Search

CN-121978587-A - Real-time integration and fault diagnosis method and system for heterogeneous data of machine room power environment

CN121978587ACN 121978587 ACN121978587 ACN 121978587ACN-121978587-A

Abstract

The embodiment of the disclosure provides a real-time integration and fault diagnosis method and system for heterogeneous data of a power environment of a machine room, wherein the method comprises the following steps of collecting current sampling data of all branch circuits of a power distribution cabinet in the machine room, and classifying and marking the current sampling data according to current sampling precision grades; the method comprises the steps of classifying current sampling data after marking, constructing a dynamic out-of-limit alarm threshold model, comparing current values in real time according to the dynamic out-of-limit alarm threshold model, identifying whether out-of-limit alarm events exist or not, and carrying out multi-source fusion analysis by combining a preliminary fault positioning result with other power environment monitoring data in a machine room. According to the scheme of the embodiment of the disclosure, the problem of inaccurate fault positioning under a high-load working condition can be solved by means of grading 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.

Inventors

  • MA JIANWEI

Assignees

  • 上海耀腾信息科技发展有限公司

Dates

Publication Date
20260505
Application Date
20260107

Claims (10)

  1. 1. The real-time integration and fault diagnosis method for heterogeneous data of a machine room power environment is characterized by comprising the following steps of: collecting current sampling data of each branch circuit of a power distribution cabinet in a machine room, and classifying and marking the current sampling data according to the current sampling precision grade, wherein the high-precision sampling data are marked as H types, and the common precision sampling data are marked as N types; based on the current sampling data after classification marking, a dynamic out-of-limit alarm threshold model is constructed, and the model self-adaptively adjusts the alarm threshold according to the current sampling precision level and the current load change trend, and specifically comprises the following steps: Setting a first threshold interval for H-class data, and dynamically tightening a threshold boundary according to the recent load fluctuation rate; Setting a second threshold interval for N types of data, and introducing a load change trend prediction compensation quantity to adapt to load step change; If the current load change rate exceeds the preset impact tolerance limit, starting an instantaneous load impact tolerance window, and suspending out-of-limit alarm triggering in a window period; according to the dynamic out-of-limit alarm threshold model, comparing the current value in real time, identifying whether an out-of-limit alarm event exists or not, and carrying out preliminary fault positioning according to the topological connection relation of the out-of-limit branch circuit; And combining the preliminary fault positioning result with other power environment monitoring data in the machine room, performing multisource fusion analysis, wherein the other power environment monitoring data at least comprise temperature and humidity, equipment vibration, partial discharge and insulation state data, giving different confidence weights to data with different precision grades through a weighted fusion strategy, and finishing accurate judgment and grading response processing of faults under high-load working conditions, wherein the steps of: If the fault judgment confidence is higher than a first threshold, triggering a first-level response, and executing automatic isolation and alarming; If the fault judgment confidence is between the first threshold and the second threshold, triggering a second-level response, pushing an overhaul work order and starting an enhanced monitoring mode; if the fault determination confidence is lower than the second threshold, only the event is recorded and the data change trend is continuously tracked.
  2. 2. The method for real-time integration and fault diagnosis of heterogeneous data of a power environment of a machine room according to claim 1, wherein constructing a dynamic out-of-limit alarm threshold model based on the current sampling data after classification marking further comprises: acquiring the average load fluctuation rate alpha of class H data at the current moment; Calculating a dynamic contraction coefficient beta according to a formula beta= (alpha multiplied by K1)/(T0), wherein K1 is a data coefficient of class H, and T0 is a sampling period; Adjusting an upper limit u_h_upper=u_h_initial+β of the first threshold interval; The lower limit l_h_lower=l_h_initial- β is adjusted to form an adaptive alert boundary.
  3. 3. The method for real-time integration and fault diagnosis of heterogeneous data of a power environment of a machine room according to claim 1, wherein the introducing the predicted compensation amount of the load variation trend further comprises: Acquiring a trend compensation coefficient gamma of the current N-type data, and defining the trend compensation coefficient gamma as the ratio of delta P/delta t, wherein delta P is a load increment, and delta t is a time window length; bringing gamma into a compensation quantity formula of C_N=gamma×K2, wherein K2 is N-class data compensation scale factors; adding the compensation amount to the original limit of the second threshold interval; a compensated target threshold U _ N _ target = U _ N _ initial + C _ N is set, for subsequent alarm comparison.
  4. 4. The method for real-time integration and fault diagnosis of machine room dynamic environment heterogeneous data according to claim 1, wherein starting the transient load shock tolerance window further comprises: Detecting a current load change rate dp_dt=Δp/Δt, and activating an impact tolerance mechanism when dp_dt > θ, wherein θ is a preset impact tolerance limit; setting the tolerance window size as W_window= (dP_dt theta) x t0, wherein t0 is a default tolerance duration constant; Prohibiting all overrun alarm triggers in a window period; and after the window period is finished, the normal alarm judgment logic is restored.
  5. 5. The method for real-time integration and fault diagnosis of heterogeneous data of a power environment of a machine room according to claim 1, wherein locating the preliminary fault according to the topological connection relationship comprises: obtaining connection topology information of each branch circuit of a power distribution cabinet; Performing fault signal matching on the abnormal current of each line; The bias rate is determined using the formula f= (i_real-i_threshold)/i_base x 100%, where i_real is the measured current, i_threshold is the threshold current, i_base is the reference current; The deviation rate is assigned to different branch circuits as a fault weight factor.
  6. 6. The method for real-time integration and fault diagnosis of heterogeneous data of a power environment of a machine room according to claim 1, wherein the multi-source fusion analysis combined with other power environment monitoring data further comprises: Collecting an output value T_sample and humidity H_level of a temperature and humidity sensor; Acquiring the equipment vibration frequency F_vib and the local discharge signal amplitude P_discharge; By weight calculation: w_total=w_t×t_event+w_h×h_level +w_vib+F_vib+w_dis×P_discharge, wherein w_T, w_H, w_vib, w_dis represent the confidence weights of the temperature, humidity, vibration and discharge data respectively; and combining the W_total with the load state of the operation scene to judge whether the fault triggering condition is met.
  7. 7. The machine room dynamic environment heterogeneous data real-time integration and fault diagnosis method according to claim 6, wherein the method further comprises: assigning different confidence weights to each dimension data, wherein the confidence weights are determined based on the statistical characteristics of the historical data; calculating the comprehensive fault determination confidence D_confidence = (w_h × I_confidence + w_n × N_confidence + w_temp × Temp_conf + w_vib × Vib_conf) × k,k according to a formula to be a normalization factor; comparing the D_confidence with a first threshold T1 to judge a first-level response; if D_confidence is smaller than T1 and larger than T2, the second-level response flow is entered.
  8. 8. The method for real-time integration and fault diagnosis of heterogeneous data of a power environment of a machine room according to claim 7, wherein the primary response processing comprises: triggering relay control logic when the confidence D_confidence > T1; Determining the isolation action force by using a formula Isolate = (d_confidence-T1)/(T1-T2) ∈ [0,1 ]; Isolate when approaching 1, enabling full path isolation, otherwise, starting a local isolation mode; and sending an isolation instruction through the breaker module to block a corresponding fault loop.
  9. 9. The method for real-time integration and fault diagnosis of heterogeneous data of a power environment of a machine room according to claim 8, wherein the secondary response process comprises: Entering a work order generation stage when the confidence coefficient is between T1 and T2; Setting the enhanced sampling frequency to be f_enhanced=f_initial× (1+λ×log (d_enhancement), where λ is a sampling enhancement factor; enabling high frequency data collection and storing logs to assist in post-hoc retrospective analysis; and pushing maintenance work list notification containing fault level, position and related data at the monitoring terminal.
  10. 10. The method for real-time integration and fault diagnosis of heterogeneous data of a power environment of a machine room according to claim 9, wherein when the confidence of fault determination is lower than a second threshold, the method comprises: Setting the recording interval as tau_recording, and analyzing the slope of a current curve through a time sequence algorithm; Calculate the short-term current Slope using the formula slope= (i_current-i_previous)/τ_recovery; if Slope exceeds the set threshold sigma, upgrading the event into an object of interest and adding the object of interest into an early warning database; The load curve change over the subsequent 30 minutes is continuously tracked, and potential risks are identified.

Description

Real-time integration and fault diagnosis method and system for heterogeneous data of machine room power environment Technical Field The application relates to the technical field of intelligent operation and maintenance, in particular to a real-time integration and fault diagnosis method and system for heterogeneous data of a machine room dynamic environment. Background In key power scenes such as data centers, communication machine rooms and the like, current monitoring of branch circuits of a power distribution cabinet is an important means for guaranteeing stable operation of a system and preventing overload faults. However, when the existing machine room power environment monitoring system is used for dealing with high-load and dynamic change working conditions, the accuracy of fault diagnosis of the existing machine room power environment monitoring system faces challenges. The traditional out-of-limit alarming method generally adopts a fixed current threshold value, and cannot fully consider the difference of sampling precision levels of current sensors in actual deployment (such as a high-precision sensor and a common precision sensor), so that the sensing sensitivity and the reliability of the same electrical event are different, and the fairness and the effectiveness of alarming criteria are further affected. Particularly, under the working condition that the load fluctuates severely or transient impact (such as starting and stopping of a server cluster) exists, the fixed threshold value is easy to cause false alarm or missing alarm, so that fault positioning is fuzzy, and real overload abnormality and normal transient load change cannot be accurately distinguished. In addition, the existing scheme focuses on the out-of-limit judgment of a single data source, lacks deep fusion and confidence weighting analysis of multi-source heterogeneous monitoring data such as temperature, vibration, partial discharge and the like, and is difficult to realize closed loop fault management from abnormal alarm to accurate positioning to hierarchical treatment. Therefore, a fault diagnosis method capable of adaptively adjusting an alarm threshold value according to a current sampling precision level, intelligently tolerating load impact, and fusing multi-source data to perform confidence assessment is needed to improve the accuracy of fault positioning and the rationality of a response strategy under a high-load complex working condition. Disclosure of Invention In view of the above, the embodiments of the present disclosure provide a method and a system for real-time integration and fault diagnosis of heterogeneous data of a power environment of a machine room, which at least partially solve the problems existing in the prior art. The application discloses a real-time integration and fault diagnosis method for heterogeneous data of a machine room power environment, which comprises the following steps: collecting current sampling data of each branch circuit of a power distribution cabinet in a machine room, and classifying and marking the current sampling data according to the current sampling precision grade, wherein the high-precision sampling data are marked as H types, and the common precision sampling data are marked as N types; based on the current sampling data after classification marking, a dynamic out-of-limit alarm threshold model is constructed, and the model self-adaptively adjusts the alarm threshold according to the current sampling precision level and the current load change trend, and specifically comprises the following steps: Setting a first threshold interval for H-class data, and dynamically tightening a threshold boundary according to the recent load fluctuation rate; Setting a second threshold interval for N types of data, and introducing a load change trend prediction compensation quantity to adapt to load step change; If the current load change rate exceeds the preset impact tolerance limit, starting an instantaneous load impact tolerance window, and suspending out-of-limit alarm triggering in a window period; according to the dynamic out-of-limit alarm threshold model, comparing the current value in real time, identifying whether an out-of-limit alarm event exists or not, and carrying out preliminary fault positioning according to the topological connection relation of the out-of-limit branch circuit; And combining the preliminary fault positioning result with other power environment monitoring data in the machine room, performing multisource fusion analysis, wherein the other power environment monitoring data at least comprise temperature and humidity, equipment vibration, partial discharge and insulation state data, giving different confidence weights to data with different precision grades through a weighted fusion strategy, and finishing accurate judgment and grading response processing of faults under high-load working conditions, wherein the steps of: If the fault judgment confidence