

[0010] In one specific embodiment, the method further includes: Different confidence weights are assigned to the data in each dimension: w_h, w_n, w_temp, w_vib. The confidence weights are determined based on the statistical characteristics of historical data. The overall fault determination confidence level is calculated according to the formula: D_confidence = (w_h × I_confidence + w_n × N_confidence + w_temp × Temp_conf + w_vib × Vib_conf) × k, where k is the standardization factor. The first-level response is determined by comparing D_confidence with the first threshold T1. If D_confidence is less than T1 and greater than T2, then proceed to the second-level response process.

[0011] In one specific implementation, the first-level response processing includes: The relay control logic is triggered when the confidence level D_confidence > T1; The strength of the isolation action is determined using the formula Isolate = (D_confidence-T1) / (T1-T2) ∈ [0,1]. When Isolate is close to 1, full-path isolation is enabled; otherwise, local isolation mode is enabled. The circuit breaker module sends an isolation command to block the corresponding faulty circuit.

[0012] In one specific implementation, the secondary response processing includes: When the confidence level is between T1 and T2, the work order generation stage begins; Set the enhanced sampling frequency to f_enhanced = f_initial × (1 + λ × log(D_confidence)), where λ is the sampling enhancement factor; Enable high-frequency data collection and log storage to aid in post-event retrospective analysis; The monitoring terminal pushes a maintenance work order notification containing the fault level, location, and related data.

[0013] In one specific implementation, the fault determination confidence level being lower than the second threshold includes: The recording interval is set to τ_recording, and the slope of the current curve is analyzed using a time series algorithm. The short-term current slope is calculated using the formula Slope = (I_current-I_previous) / τ_recording. If the slope exceeds the set threshold σ, the event will be escalated to a concern and added to the alert database. Continuously monitor the load curve changes over the next 30 minutes to identify potential risks.

[0014] This disclosure provides a method for real-time integration and fault diagnosis of heterogeneous data in a data center's power environment, including the following steps: collecting current sampling data from each branch circuit of the power distribution cabinet in the data center, and classifying and labeling the current sampling data according to the current sampling accuracy level, wherein high-precision sampling data is labeled as Class H and ordinary-precision sampling data is labeled as Class N; based on the classified and labeled current sampling data, constructing a dynamic over-limit alarm threshold model, which adaptively adjusts the alarm threshold according to the current sampling accuracy level and the current load change trend, specifically including: for Class H data, setting a first threshold interval and dynamically tightening the threshold boundary according to the recent load fluctuation rate; for Class N data, setting a second threshold interval and introducing a load change trend prediction compensation amount to adapt to load step changes; if the current load change rate exceeds a preset impact tolerance limit, then activating an instantaneous load impact tolerance window, temporarily suspending the triggering of over-limit alarms during the window period. Based on the dynamic over-limit alarm threshold model, the current current value is compared in real time to identify whether an over-limit alarm event exists, and preliminary fault location is performed based on the topological connection relationship of the over-limit branch circuit. Combining the preliminary fault location result with other power and environmental monitoring data in the computer room, multi-source fusion analysis is performed. The other power and environmental monitoring data includes at least temperature and humidity, equipment vibration, partial discharge, and insulation status data. A weighted fusion strategy assigns different confidence weights to data of different accuracy levels to achieve accurate fault determination and graded response processing under high load conditions. Specifically: if the fault determination confidence is higher than the first threshold, a first-level response is triggered, executing automatic isolation and alarm; if the fault determination confidence is between the first and second thresholds, a second-level response is triggered, pushing a maintenance work order and activating enhanced monitoring mode; if the fault determination confidence is lower than the second threshold, only the event is recorded and the data change trend continues to be tracked. The solution of this embodiment can solve the problem of inaccurate fault location under high load conditions by graded adjustment of the over-limit alarm criterion based on the current sampling accuracy level of the distribution cabinet branch circuit. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the exemplary embodiments of this disclosure, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this disclosure and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1This is a flowchart of a method for real-time integration and fault diagnosis of heterogeneous data on the power and environmental conditions of a computer room; Figure 2 This is a flowchart of a system for real-time integration and fault diagnosis of heterogeneous data on the power and environmental conditions of a computer room. Detailed Implementation

[0017] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0018] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0019] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0020] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0021] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0022] This implementation aims to detail the complete operational process of a method for real-time integration and fault diagnosis of heterogeneous data in a data center's power environment. This method, through intelligent data processing and decision-making mechanisms, achieves closed-loop management of the data center's power system, especially the power distribution circuits, from data acquisition and anomaly identification to fault determination and response. Figure 1 As shown, the method of the present invention specifically includes: Step S101: Collect current sampling data of each branch circuit of the power distribution cabinet in the computer room, and classify and label the current sampling data according to the current sampling accuracy level.

[0023] Specifically, firstly, current sensors (such as CT coils or Hall sensors) are installed on the busbars or wires of each branch circuit (such as server rack PDU circuit, air conditioning unit circuit, lighting circuit, etc.) in the computer room power distribution cabinet to collect the effective value or instantaneous value of AC current in real time at a fixed sampling period (e.g., once per second) to form the original current sampling data stream.

[0024] After data is acquired, it is categorized and labeled according to the current sampling accuracy level. Specifically, an accuracy level mapping table can be built into the system. This table associates each data acquisition channel with an accuracy level based on the sensor model, calibration certificate, or preset configuration. For example, loop data using a 0.5-level accuracy sensor is classified as high-precision sampling data; loop data using a 2.5-level accuracy sensor is classified as ordinary-precision sampling data.

[0025] Before the data stream undergoes further transmission and processing, a classification label field is added to each data point. For example, high-precision sampled data is labeled as "Class H," and normal-precision sampled data is labeled as "Class N." This label serves as a meta-attribute of the data.

[0026] Step S102: Based on the current sampling data after classification and labeling, construct a dynamic over-limit alarm threshold model.

[0027] In this invention, for H-type data, an initial first threshold range [L_H_initial, U_H_initial] is set, for example, 80% to 105% of the rated current, and the threshold boundary is dynamically tightened according to the recent load fluctuation rate. The system calculates the standard deviation or average rate of change of H-type current data over the most recent N periods (e.g., the past 5 minutes) as the volatility (α). If the volatility (α) is small, it indicates that the load is stable, so the threshold boundary is slightly tightened to improve monitoring sensitivity; if the volatility (α) increases, the boundary is correspondingly widened to prevent normal fluctuations from triggering alarms. Specifically, this can be achieved through a dynamic contraction coefficient (β), for example: U_H_current = U_H_initial - β × S, L_H_current = L_H_initial + β × S, where S is the adjustment range calculated based on (α).

[0028] For N-type data, a relatively wider second threshold range [L_N_initial, U_N_initial] is set, for example, 75% to 110% of the rated current. To address its relatively low accuracy, a load change trend prediction compensation (C_N) is introduced. The system predicts the load change trend for the next cycle by analyzing short-term historical data (e.g., the past minute) (e.g., obtaining the slope γ through linear fitting). If the predicted load shows an upward trend, the upper limit of the threshold is temporarily increased by a compensation amount C_N = γ × K (K is an empirical coefficient) to tolerate reasonable load step increases and avoid instantaneous false alarms.

[0029] Furthermore, for instantaneous load surges, the load change rate dP / dt of the global or critical loops can be calculated in real time. If dP / dt exceeds the preset surge tolerance limit (θ) (e.g., an increase exceeding 20% ​​per second of the rated value, possibly due to the startup of large equipment), an instantaneous load surge tolerance window is activated. During this window period (e.g., lasting 10 seconds), all threshold-based over-limit alarm judgments are suspended or delayed. After the window period ends, the system re-enables the alarm judgment logic based on the latest, stable load data.

[0030] Step S103: Based on the dynamic over-limit alarm threshold model, the current current value is compared in real time to identify whether there is an over-limit alarm event, and preliminary fault location is performed based on the topological connection relationship of the over-limit branch circuit.

[0031] Specifically, the system compares the current current value collected and marked in step S101 with the dynamic adaptive alarm threshold calculated in step S102 in real time. If the current value continuously (e.g., for more than 3 sampling periods) exceeds the threshold range of its corresponding category, an "over-limit alarm event" is determined to have occurred.

[0032] Once an alarm event is detected, preliminary fault location is immediately initiated. The system invokes a preset power distribution network topology diagram (describing the hierarchical connections between circuit breakers, buses, and branch lines). By analyzing the current status of the alarm point (faulty branch) and its upstream, downstream, and parallel lines, the scope of the fault's impact can be preliminarily inferred. For example, if a branch circuit experiences overcurrent, but the total current of its upstream bus does not increase abnormally, the problem may be a load fault or measurement issue within that branch itself; if the upstream total current also becomes abnormal, the problem may be a power supply issue further upstream or uneven load distribution.

[0033] Step S104: Combine the preliminary fault location results with other power and environmental monitoring data in the computer room to perform multi-source fusion analysis.

[0034] Specifically, the initial location is based solely on current information, which may introduce uncertainties. This step incorporates other power environment monitoring data within the computer room for collaborative analysis to achieve accurate judgment. This data includes at least: temperature and humidity data (from temperature and humidity sensors within the computer room, reflecting the heat dissipation environment), equipment vibration data (from vibration sensors installed on precision air conditioners, UPS, and other equipment, reflecting mechanical conditions), and partial discharge and insulation status data (from UHF or ultrasonic partial discharge detectors, reflecting electrical insulation degradation).

[0035] In this application, different confidence weights (w) are assigned to data from different sources and with different levels of precision. For example, the weight w_H of high-precision (H-class) current data is higher than the weight w_N of ordinary-precision (N-class) current data; the partial discharge signal has the highest weight w_pd when determining insulation faults. A comprehensive fault determination confidence level (D_confidence) is calculated through a weighted decision function (e.g., D = w_H×I_H + w_N×I_N + w_T×T + w_v×V + w_pd×PD, where I is the current anomaly, T is the temperature and humidity anomaly, V is the vibration anomaly, and PD is the discharge intensity).

[0036] In this application, a tiered response process is also implemented.

[0037] Specifically, if the fault determination confidence level is higher than the first threshold (T1) (e.g., D_confidence>0.85), it indicates that the fault is confirmed and urgent. The system immediately triggers a first-level response: it sends a trip command to the corresponding smart circuit breaker through the smart interface to perform automatic isolation, cut off the faulty circuit, and prevent the accident from escalating; at the same time, it sends the highest-level audible and visual alarm and SMS / APP push to the monitoring center.

[0038] If the fault determination confidence level is between the first threshold (T1) and the second threshold (T2) (e.g., 0.6 < D_confidence ≤ 0.85), it indicates that there is a relatively high fault risk but it has not been fully confirmed. The system triggers a secondary response: automatically generate and push a maintenance work order on the operation and maintenance management platform, prompting the operation and maintenance personnel to go for inspection; at the same time, for the suspected fault circuit and its associated devices, start an enhanced monitoring mode, that is, temporarily increase the data sampling frequency and add monitoring dimensions (such as synchronously retrieving videos) to provide more intensive data support for manual decision-making.

[0039] If the fault determination confidence level is lower than the second threshold (T2) (e.g., D_confidence ≤ 0.6), it indicates that the current data anomaly may be caused by instantaneous interference or normal fluctuations. The system does not trigger an active alarm, but completely records all relevant data (such as current values, thresholds, environmental data, confidence levels, etc.) of this over-limit event in the event log library. The system will continue to track the data change trend of this circuit for a subsequent period of time (such as 30 minutes). If the abnormal pattern continues or deteriorates, the system may recalculate and upgrade its confidence level.

[0040] Through the above steps, the present invention realizes a complete closed-loop fault handling from heterogeneous data acquisition, intelligent dynamic warning, preliminary analysis and positioning, to multi-source information fusion and accurate diagnosis, and finally completes automated and hierarchical response, significantly improving the intelligent level and operation and maintenance reliability of the computer room power environment monitoring.

[0041] Next, describe the steps of constructing a dynamic over-limit alarm threshold model based on the current sampling data after classification marking in an embodiment of the present invention. Its purpose is to realize the dynamic adaptive adjustment of the alarm threshold for high-precision (H-class) current data to balance the monitoring sensitivity and anti-interference ability. The specific implementation process is as follows: The first step: Obtain the average load volatility α of the H-class data at the current moment.

[0042] The system maintains a first-in-first-out (FIFO) historical data buffer for each data channel marked as H-class, which is used to store the current sampling values within a recent time window (e.g., the past 300 sampling points, corresponding to 5 minutes of data). Based on the data in this buffer, calculate its average load volatility (α). In this embodiment, the volatility (α) is not simply the standard deviation, but the mean absolute deviation (MAD) after smoothing processing or the coefficient of variation after removing the influence of extreme values, so as to more robustly reflect the stability of the load. For example, calculate the ratio of the mean absolute deviation of the current value within the past 5 minutes to the average value as the volatility (α). The smaller the α value, the more stable the load of this circuit; the larger the α value, the more剧烈 the load fluctuation is.

[0043] Step 2: Calculate the dynamic shrinkage coefficient β according to the formula β = (α × K1) / (T0), where K1 is the coefficient of H-class data and T0 is the sampling period.

[0044] After obtaining the volatility (α), input it into a dedicated dynamic contraction coefficient calculation module. This module calculates β using the following formula: β = (α × K1) / T0 Where: K1 is the H-type data coefficient, a preset empirical constant used to adjust the influence of volatility on threshold adjustment. Its value is usually obtained through historical data analysis and machine learning tuning. For example, in a typical data center scenario, K1 can be set between 0.05 and 0.15. T0 is the sampling period, i.e., the time interval (in seconds) of data collection. For example, if sampling is done once per second, then T0=1. The purpose of introducing T0 is to standardize the calculation result of coefficient β with the sampling period, ensuring consistent adjustment behavior under different sampling frequencies. The design intent of this formula is that the larger the load volatility (α), the larger the calculated contraction coefficient (β), but the overall adjustment magnitude is constrained by K1 and T0, ensuring that the adjustment is smooth and controlled.

[0045] Step 3: Adjust the upper limit of the first threshold interval U_H_upper = U_H_initial - β, and adjust the lower limit L_H_lower = L_H_initial + β.

[0046] The system presets an initial alarm threshold range [L_H_initial, U_H_initial] for H-type data, for example, a lower limit of 85% of the rated current and an upper limit of 105%. The currently effective threshold boundaries are adjusted in real time using a calculated dynamic contraction coefficient (β). Adjust the upper limit: U_H_current = U_H_initial + β Adjust the lower limit: L_H_current = L_H_initial - β This operation means that when the load is stable (small α, small β), the threshold range remains relatively wide initially, focusing on preventing false alarms; while when the load fluctuates more significantly (large α, large β), the threshold range dynamically widens, with the upper and lower limits expanding outwards to form a wider "buffer zone," thereby avoiding frequent alarms caused by normal load fluctuations. Ultimately, an adaptive alarm boundary [L_H_current, U_H_current] that changes in real time with load stability is formed for subsequent real-time data comparison.

[0047] In one embodiment of the present invention, the introduction of load change trend prediction compensation specifically includes: Step 1: Obtain the trend compensation coefficient γ of the current N types of data, defined as the ratio of ΔP / Δt, where ΔP is the load increment and Δt is the time window length.

[0048] The system sets a short-term trend analysis time window (e.g., the most recent 60 seconds) for each of the N types of data channels. Within this window, a continuous current sampling sequence is acquired. First, the load increment (ΔP) within the window is calculated, which is the difference between the average current value at the end of the window and the average current value at the beginning of the window. Meanwhile, the time window length (Δt) is a preset fixed value (in seconds). Then, the trend compensation coefficient (γ) is calculated using the formula: γ = ΔP / Δt The coefficient γ essentially represents the average rate of change in recent load (unit: amperes / second). γ > 0 indicates an upward trend in load; γ < 0 indicates a downward trend in load; γ ≈ 0 indicates a relatively stable load.

[0049] Step 2: Substitute γ into the compensation formula: C_N = γ × K2, where K2 is the compensation ratio factor for N types of data.

[0050] After obtaining the trend compensation coefficient (γ), it is input into the compensation amount generation module. This module calculates the specific compensation amount according to the following formula: C_N = γ × K2 Wherein, K2 is the N-type data compensation scaling factor, a key configurable parameter. K2 determines the system's confidence level or compensation strength in the predicted trend. Its value is typically determined based on the typical pattern of load changes in this type of loop and the alarm delay tolerance time. For example, for an air conditioning compressor loop known to have a slow ramp-up process, K2 might be set to 10; while for certain IT equipment loops with drastic load changes, K2 might be set to 5. The product of K2 and the prediction time window (K2 × Δt) can be understood as the lead time that the system reserves for load changes.

[0051] Step 3: Add the compensation amount to the original boundary of the second threshold interval, and set the compensated target threshold U_N_target = U_N_initial + C_N for subsequent alarm comparison.

[0052] The system presets an initial second threshold range for N types of data. Typically, the upper limit U_N_initial is set to 110% or higher of the rated current, and the lower limit L_N_initial is set to 70% or lower of the rated value. When introducing trend prediction compensation, the upper threshold is mainly adjusted to adapt to the increasing load trend. The compensated target threshold upper limit is: U_N_target = U_N_initial + C_N.

[0053] For example, if the initial upper limit is 110A and C_N is calculated to be +5A, then the current target upper limit is adjusted to 115A. This means that when the system detects that the load is steadily increasing, it will automatically and temporarily raise the alarm upper limit, allowing the current to grow to a higher level for a period of time without triggering false alarms. This compensation is dynamically calculated and updated every trend analysis window (Δt). If the load trend changes to a decrease or stabilizes, C_N will decrease or become zero / negative, and the threshold upper limit will also recover or decrease accordingly, thus maintaining monitoring sensitivity during the load decrease phase. This U_N_target will serve as the basis for the next stage of real-time alarm comparison.

[0054] In one embodiment of the present invention, in order to handle the transient current surges caused by common, short-lived, high-power loads in computer rooms (such as large UPS switching or simultaneous power-on of multiple servers), prevent such non-fault-related surges from triggering unnecessary alarms, and improve the intelligence level of the monitoring system, the present invention allows the activation of a transient load surge tolerance window, specifically including: Step 1: Detect the current load change rate dP_dt = ΔP / Δt. When dP_dt > θ, activate the impact tolerance mechanism, where θ is the preset impact tolerance limit.

[0055] The system monitors the current data of key buses or summary loops (or all loops) in real time and calculates their real-time load change rate (dP_dt). This rate is obtained by calculating the ratio of the change in current (ΔP) within the current sampling period to the sampling interval (Δt): dP_dt = ΔP / Δt, in amperes per second or percentage per second. The system presets an impulse tolerance limit (θ), which represents the boundary between the normal maximum rate of change and abnormal or impulse rates of change that the system can distinguish. The θ value is set according to the characteristics of the power equipment in the computer room; for example, it can be set to 15% of the rated current increase per second. When the system detects that dP_dt > θ, it determines that a transient load impulse event has occurred and immediately activates the impulse tolerance mechanism.

[0056] Step 2: Set the tolerance window size to W_window = (dP_dt θ) × t0, where t0 is the default tolerance duration constant.

[0057] Once an impact event is confirmed, the system does not immediately issue an alarm, but instead opens a dynamic tolerance window. The duration of the window, W_window, is not fixed, but is positively correlated with the severity of the impact, calculated as: W_window = (dP_dt - θ) × t0, where (dP_dt - θ) represents the impact intensity exceeding the baseline tolerance limit. The more violent the impact, the greater the excess value.

[0058] t0 is the default tolerance duration constant (in seconds), for example, set to 2 seconds. This is a basic unit of time.

[0059] The physical meaning of this formula is: for slight exceedances (slightly greater than θ), the window period is very short; for strong shocks (much greater than θ), the system provides a longer tolerance time to allow the shock process to transition smoothly. For example, if dP_dt is 20% / second, θ is 15% / second, and t0 is 2 seconds, then W_window = (20-15) × 2 = 10 seconds. The system immediately starts a tolerance window timer with a duration of 10 seconds.

[0060] Step 3: Disable all over-limit alarms during the window period; restore normal alarm judgment logic after the window period ends. Within the entire window period defined by W_window, the system executes specific alarm suppression logic. Specifically, for all affected loops (which could be the source loop or an associated upstream loop), all threshold comparison-based over-limit alarm triggering functions are suspended or disabled. This means that even if the current value exceeds its dynamically adjusted threshold during this period, the system will not generate or report an alarm event. All data collected within the window period is marked with a special flag and stored normally for post-event analysis.

[0061] When the W_window timer reaches zero, the window period officially ends. The system then resumes normal alarm judgment and triggering logic. After recovery, the system will immediately use the latest current sampling value after the window period ends and compare it with the current dynamic threshold (which may have been updated based on other factors during the window period) to determine whether the load has stabilized at a normal or new level after the impact, or whether there is still a fault overload.

[0062] Next, the steps for locating a preliminary fault based on topological connections according to an embodiment of the present invention will be described. This step aims to utilize the physical connection logic of the power distribution system to map a single current over-limit alarm to a specific fault impact area. The specific implementation process is as follows: Step 1: Obtain the connection topology information of each branch circuit of the power distribution cabinet.

[0063] The system internally acquires complete connection topology information of the power distribution cabinets in the computer room, either built-in or synchronously from external management systems (such as BIM or DCIM systems). This information is stored as structured data (such as XML, JSON, or graph database node relationships), accurately describing the connection relationships between various electrical components. Its core content includes: the main incoming circuit breaker, busbars at all levels, distribution circuit breakers (air switches or fuses), and the hierarchical relationships of the final load branch circuits. For example, a typical topology record is: “Bus_BusA” --Power Supply--> “Circuit Breaker_CB201” --Connection--> “Branch Circuit_Line_Server_Rack_05”. The system builds and maintains this topology graph model in memory for real-time querying.

[0064] Step 2: Use the formula F = (I_real I_threshold) / I_base × 100% to determine the deviation rate, where I_real is the measured current, I_threshold is the threshold current, and I_base is the reference current.

[0065] When a branch circuit (denoted as Line_X) triggers a current over-limit alarm, the system activates the location analysis engine. First, it confirms that Line_X's own current I_real continuously exceeds its dynamic threshold I_threshold. Next, it calculates the current deviation rate (F) of that line: F = (I_real - I_threshold) / I_base × 100%, where I_real is the measured current value of the line; I_threshold is the currently applied dynamic alarm threshold (upper limit) for the line; and I_base is the reference current for the line, typically selected as its rated current or long-term historical average operating current. This denominator normalizes the absolute difference to a relative percentage, making circuits of different specifications comparable.

[0066] The calculated F value intuitively reflects the severity of the exceedance. For example, F=15% means that the measured current exceeds the threshold by 15% of the reference value.

[0067] Step 3: Assign the deviation rate as a fault weighting factor to different branch circuits.

[0068] The system takes alarm point Line_X as the starting point, performs traversal analysis on its topology map, and allocates the calculated deviation rate F as a core fault weighting factor: Direct assignment: The alarm point Line_X itself receives the highest fault weight, which can be denoted as Weight(Line_X) = F.

[0069] Upstream backtracking: The system traces upstream along the topology to find the power supply upstream of Line_X (such as the directly powered circuit breaker CB_X, and the even upstream bus Bus_A). The fault weights of these upstream nodes are aggregated based on the conditions of all abnormal branches downstream. For example, the weight of an upstream node may be equal to the weighted sum of the deviation rates of all its alarm sub-circuit circuits. If the weight of a certain upstream node also increases significantly, it suggests that the fault impact may be larger.

[0070] Parallel Analysis: Simultaneously, the system checks parallel sibling circuits under the same upstream node as Line_X. If the current of these sibling circuits is normal, the fault is most likely due to a load fault in Line_X itself. If multiple parallel circuits simultaneously exceed their limits, the fault point may be upstream of them (such as poor contact in the distribution module or bus).

[0071] Finally, the system generates a preliminary fault location report, which lists the suspected fault points (such as Line_X) and their associated components, and marks each point with a calculated "fault probability weight", thereby guiding maintenance personnel to prioritize the location with the highest weight.

[0072] Next, the steps of multi-source fusion analysis combining other power environment monitoring data according to an embodiment of the present invention will be described. The core of this step lies in constructing a multi-dimensional fault determination system by integrating heterogeneous environmental and state sensor data, greatly improving the accuracy and reliability of fault diagnosis. The specific process is as follows: Step 1: Collect the output values ​​T_ambient and H_level of the temperature and humidity sensor; obtain the device vibration frequency F_vib and the amplitude of the partial discharge signal P_discharge.

[0073] While the current data triggers an over-limit alarm and performs preliminary location analysis, the system automatically triggers an enhanced data acquisition command for the associated area, synchronously acquiring at least the following four types of real-time data from the computer room monitoring network: Temperature and humidity data: The current ambient temperature T_ambient (unit: °C) and relative humidity H_level (unit: %RH) are read from sensors deployed near the alarm cabinet or in the corresponding air conditioning supply / return air duct.

[0074] Equipment vibration data: Read the effective values ​​of vibration acceleration or velocity from vibration sensors installed on associated precision air conditioning compressors, UPS internal fans, or generator sets, and further analyze their dominant frequency F_vib (unit: Hz). Abnormal frequencies (such as power frequency octaves) may indicate mechanical imbalance or bearing wear.

[0075] Partial discharge data: Using ultra-high frequency (UHF) or ultrasonic (AE) sensors, collect partial discharge signals from switch contacts, busbar connections, or cable terminals in the distribution cabinet, and obtain the quantized value P_discharge (such as dBμV or pulses / second) of the pulse amplitude or frequency.

[0076] Step 2: Calculate by weighting: W_total = w_T × T_ambient + w_H × H_level + w_vib × F_vib + w_dis × P_discharge, where w represents the respective confidence weight.

[0077] First, the raw values ​​from sensors with different dimensions are normalized to a scale range of [0, 1] or [0, 100], where 0 represents complete normality and a larger value indicates a higher degree of abnormality. For example, temperature values ​​can be linearly or non-linearly mapped based on a set threshold (e.g., 25°C) and an alarm threshold (e.g., 30°C).

[0078] Then, the mixture is fused according to the following formula: W_total = w_T × T_ambient_norm + w_H × H_level_norm + w_vib × F_vib_norm + w_dis × P_discharge_norm Here, w_T, w_H, w_vib, and w_dis represent the confidence weights of temperature, humidity, vibration, and discharge data, respectively. These weights are set based on the different data types' ability to characterize specific faults. For example, w_T has a higher weight when determining overload caused by poor heat dissipation; while w_dis has the highest weight when determining internal insulation faults in switchgear.

[0079] Step 3: Combine W_total with the load status of the job scenario to determine whether the fault triggering conditions are met.

[0080] The calculated comprehensive anomaly index W_total is combined with the current system load status (e.g., high load, light load at night, or system testing phase) and preliminary fault location information to make a final judgment. The system has preset fault trigger condition thresholds for different scenarios. For example: Under high load conditions, if W_total exceeds a relatively high threshold and the current exceeds the limit for a long period of time, it is judged as an "overload fault that needs to be dealt with immediately".

[0081] Under light load conditions, if the current slightly exceeds the limit but P_discharge_norm (discharge data) is abnormally high, it may be judged as "insulation degradation fault", which is also a very high risk level.

[0082] The output of this step is a fault diagnosis conclusion with enhanced confidence. It associates current anomalies with environmental anomalies, effectively distinguishing between "true overload" (accompanied by temperature rise) and "false overload" (such as measurement fault), or identifying complex faults such as "mechanical fault causing current anomaly" (accompanied by specific vibration).

[0083] Next, another embodiment of the method of the present invention will be described. It specifically includes the following steps: Step 1: Assign different confidence weights to the data in each dimension: w_h, w_n, w_temp, w_vib. The confidence weights are determined based on the statistical characteristics of historical data.

[0084] The system pre-determines a basic confidence weight for each data dimension involved in the final judgment. These weights are not assigned arbitrarily, but are determined based on the statistical characteristics of historical data. For example: High-precision current data confidence weight (w_h): By analyzing H-type data of long-term series, its measurement stability and correlation coefficient with the real load are evaluated and assigned a higher weight (e.g., 0.35).

[0085] Confidence weight (w_n) for ordinary precision current data: It is given a low weight (e.g., 0.20) based on its relatively high noise and possible bias.

[0086] Temperature data confidence weight (w_temp): set according to the strength of the causal relationship between temperature rise and electrical overload fault and the historical consistency (e.g., 0.25).

[0087] Vibration data confidence weight (w_vib): set according to the probability of a specific mechanical fault mode and current anomaly occurring simultaneously (e.g., 0.20).

[0088] The weights are typically summed to 1, reflecting the relative importance of each piece of evidence in the decision-making process. These weights can be optimized through offline machine learning, such as classifiers trained on historical failure cases.

[0089] Step 2: Calculate the overall fault determination confidence level D_confidence = (w_h × I_confidence + w_n × N_confidence + w_temp × Temp_conf + w_vib × Vib_conf) × k, where k is the standardization factor.

[0090] In each diagnostic cycle, the system collects real-time data from various dimensions and their normalized abnormality scores (for example, the percentage of current exceeding the limit, the percentage of temperature rise exceeding the threshold, etc., which are all converted to scores from 0 to 100). Then, weighted synthesis is performed according to the following formula: D_confidence = (w_h × I_confidence_h + w_n × I_confidence_n + w_temp × Temp_conf + w_vib × Vib_conf) × k, where I_confidence_h, I_confidence_n, Temp_conf, and Vib_conf are the real-time abnormality confidence scores of high-precision current, ordinary-precision current, temperature, and vibration data respectively.

[0091] k is a normalization factor, whose role is to ensure that the finally calculated D_confidence falls within a preset range (such as 0 to 100), facilitating comparison with a fixed threshold. k can be a simple scaling factor or a more complex function used to compensate for sensitivity differences under different working conditions.

[0092] Step 3: Compare D_confidence with the first threshold T1 to judge the first-level response; if D_confidence is less than T1 and greater than T2, then enter the second-level response process.

[0093] The system presets two key confidence thresholds: the first threshold (T1) and the second threshold (T2), and T1 > T2 (for example, T1 = 80, T2 = 60).

[0094] First-level response judgment: Compare the calculated D_confidence with the first threshold T1 in real time. If D_confidence > T1, it is determined as a high-confidence fault, and the system immediately enters the first-level response process, triggering automatic isolation and the highest-level alarm.

[0095] Second-level response judgment: If D_confidence is less than or equal to T1 and greater than T2 (that is, T2 < D_confidence ≤ T1), it is determined as a medium-confidence fault or a potential high risk. The system enters the second-level response process, generates a maintenance work order, and starts enhanced monitoring.

[0096] Continuous tracking: If D_confidence ≤ T2, it is determined that the current evidence is insufficient or only a low-risk abnormality. The system does not trigger active handling, but records the event in the database and continuously tracks the data trend. This quantitative decision-making mechanism ensures that the operation and maintenance response resources can accurately match the actual risk level of the fault.

[0097] For Level 1 response handling, the system immediately sends a command to the intelligent power distribution management unit to activate the relay control logic associated with the faulty circuit. This logic is not a simple "on / off" signal, but a composite control command that includes parameters such as fault location and isolation strength. Simultaneously, the system locks the state of the faulty circuit, preparing for isolation.

[0098] Before triggering isolation, the system first calculates a dynamic isolation force coefficient (Isolate), which determines the thoroughness and selectivity of the isolation. The calculation is based on the following formula: Isolate = (D_confidence - T1) / (T1 - T2) The numerator (D_confidence - T1) represents the excess risk that the current fault confidence exceeds the emergency threshold T1. The greater the excess, the higher the certainty of the fault or the greater the risk.

[0099] Denominator (T1 - T2): is the preset confidence decision interval width, which standardizes the excess risk.

[0100] The calculated Isolate value is strictly limited to the range [0,1]. Its magnitude directly reflects the strength of the required isolation measures.

[0101] Then, differentiated isolation strategies are implemented based on the level of isolation required. The system performs tiered isolation operations based on the calculated Isolate value: Strong Isolation Mode (Isolate close to 1, e.g., ≥ 0.8): When the fault confidence level is much higher than T1, it indicates that the fault is certain and severe. The system enables full-path isolation. This means that control commands will not only disconnect the intelligent circuit breaker (such as a miniature circuit breaker MCB) of the faulty branch circuit itself, but also, as needed, sequentially or simultaneously disconnect the upstream distribution circuit breaker (such as a molded case circuit breaker MCCB), and even logically isolate the associated bus tie switches, ensuring that the fault point is completely isolated from all possible power supply sides, forming an absolute electrical island, preventing the fault from spreading or requiring circuitous power supply.

[0102] Selective Isolation Mode (lower Isolate value, e.g., <0.8): When the fault confidence level just exceeds T1, the risk is relatively clear but may be minor. The system activates a partial isolation mode. In this mode, the system prioritizes and only operates the nearest downstream smart circuit breaker to the fault point. For example, only the final circuit breaker supplying power to the cabinet is disconnected, while its upstream switch remains closed. This minimizes the scope of the power outage, affecting only the identified faulty loads and protecting the continued operation of other normal loads on the same busbar.

[0103] Finally, the isolation decision module sends precise isolation instructions to the corresponding intelligent circuit breaker or communication relay module through standard industrial communication protocols (such as Modbus TCP, IEC 61850 MMS). This instruction includes the target device address, operation command ("switch off"), and necessary security verification codes. After receiving the instruction, the execution unit drives the internal operating mechanism to physically disconnect the contacts and block the current path of the corresponding faulty circuit.

[0104] When the system determines that the fault confidence level D_confidence satisfies the condition T2 < D_confidence ≤ T1 (for example, T2 = 60, T1 = 80), it automatically enters the secondary response process. First, the work order generation engine is triggered. This engine automatically extracts relevant information from the system and structurally generates a maintenance work order. The work order content includes at least: fault event ID, trigger time, suspected fault location (based on topological positioning), snapshots of current monitoring data (such as current, temperature, etc.), calculated confidence value, and associated device asset code. Subsequently, the system pushes this work order to the operation and maintenance management platform (such as the ITSM system), mobile APP, or SMS platform in real time through the integration interface to notify the designated operation and maintenance engineer or team, clarify the alarm location and event nature, and guide them to go to the site or conduct remote verification.

[0105] While pushing the work order, the system immediately starts the enhanced monitoring mode for the suspected faulty circuit and associated devices. In this mode, the system dynamically adjusts according to the current risk level (i.e., D_confidence). Specifically, it is calculated according to the following formula: f_enhanced = f_initial × (1 + λ × log(D_confidence)) f_initial: is the normal monitoring sampling frequency of this circuit (such as 1 time / second).

[0106] λ: is the sampling enhancement factor, an adjustable parameter (such as λ = 2), which controls the intensity of frequency increase.

[0107] log(D_confidence): uses the logarithmic function to ensure that the frequency increases with the confidence level, but the increase curve is smooth to avoid a sharp increase in resources. For example, when D_confidence = 70, the calculated frequency may increase to 2 - 5 times / second.

[0108] In addition, the system uses f_enhanced to collect high-frequency data on the target circuit's current, voltage, and associated temperature, humidity, and vibration sensors. The acquired raw waveforms or high-density sampled values ​​are synchronously stored in a dedicated high-performance log database or time-series database.

[0109] All data collected in enhanced monitoring mode is stored in a complete and organized log. This provides a valuable data foundation for post-fault retrospective analysis. Once a subsequent fault develops into a Level 1 event, or an engineer confirms the problem on-site, this high-frequency data can be used to accurately analyze the fault's evolution and capture transient anomalies, thereby deepening the understanding of the root cause of the fault. The stored data index is associated with previously generated maintenance work order IDs for easy retrieval.

[0110] Finally, the system will push composite alarm information, including the fault level, precise location (e.g., circuit A of distribution cabinet 3, circuit of cabinet 5), and relevant multi-source data (current value, trend chart), in the form of pop-ups, list highlights, or dedicated panels. This allows monitoring personnel to obtain a comprehensive situational awareness of potential faults without leaving their posts, thereby making more efficient dispatching decisions.

[0111] When the fault determination confidence level D_confidence is lower than or equal to the second threshold T2 (e.g., T2=60), the system determines that there is no urgent or high-risk fault. However, this does not mean ignoring it. The system sets a periodic recording interval τ_recording for this event (e.g., recording summary data every 30 seconds). Simultaneously, a dedicated time-series tracking and analysis thread is started to continuously monitor and analyze the current data of this loop (primarily the data source that triggered the event).

[0112] Specifically, at each recording point (or a shorter internal calculation cycle), the system calculates the short-term current slope according to the following formula: Slope = (I_current - I_previous) / τ_recording I_current: The current value at the current sampling moment.

[0113] I_previous: The current value at the previous recording interval (before τ_recording).

[0114] τ_recording: The time difference, in seconds.

[0115] The calculated slope (unit: amperes per second or percentage per second) visually reflects the average rate of change of current over a short period of time. A positive and large slope indicates that the current is rising rapidly; a negative value indicates a decline; and a value close to zero indicates stability.

[0116] The system compares the calculated slope with a preset trend concern threshold σ. σ is an empirical value set based on loop characteristics and historical stable operation data (e.g., the rate of change per second corresponding to 0.5% / minute of rated current).

[0117] If the absolute value of Slope exceeds a set threshold σ (e.g., |Slope|>σ), it indicates a noteworthy abnormal trend in the current (whether a sudden increase or decrease). In this case, the system escalates this event from a regular log entry to a "priority event." The system creates or updates a priority event record and adds it to a dedicated alert database or priority list. This record is associated with all the contextual information of the initial event and begins to flag its trend data.

[0118] If the slope does not exceed σ, the current low-risk recording status is maintained, and only regular snapshots are recorded at intervals.

[0119] For events marked as "objects of concern," the system initiates an extended tracking period (e.g., continuous tracking for the next 30 minutes). During this period, the system continues to monitor the current, synchronizes associated environmental data (temperature, etc.) at a lower frequency, and continuously calculates and records changes in the slope value. By analyzing the load curve changes over these 30 minutes, the system can identify various potential risk patterns, such as: Slowly rising type: The slope remains positive with small increments, which may lead to the alarm threshold being reached after a period of time.

[0120] Step-hold type: The current remains high after a sudden increase, although it has not exceeded the dynamic threshold, it is already on the verge of high risk.

[0121] Increased periodicity: The amplitude or frequency of fluctuations increases.

[0122] The system summarizes these trend analysis results and updates them to the event log. If the trend deteriorates during the tracking period, the confidence level is recalculated, and it may exceed T2, the event will automatically escalate, triggering a higher-level response. This mechanism enables "long-term monitoring" and intelligent early warning of emerging risks.

[0123] In addition, such as Figure 2 As shown, this application also provides a real-time integration and fault diagnosis system 200 for heterogeneous data of computer room power environment, the system comprising: The data acquisition module 201 collects current sampling data from each branch circuit of the power distribution cabinet in the computer room, and classifies and labels the current sampling data according to the current sampling accuracy level. High-precision sampling data is labeled as Class H, and ordinary-precision sampling data is labeled as Class N. Model building module 202 constructs a dynamic over-limit alarm threshold model based on the classified and labeled current sampling data. This model adaptively adjusts the alarm threshold according to the current sampling accuracy level and the current load change trend, specifically including: For H-type data, a first threshold range is set, and the threshold boundary is dynamically tightened based on recent load volatility. For N types of data, a second threshold range is set, and a load change trend prediction compensation amount is introduced to adapt to load step changes; If the current load change rate exceeds the preset impact tolerance limit, the instantaneous load impact tolerance window will be activated to temporarily suspend the triggering of the over-limit alarm during the window period. The comparison module 203 compares the current current value in real time according to the dynamic over-limit alarm threshold model, identifies whether there is an over-limit alarm event, and performs preliminary fault location based on the topological connection relationship of the over-limit branch circuit. Monitoring module 204, combining the preliminary fault location results with other power and environmental monitoring data within the computer room, performs multi-source fusion analysis. This other power and environmental monitoring data includes at least temperature and humidity, equipment vibration, partial discharge, and insulation status data. A weighted fusion strategy assigns different confidence weights to data of different accuracy levels, enabling accurate fault determination and graded response processing under high-load conditions. If the confidence level of the fault determination is higher than the first threshold, a level 1 response is triggered, and automatic isolation and alarm are executed. If the confidence level of the fault determination is between the first threshold and the second threshold, a level 2 response is triggered, a maintenance work order is pushed and the enhanced monitoring mode is started. If the confidence level for fault determination is lower than the second threshold, only the event is recorded and the data change trend is continued to be tracked.

[0124] The functions and implementation methods of each module of the computer room power environment heterogeneous data real-time integration and fault diagnosis system 200 of this application have been described above with reference to the attached drawings, and will not be repeated here.

[0125] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for real-time integration and fault diagnosis of heterogeneous data on power and environmental conditions in a computer room, characterized in that, Includes the following steps: The current sampling data of each branch circuit of the power distribution cabinet in the computer room is collected, and the current sampling data is classified and labeled according to the current sampling accuracy level. High-precision sampling data is labeled as Class H, and ordinary-precision sampling data is labeled as Class N. Based on the categorized and labeled current sampling data, a dynamic over-limit alarm threshold model is constructed. This model adaptively adjusts the alarm threshold according to the current sampling accuracy level and the current load change trend, specifically including: For H-type data, a first threshold range is set, and the threshold boundary is dynamically tightened based on recent load volatility. For N types of data, a second threshold range is set, and a load change trend prediction compensation amount is introduced to adapt to load step changes; If the current load change rate exceeds the preset impact tolerance limit, the instantaneous load impact tolerance window will be activated to temporarily suspend the triggering of the over-limit alarm during the window period. Based on the dynamic over-limit alarm threshold model, the current current value is compared in real time to identify whether there is an over-limit alarm event, and preliminary fault location is performed based on the topological connection relationship of the over-limit branch circuit. Combining the preliminary fault location results with other power and environmental monitoring data within the computer room, a multi-source fusion analysis is performed. This other power and environmental monitoring data includes at least temperature and humidity, equipment vibration, partial discharge, and insulation status data. A weighted fusion strategy is used to assign different confidence weights to data of different accuracy levels, enabling accurate fault identification and graded response processing under high-load conditions. If the confidence level of the fault determination is higher than the first threshold, a level 1 response is triggered, and automatic isolation and alarm are executed. If the confidence level of the fault determination is between the first threshold and the second threshold, a level 2 response is triggered, a maintenance work order is pushed and the enhanced monitoring mode is started. If the confidence level for fault determination is lower than the second threshold, only the event is recorded and the data change trend is continued to be tracked.

2. The method for real-time integration and fault diagnosis of heterogeneous data of computer room power environment according to claim 1, characterized in that, The dynamic over-limit alarm threshold model constructed based on the current sampling data after classification and labeling further includes: Obtain the average load volatility α of H-type data at the current moment; The dynamic shrinkage coefficient β is calculated using the formula β = (α × K1) / (T0), where K1 is the coefficient of H-class data and T0 is the sampling period. Adjust the upper limit of the first threshold interval U_H_upper = U_H_initial + β; Adjust the lower limit L_H_lower = L_H_initial - β to form an adaptive alarm boundary.

3. The method for real-time integration and fault diagnosis of heterogeneous data of computer room power environment according to claim 1, characterized in that, The introduced load change trend prediction compensation amount further includes: Obtain the trend compensation coefficient γ of the current N types of data, defined as the ratio of ΔP / Δt, where ΔP is the load increment and Δt is the time window length; Substituting γ into the compensation formula: C_N = γ × K2, where K2 is the compensation ratio factor for N types of data; The compensation amount is added to the original boundary of the second threshold interval; Set the compensated target threshold U_N_target = U_N_initial + C_N for subsequent alarm comparison.

4. The method for real-time integration and fault diagnosis of heterogeneous data of computer room power environment according to claim 1, characterized in that, Activating the instantaneous load shock tolerance window further includes: The current load change rate dP_dt = ΔP / Δt is detected. When dP_dt > θ, the impact tolerance mechanism is activated, where θ is a preset impact tolerance limit. Set the tolerance window size to W_window = (dP_dt θ) × t0, where t0 is the default tolerance duration constant; During the window period, all over-limit alarms must be disabled. The alarm judgment logic will be restored to normal after the window period ends.

5. The method for real-time integration and fault diagnosis of heterogeneous data of computer room power environment according to claim 1, characterized in that, Preliminary fault location based on topological connectivity includes: Obtain the connection topology information of each branch circuit of the power distribution cabinet; Fault signal matching is performed for current anomalies on each line; The deviation rate is determined using the formula F = (I_real-I_threshold) / I_base × 100%, where I_real is the measured current, I_threshold is the threshold current, and I_base is the reference current. The deviation rate is used as a fault weighting factor and assigned to different branch circuits.

6. The method for real-time integration and fault diagnosis of heterogeneous data of computer room power environment according to claim 1, characterized in that, Multi-source fusion analysis, combined with other power environment monitoring data, further includes: Collect the output values ​​T_ambient and H_level of the temperature and humidity sensors; Obtain the equipment vibration frequency F_vib and the partial discharge signal amplitude P_discharge; The weighted calculation is as follows: W_total = w_T×T_ambient + w_H × H_level + w_vib × F_vib + w_dis × P_discharge, where w_T, w_H, w_vib, and w_dis represent the confidence weights of the temperature, humidity, vibration, and discharge data, respectively. Combine W_total with the load status of the job scenario to determine whether the fault triggering conditions are met.

7. The method for real-time integration and fault diagnosis of heterogeneous data of computer room power environment according to claim 6, characterized in that, The method further includes: Different confidence weights are assigned to the data in each dimension: w_h, w_n, w_temp, w_vib. The confidence weights are determined based on the statistical characteristics of historical data. The overall fault determination confidence level D_confidence is calculated according to the formula: D_confidence = (w_h × I_confidence + w_n × N_confidence + w_temp × Temp_conf + w_vib × Vib_conf) × k, where k is the standardization factor. The first-level response is determined by comparing D_confidence with the first threshold T1. If D_confidence is less than T1 and greater than T2, then proceed to the second-level response process.

8. The method for real-time integration and fault diagnosis of heterogeneous data of computer room power environment according to claim 7, characterized in that, The first-level response processing includes: The relay control logic is triggered when the confidence level D_confidence > T1; The strength of the isolation action is determined using the formula Isolate = (D_confidence-T1) / (T1-T2) ∈ [0,1]. When Isolate is close to 1, full-path isolation is enabled; otherwise, local isolation mode is enabled. The circuit breaker module sends an isolation command to block the corresponding faulty circuit.

9. The method for real-time integration and fault diagnosis of heterogeneous data of computer room power environment according to claim 8, characterized in that, The secondary response processing includes: When the confidence level is between T1 and T2, the work order generation stage begins; Set the enhanced sampling frequency to f_enhanced = f_initial × (1 + λ × log(D_confidence)), where λ is the sampling enhancement factor; Enable high-frequency data collection and log storage to aid in post-event retrospective analysis; The monitoring terminal pushes a maintenance work order notification containing the fault level, location, and related data.

10. The method for real-time integration and fault diagnosis of heterogeneous data of computer room power environment according to claim 9, characterized in that, When the fault determination confidence level is lower than the second threshold, it includes: The recording interval is set to τ_recording, and the slope of the current curve is analyzed using a time series algorithm. The short-term current slope is calculated using the formula Slope = (I_current-I_previous) / τ_recording. If the slope exceeds the set threshold σ, the event will be escalated to a concern and added to the alert database. Continuously monitor the load curve changes over the next 30 minutes to identify potential risks.

Images