Intelligent monitoring method for operation state of data center bus duct

Abstract

The present application belongs to the technical field of data center power distribution, and particularly relates to a data center bus duct operation state intelligent monitoring method, comprising: obtaining a temperature distribution curve through a distributed temperature measurement optical fiber, locating a high temperature suspected section and collecting joint lap joint pressure; when the first order difference absolute value of the total harmonic current distortion rate exceeds a trigger threshold, triggering a high-density sampling event, obtaining multi-parameter synchronous frozen data to form a harmonic pulse event slice; based on the harmonic pulse event slice and the joint thermal time constant, calculating a joint cumulative thermal impact index with an exponential decay kernel; sending the joint cumulative thermal impact index into a neural network fusion model to output an insulation aging score, and when the score exceeds a grading threshold, generating a warning signal and executing an intervention action. Through event-driven sampling and cumulative thermal impact calculation, the joint thermal damage accumulation trend is perceived in advance, and the warning opportunity is moved forward.

Description

Technical Field

[0001] This invention relates to the field of data center power distribution technology. More specifically, this invention relates to an intelligent monitoring method for the operating status of data center busbar trunking. Background Technology

[0002] Data centers, as critical infrastructure supporting massive computing power, typically employ dense busbar trunking as their main power supply channel to deliver high current to loads such as server racks, precision air conditioners, and UPS systems. Due to the characteristics of data center loads, including dense IT equipment switching power supplies, a high proportion of nonlinear loads, and significant diurnal business fluctuations, busbar trunking is subjected to complex thermal stress from the fundamental current superimposed with 3rd to 13th harmonic currents over long periods. Weak points such as connectors, insulating gaskets, and terminal boxes are highly susceptible to insulation degradation. Once insulation breakdown occurs, it will cause a complete power outage, directly triggering a business-level shutdown.

[0003] Existing technologies for monitoring the operating status of busbar trunking typically involve laying distributed temperature-measuring optical fibers or deploying infrared temperature probes on the surface of the busbar trunking to capture temperature rises. Simultaneously, current transformers and voltage transformers are connected to collect the effective values ​​of operating current and voltage in real time. Furthermore, harmonic current sensors, contact resistance monitoring units, vibration sensors, and temperature and humidity probes are introduced. Multiple parameters, including temperature, current, voltage, harmonics, contact resistance, and vibration, are integrated into a multi-parameter fusion model such as a BP neural network. By weighted fusion of these multi-source parameters, an insulation aging score and a predicted insulation resistance value are output, thereby enabling the prediction of insulation aging trends and compliance testing.

[0004] However, because the harmonic currents of data center IT loads are not steady-state but exhibit burst pulse characteristics on a second-level scale due to virtual machine migration, batch task scheduling, and hot / cold aisle switching, the total distortion rate of harmonic currents in the same busbar section can surge from 7% to 18% and then fall back within two adjacent 30-second sampling windows. Existing multi-parameter fusion models based on fixed sampling periods and equal weighting of all parameters will, on a long-term scale, mix and average the energy of short-term high-amplitude harmonic pulses that actually induce cumulative thermal damage to joints with the thermal contribution of steady-state fundamental waves. This results in the insulation aging score output by the model showing a slow linear increase, failing to identify the cumulative thermal shock of joint micro-areas formed by the superposition of dozens of sub-minute harmonic pulses. Ultimately, within the critical warning window before the insulation resistance falls below the 50MΩ compliance threshold, a health judgment is always given, causing maintenance personnel to miss the best intervention opportunity for joint tightening and harmonic mitigation. Summary of the Invention

[0005] To address the technical problem that existing multi-parameter fusion monitoring models, due to their equal-interval sampling and steady-state fusion, cannot capture the cumulative thermal shock of sub-minute harmonic pulses at the joints of data center busbar trunking, resulting in insulation aging warnings lagging behind the insulation resistance compliance critical point, this invention provides an intelligent monitoring method for the operating status of data center busbar trunking.

[0006] This invention provides an intelligent monitoring method for the operating status of data center busbar trunking, comprising: S1: acquiring the temperature distribution curve of the busbar trunking by laying distributed temperature-measuring optical fibers, locating suspected high-temperature sections based on sliding scanning and historical temperature distribution statistical characteristics, and deploying sensors in the suspected high-temperature sections to collect joint lap pressure; S2: deploying an event triggering circuit at the front end of the harmonic current sensor to calculate the total harmonic current distortion rate in real time, and triggering a high-density sampling event when the absolute value of the first-order difference of the total harmonic current distortion rate exceeds the trigger threshold calibrated based on the statistical characteristics of the silent section, thereby acquiring data including the joint lap pressure. S3: Based on the harmonic pulse event slice, extract the root mean square value of harmonic current, contact resistance and equivalent duration of the event, and combine it with the joint thermal time constant calibrated by the step current injection test to calculate the joint cumulative thermal shock index with exponential decay kernel; S4: Use the joint cumulative thermal shock index as the input channel to feed into the neural network fusion model to output the insulation aging score. When the insulation aging score exceeds the classification threshold determined based on the statistical distribution of historical monitoring data, generate an early warning signal and execute the corresponding intervention action.

[0007] By adopting the above technical solution, this invention introduces an event-driven sampling method based on the binding of first-order differential and silent segment statistical characteristics at the front end of the harmonic current sensor. This breaks through the time-averaged masking of sub-minute harmonic pulses by equal-interval sampling, allowing the originally diluted short-time high-amplitude harmonic energy to be completely preserved. At the same time, it introduces an improved accumulation form with thermal memory attenuation kernel, which objectively reflects the true accumulation state of undissipated harmonic heat energy in the junction micro-region, eliminates the systematic overestimation of residual thermal damage of distant historical pulses, and significantly advances the early warning timing.

[0008] Preferably, the step of locating the suspected high-temperature section based on sliding scanning and historical temperature distribution statistical characteristics includes: performing a sliding scan of the temperature distribution of the entire line at preset time intervals to obtain the standard deviation and average temperature of the historical temperature distribution of the same busbar under the same load current level; marking sampling points that exceed the sum of the standard deviations of the historical temperature distribution and the average temperature by a preset multiple as abnormal points; and finally obtaining the suspected high-temperature section when the spatial span of consecutive abnormal points is not less than the actual coverage length of the crimped section of the busbar connector joint.

[0009] By adopting the above technical solution, this invention combines the one-dimensional spatial profile of distributed optical fiber temperature measurement with adaptive current grading threshold determination, thereby achieving accurate spatial positioning of suspected high-temperature sections of joints under load tidal fluctuations. This avoids a large number of false alarms and missed alarms under light and heavy load conditions with fixed thresholds, and provides accurate spatial anchor points for subsequent event-driven sampling.

[0010] Preferably, the step of arranging sensors in the suspected high-temperature section to collect the joint lap pressure includes: arranging a multi-channel infrared temperature sensor at the connector joint for the calibrated suspected high-temperature section; using the multi-channel infrared temperature sensor to measure the temperature of the lap surfaces of the multiphase conductor and the neutral conductor respectively; and simultaneously arranging a thin-film contact pressure sensor on the bolt clamping surface of the joint to finally obtain the joint lap pressure.

[0011] By adopting the above technical solution, this invention achieves simultaneous acquisition of the temperature of the lap joint surface of the multiphase conductor and the neutral conductor and the lap joint pressure by arranging multi-channel infrared temperature sensors and thin-film contact pressure sensors at the connector joint in the high-temperature suspected section. This allows abnormal temperature rise of the joint to be corroborated with changes in lap joint pressure, thereby improving the ability to identify increased contact resistance and local abnormal heating caused by joint loosening and insufficient clamping force. This provides more accurate monitoring data for subsequent joint thermal shock assessment and abnormal cause judgment.

[0012] Preferably, the trigger threshold based on the statistical characteristics of the silent segment includes: continuously collecting the baseline sequence of total harmonic current distortion rate during off-peak hours, removing data aligned with known abrupt change segments, and retaining the silent segments related to normal switching noise; calculating the mean and standard deviation of the absolute value of the first-order difference of the silent segment; and finally forming the trigger threshold by summing the mean and the standard deviation by a preset multiple.

[0013] Preferably, the step of triggering a high-density sampling event to acquire multi-parameter synchronous frozen data, including the joint lap pressure, and forming a harmonic pulse event slice includes: after the high-density sampling event is triggered, extending the time before and after the event occurrence by a preset duration to form a high-density sampling window; synchronously freezing and collecting temperature, current, voltage, joint lap pressure, contact resistance, and vibration value within the high-density sampling window to acquire the multi-parameter synchronous frozen data; and integrating the multi-parameter synchronous frozen data to finally form the harmonic pulse event slice.

[0014] By adopting the above technical solution, the present invention ensures that the transient peak values ​​of multiple physical quantities with different response delays can fall completely within the same harmonic pulse event slice through the high-density synchronous freezing acquisition method of front and rear extended windows. This avoids the time sequence misalignment of multiple physical quantities caused by unilateral truncation and provides an undiluted raw data base for the accurate calculation of cumulative thermal shock.

[0015] Preferably, the step current injection test calibrated to determine the joint thermal time constant includes: performing a step current injection test on the target bus joint in a calibration state isolated from the data center operating loop, removing the step current after the temperature rise of the target bus joint reaches a stable state; observing the natural cooling temperature rise curve of the joint collected by the infrared temperature sensor after the step current is removed; recording the time taken for the natural cooling temperature rise curve of the joint to decay from the steady-state temperature rise when the step current is removed to a preset proportion of the steady-state temperature rise, and finally obtaining the joint thermal time constant.

[0016] Preferably, the calculation of the joint cumulative thermal shock index with an exponential decay kernel includes: calculating the product of the square of the root mean square value of the harmonic current, the contact resistance, and the equivalent duration of the event to obtain the Joule heat of a single event; calculating the time difference between the current moment and the moment the event occurred, dividing the time difference by the joint thermal time constant and taking a negative value, using this as an exponential term to calculate the exponential decay kernel; multiplying the Joule heat of a single event by the exponential decay kernel, and summing the results over multiple historical events to finally obtain the joint cumulative thermal shock index.

[0017] By adopting the above technical solution, this invention improves the Joule heating formula into a pulse sequence accumulation form with an exponential decay kernel, which truly restores the physical fact that the harmonic pulse closer to the current time contributes more to the marginal thermal damage of the joint, and the earlier pulses contribute to the decay due to thermal diffusion. This provides a quantitative thermal damage input with clear physical meaning for the fusion model.

[0018] Preferably, the grading threshold determined based on the statistical distribution of historical monitoring data includes: tracing back samples from historical monitoring data that have been confirmed to have experienced a decrease in insulation resistance; statistically analyzing the distribution range of insulation aging scores within a preset time period before the occurrence of the sample; extracting the lower quartile, upper quartile, and statistical upper bound of the compliance segment score sequence of the distribution range, and finally obtaining the grading threshold.

[0019] Preferably, the step of generating the early warning signal and executing the corresponding intervention action includes: when the insulation aging score exceeds the grading threshold, generating the early warning signal according to different levels; determining the harmonic number corresponding to the largest harmonic current root mean square value in the harmonic pulse event slice as the dominant harmonic number based on the root mean square values ​​of different harmonic currents in the harmonic pulse event slice; pushing the early warning signal, along with the spatial coordinates of the corresponding high-temperature suspected section, the timestamp of the harmonic pulse event slice, and the dominant harmonic number, to the operation and maintenance platform; using the phase comparison method to lock the harmonic source branch; issuing a compensation switching signal or load migration suggestion to the harmonic source branch; generating a bolt tightening work order for the section with abnormal joint lap pressure; and finally executing the corresponding intervention action.

[0020] Preferably, the step of using the phase comparison method to locate the harmonic source branch includes: acquiring the current transformer data of each outgoing branch on the low-voltage side of the distribution cabinet; finding the branch in the current transformer data that best matches the phase of the harmonic current of the main busbar at the dominant harmonic order; and finally locking the cabinet group corresponding to the best-matching branch as the harmonic source branch.

[0021] The technical solution of the present invention has the following beneficial technical effects:

[0022] This invention starts with the underlying data acquisition mechanism, changing traditional time-driven periodic sampling to energy-driven event sampling, effectively capturing the sub-minute harmonic pulses unique to data centers. This approach avoids the averaging of high-frequency thermal shock energy over a long time window, ensuring that subsequent thermal damage assessments are based on real, high-fidelity transient data, significantly improving the monitoring system's sensitivity to early-stage hazards.

[0023] Furthermore, the deep integration of the joint cumulative thermal shock index constructed in this invention with the neural network fusion model not only enables early prediction of insulation aging trends but also establishes a complete link from state perception to proactive intervention through data-driven hierarchical thresholds and automatic traceability. Maintenance personnel can accurately implement harmonic mitigation and physical tightening before substantial damage to insulation performance occurs, greatly ensuring the operational continuity of the data center power distribution system. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating an intelligent monitoring method for the operating status of a data center busbar according to the present invention; Figure 2 This is a schematic diagram showing the comparison curves of the cumulative thermal shock index of the joint based on the exponential decay kernel and the equal weight superposition method in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0026] This invention discloses an intelligent monitoring method for the operating status of data center busbar trunking, referring to... Figure 1 This includes steps S1-S4: S1. Obtain the temperature distribution curve to collect the joint lap pressure.

[0027] In an optional embodiment, distributed temperature-measuring optical fibers are spirally wound along the busbar routing on the surfaces of the A, B, C phases and neutral line outer shell of the data center's dense busbar trunking. Utilizing the Stokes and anti-Stokes intensity ratio of the backscattered Raman light in the fiber, temperature distribution curves at multiple sampling points along the route are demodulated, covering the entire busbar section from the distribution cabinet outlet to the terminal plug-in box, forming a one-dimensional spatial image of the busbar trunking's temperature field. A sliding scan of the entire line's temperature distribution is performed at 30-second intervals, based on the standard deviation of the historical temperature distribution of this busbar section under the same load current. With average temperature This historical temperature distribution statistical characteristic will exceed the average temperature of the same section of the busbar. and The sampling points of the sum are marked as outliers, i.e., those that satisfy... The sampling points are considered outliers.

[0028] In the formula, For the current sliding scan time, the th The measured temperature at each sampling point is in °C. Number the temperature sampling points arranged along the busbar trunking; In order to be with the first The historical average temperature of each sampling point under the same load current range for the bus section to which it belongs is expressed in °C. 3 represents the standard deviation of historical temperature distribution in the busbar section under the same load current rating, in °C; 3 represents the multiple of the standard deviation of historical temperature distribution used in this embodiment.

[0029] Furthermore, when the spatial span of consecutive anomalies is not less than the actual coverage length of the crimped section of the busbar connector of that model, the section is identified as a suspected high-temperature section, and the spatial coordinates of the suspected high-temperature section are sent to the next-level monitoring device. Due to different busbar specifications, the actual coverage length of the crimped section varies; for example, this length can be 15 to 30 centimeters. Distributed temperature-measuring optical fibers form multiple temperature sampling points along the busbar, and the suspected high-temperature section is determined by comparing the actual spatial span of consecutive anomalies with the actual coverage length of the crimped section. The load current is categorized based on the measured busbar current read from the current transformer at the same sampling time, divided into 200A increments, and the historical average temperature and the historical temperature distribution standard deviation are also considered. All data are obtained from historical records of the corresponding current range, thus enabling the threshold to adaptively fluctuate with the actual load level and eliminating misjudgments caused by load tidal fluctuations.

[0030] Next, for the identified high-temperature suspected section, a four-channel infrared temperature sensor was installed at the connector joint, pointing towards the lap surfaces of the A-phase, B-phase, C-phase conductors and the neutral conductor, respectively. Simultaneously, a thin-film contact pressure sensor was installed on the bolt clamping surface of the joint to collect the joint lap pressure. Because the neutral line in a data center carries zero-sequence harmonic currents superimposed by three-phase nonlinear loads, the neutral conductor often experiences the most significant temperature rise. Four-channel infrared thermography can accurately identify whether the neutral conductor exhibits an abnormal temperature rise higher than that of the phase conductors. Combined with joint lap pressure data, it can determine whether the temperature rise is caused by deterioration of contact resistance due to bolt loosening, providing spatial location and physical quantity basis for subsequent cause tracing. It should be noted that there are various other methods in this field that can achieve temperature distribution demodulation along the line; those skilled in the art can choose different distributed temperature measurement schemes according to actual needs.

[0031] In this way, by combining the one-dimensional spatial profile of distributed fiber optic temperature measurement with adaptive current grading threshold determination, accurate spatial positioning of suspected high-temperature sections is achieved under the background of load tidal fluctuations. This avoids a large number of false alarms and missed alarms under light and heavy load conditions with fixed thresholds, and provides accurate spatial anchors for subsequent event-driven sampling.

[0032] S2, triggering high-density sampling events to form harmonic pulse event slices.

[0033] In an optional embodiment, the harmonic current sensor continuously samples the bus current at a rate of 2kHz and calculates the root mean square (RMS) values ​​of the 3rd, 5th, 7th, 9th, 11th, and 13th harmonic currents using FFT. The harmonic order corresponding to the maximum value among these three harmonic current RMS values ​​is determined as the dominant harmonic order of the current harmonic pulse event. A hardware-level event triggering circuit deployed at the front end of the harmonic current sensor calculates the total harmonic current distortion rate in real time. ,once The absolute value of the first-order difference exceeds the trigger threshold within any 100ms window. This immediately triggers a high-density sampling event.

[0034] Furthermore, trigger threshold The determination of the value is not based on human experience, but is defined by objective statistical methods: data is collected continuously for 24 hours during the nighttime off-peak business hours. The baseline sequence is used to remove abrupt changes that align with known virtual machine batch migration scheduling times, retaining silent segments related to normal IT equipment switching noise. The mean of the absolute values ​​of the first-order differences of these silent segments is then used. with standard deviation Based on this, let

[0035] In the formula, This is the trigger threshold corresponding to the absolute value of the first-order difference of the total harmonic current distortion rate; It is the mean of the absolute values ​​of the first-order difference of the total harmonic current distortion rate within the silent segment; The standard deviation of the absolute value of the first-order difference of the total harmonic current distortion rate during the silent segment; 3 represents the statistical data of the silent segment; 3 is the multiple of the standard deviation of the silent segment used in this embodiment. , and It adopts the same dimension as the absolute value of the first-order difference of the total harmonic current distortion rate.

[0036] When the absolute value of the first-order difference of the total harmonic current distortion rate exceeds the trigger threshold, a high-density sampling event is triggered. The system then extends the sampling window by a preset duration before and after the event, forming a high-density sampling window. For example, the extension duration before and after the event can be 10 seconds, corresponding to a high-density sampling window of 20 seconds. Within the window, temperature, current, voltage, joint lap pressure, contact resistance, and vibration values ​​are synchronously frozen and collected at a preset sampling rate. For example, this sampling rate can be 10Hz, acquiring synchronously frozen data of multiple parameters, including joint lap pressure, to form a set of harmonic pulse event slices. During periods without triggering, routine inspection sampling is maintained every 30 seconds. Since different physical quantities have different response delays when harmonic pulses occur, the response of contact resistance due to thermal expansion usually lags behind the harmonic current mutation by several seconds, while the temperature response lags even longer. The design of the expansion windows before and after ensures that the transient peak values ​​of each physical quantity fall within the same harmonic pulse event slice, avoiding the timing misalignment of multiple physical quantities caused by unilateral truncation.

[0037] In this way, the short-time harmonic pulses and their associated transient responses of multiple physical quantities, which were originally masked by time averaging, are independently preserved in the form of harmonic pulse event slices. This provides undiluted raw data for the subsequent calculation of cumulative thermal shock at the joint, fundamentally solving the problem of information loss caused by equal-interval sampling spreading sub-minute pulse energy across the entire sampling period.

[0038] S3. Calculate the cumulative thermal shock index of the joint with the exponentially decaying core.

[0039] In an optional embodiment, the existing Joule heating formula for joints in the field of power engineering is:

[0040] In the formula, The Joule heat generated at the joint is measured in J. The effective value of the current flowing through the connector is expressed in amperes (A). Contact resistance, in Ω; The duration of the current is expressed in seconds (s).

[0041] This formula only describes the instantaneous heat generation of the joint under steady-state current and cannot reflect the thermal memory effect formed by the thermal capacity of the joint metal between harmonic pulse events. Since the heat injected by earlier harmonic pulses is gradually dissipated through convection and conduction to the joint copper busbar, its marginal contribution to the current joint temperature rise decays over time. If the harmonic contributions at different times are regarded as equally weighted superposition, the residual thermal damage of distant historical pulses will be systematically overestimated, resulting in an overestimation of the cumulative thermal shock calculation.

[0042] Therefore, based on the existing Joule heating formula, this invention extends the single steady-state input to a cumulative form with thermal memory decay for pulse event sequences. It extracts the root mean square value of harmonic current, contact resistance, and equivalent duration of events based on harmonic pulse event slices. The improved relationship is as follows:

[0043] Compared with the original Joule heating formula, the key improvements are in two aspects: First, the duration of the single steady state in the original formula is reduced. Equivalent duration of each pulse event and cover by summation. The first historical event extends Joule heating calculations from single steady-state events to the time accumulation of pulse event sequences, where the product of the square of the root mean square value of the harmonic current, the contact resistance, and the equivalent duration of the event is the Joule heating of a single event; secondly, an exponentially decaying kernel is introduced for each historical event. The larger the attenuation kernel value, the closer the corresponding event is to the current moment, and the joint has not yet fully dissipated the heat injected by the pulse through convection and conduction. In this case, the marginal contribution of the event to the cumulative thermal damage is greater. The smaller the attenuation kernel value, the farther away the corresponding event is from the current moment, and most of the heat has been dissipated. In this case, the marginal contribution is smaller. This is consistent with the actual heat diffusion law of the bus joint copper bus.

[0044] In the formula, The cumulative thermal shock index of the target busbar joint at the current moment, in J; This represents the total number of historical harmonic pulse events involved in the current cumulative calculation; Let be the sequence number of the harmonic pulse event, and ; For the first The root mean square value of harmonic current within a slice of a subharmonic pulse event, in amperes (A). For the first The contact resistance obtained synchronously from subharmonic pulse event slices is expressed in Ω. For the first The equivalent duration of a subharmonic pulse event, expressed in seconds; The current moment for performing cumulative thermal shock calculations; For the first The timing of the occurrence of the subharmonic pulse event; For the first The time interval between the occurrence of a second harmonic pulse event and the current time is expressed in seconds. The thermal time constant of the target busbar joint is expressed in seconds. It is a natural constant.

[0045] Where thermal time constant Instead of being manually set, the test involves applying a constant step current to the test circuit containing the target busbar connector under a calibration state isolated from the data center's operating circuit using a dedicated test power supply. Once the temperature rise of the connector, as measured by the infrared temperature sensor, reaches a stable state, the step current is removed. The natural cooling temperature rise curve of the connector is then continuously observed, and the time it takes for the temperature rise to decay from the steady-state temperature rise after the step current is removed to 37% of the steady-state temperature rise is determined as the thermal time constant of the connector. For example, assuming a data center busbar connector with a rated current of 1600A, the thermal time constant calibrated through the step current injection test can be 900 seconds. If eight harmonic pulse events were triggered in the past hour, and assuming the root mean square of the harmonic current, contact resistance, and equivalent duration of each event can be 220A, 85μΩ, and 15 seconds respectively, then substituting these values ​​into the improved formula yields...

[0046] The calculation results, expressed in joules, objectively reflect the total amount of harmonic heat energy that has actually accumulated at the joint in the past hour and has not yet been dissipated, rather than the artificially inflated value obtained by weighting all historical pulses equally.

[0047] Thus, by modifying the existing Joule heating formula into a pulse sequence accumulation form with an exponentially decaying kernel, and by experimentally calibrating the thermal time constant of each joint using step response, the thermal time constant of each joint can be determined. A joint cumulative thermal shock index that can realistically reproduce the thermal accumulation state of the joint micro-region was constructed, providing a quantitative thermal damage input with clear physical meaning and no dilution for subsequent fusion models, filling the gap in the existing models in pulse thermal memory characterization.

[0048] S4. Output insulation aging score and execute corresponding intervention actions.

[0049] In an optional embodiment, the already constructed joint cumulative thermal shock index is used. As a newly added key input channel, the joint lap pressure, contact resistance, vibration value, voltage distortion rate, and ambient temperature and humidity, which are frozen synchronously with the distributed temperature and harmonic pulse event slices, are fed into the BP neural network fusion model. The model output layer retains two nodes: insulation aging score (0 to 100 points) and insulation resistance prediction value. The network weights of the BP neural network fusion model are obtained by adaptive learning from the training set data through backpropagation. The channel weights do not require manual prior specification. Compared to the original model that directly inputs harmonic current as instantaneous values, the improved model can accurately perceive how much harmonic heat energy has accumulated in the joint over a period of time and has not yet dissipated. Therefore, before the insulation resistance falls below the 50MΩ compliance threshold, based on... The continued rise in [the value] results in a steep increase in the insulation aging score.

[0050] Furthermore, the BP neural network fusion model employs a network structure consisting of an input layer, hidden layers, and an output layer, and constructs training samples based on historical monitoring data. For any given historical monitoring moment, the cumulative thermal shock index of the joint at that moment is calculated. Distributed temperature, joint lap pressure, contact resistance, vibration value, voltage distortion rate, and ambient temperature and humidity are used as a set of input samples, and the insulation resistance actually measured after a preset prediction time is used as the insulation resistance monitoring label corresponding to the input sample.

[0051] Simultaneously, an insulation aging score supervision label ranging from 0 to 100 is constructed based on the actual measured insulation resistance after a preset prediction period. The statistical benchmark value of insulation resistance under normal insulation conditions from historical monitoring data is used as the value corresponding to 0 points, and the 50MΩ compliance threshold is used as the value corresponding to 100 points. When the measured insulation resistance is greater than or equal to the statistical benchmark value, the insulation aging score supervision label is set to 0 points; when the measured insulation resistance is less than or equal to 50MΩ, the insulation aging score supervision label is set to 100 points. Measured insulation resistances between the statistical benchmark value and 50MΩ are normalized so that the insulation aging score increases when the insulation resistance decreases. The input samples and their corresponding insulation resistance supervision labels and insulation aging score supervision labels are input into a BP neural network fusion model for training. The network weights are iteratively adjusted through backpropagation until the error between the model output and the corresponding supervision label meets the preset convergence condition, ultimately obtaining the trained BP neural network fusion model.

[0052] Furthermore, the three-level warning thresholds for insulation aging scores are not determined manually based on experience, but rather through objective statistical methods: Samples of confirmed insulation resistance decline events are reviewed from historical monitoring data, and the distribution range of insulation aging scores in the hours preceding these events is statistically analyzed. The lower quartile of the score distribution is used as the lower bound for an orange warning, the upper quartile as the lower bound for a red warning, and the statistical upper bound of the score sequence when the insulation resistance is still within the compliant range (the mean of the compliant score plus twice the standard deviation) as the lower bound for a yellow warning. This allows the three-level thresholds to be automatically anchored by historical event data and continuously updated as subsequent operational data accumulates. When the insulation aging score exceeds the tiered thresholds determined based on the statistical distribution of historical monitoring data, the system generates warning signals at three levels: yellow, orange, and red. These warning signals, along with the corresponding spatial coordinates of the suspected high-temperature section, the timestamp of the harmonic pulse event slice, and the dominant harmonic order, are pushed to the operation and maintenance platform.

[0053] Next, targeting the dominant harmonic order, the system uses a phase comparison method to search the current transformer data of each outgoing branch on the low-voltage side of the distribution cabinet. It identifies the branch whose harmonic current phase best matches the main busbar at the dominant harmonic order, and pinpoints the corresponding cabinet group as the harmonic source. For the identified harmonic source branch, the system automatically issues active intervention actions: it sends a harmonic compensation switching signal corresponding to the dominant harmonic order to the active power filter connected to that branch, pushes a load migration suggestion for the cabinet group to the data center DCIM system to reduce nonlinear load density, and generates bolt tightening work orders for sections with abnormal joint pressure. These intervention actions reduce the triggering frequency of subsequent pulse events at the harmonic source. Subsequently, due to the exponential decay kernel, it naturally falls back, forming a complete processing chain of event sampling, thermal damage accumulation calculation, score enhancement, source intervention, and cumulative decay. It should be noted that there are many other methods in this field that can achieve harmonic source tracing and localization, and those skilled in the art can choose different tracing algorithms according to the actual network topology and sensor configuration.

[0054] Thus, by accumulating thermal shock index at the joint By introducing a BP neural network fusion model and combining it with data-driven three-level threshold automatic anchoring, the model can detect the real accumulation trend of joint thermal damage before the insulation resistance reaches the compliance critical point, thus significantly advancing the early warning time. At the same time, it links harmonic source tracing with active filter switching, load migration suggestions, and bolt tightening work orders, reducing the frequency of subsequent harmonic pulse events from the source, and enabling the entire monitoring system to have complete processing capabilities from perception to intervention.

[0055] Figure 2 This is a comparison curve of the cumulative thermal shock index of the joint based on the exponential decay kernel and the equal-weight superposition method in this embodiment of the invention. The figure shows two curves representing the evolution of the cumulative thermal shock index of the joint over time under the same harmonic pulse event sequence using the existing method and the method of the present invention: the curve of the existing method uses a solid line superimposed with square markers, calculated using an equal-weight superposition method, and increases monotonically with event accumulation; the curve of the present invention uses a solid line superimposed with circular markers, using an exponential decay kernel to weight historical thermal shock events, causing the contribution of distant historical events to decay over time, and the overall curve is lower than that of the existing method. The vertical dashed lines in the figure mark the triggering time of each harmonic pulse event, and the legend provides independent explanations for the two curves and the triggering time of the harmonic pulse events. It can be observed that the cumulative thermal shock index of the joint obtained by the method of the present invention is generally lower than the equal-weight superposition result of the existing method, indicating that the method of the present invention can effectively suppress the excessive accumulation effect of historical thermal shock events and more realistically reflect the current thermal damage state of the joint.

[0056] It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this invention, and these modifications and improvements are all within the scope of protection of this invention.

Claims

1. A method for intelligent monitoring of the operating status of data center busbar trunking, characterized in that, include: S1: The temperature distribution curve of the bus trunk is obtained by laying distributed temperature measurement optical fibers. Based on the sliding scan and historical temperature distribution statistical characteristics, the suspected high temperature section is located, and sensors are arranged in the suspected high temperature section to collect the joint lap pressure. S2: Deploy an event triggering circuit at the front end of the harmonic current sensor to calculate the total harmonic current distortion rate in real time. When the absolute value of the first-order difference of the total harmonic current distortion rate exceeds the trigger threshold calibrated based on the statistical characteristics of the silent segment, a high-density sampling event is triggered to obtain multi-parameter synchronous frozen data including the joint lap pressure, forming a harmonic pulse event slice. S3: Based on the harmonic pulse event slice, extract the root mean square value of harmonic current, contact resistance and equivalent duration of the event, and combine the joint thermal time constant calibrated by the step current injection test to calculate the cumulative thermal shock index of the joint with the exponential decay core. S4: The cumulative thermal shock index of the joint is sent as an input channel to the neural network fusion model, and the insulation aging score is output. When the insulation aging score exceeds the grading threshold determined based on the statistical distribution of historical monitoring data, an early warning signal is generated and the corresponding intervention action is executed.

2. The intelligent monitoring method for the operating status of a data center busbar trunking according to claim 1, characterized in that, The method of locating suspected high-temperature sections based on sliding scanning and historical temperature distribution statistical characteristics includes: A sliding scan of the temperature distribution across the entire line is performed at preset time intervals to obtain the historical temperature distribution standard deviation and average temperature of the same busbar under the same load current range. Sampling points that exceed the sum of the standard deviations of the historical temperature distribution by a preset multiple of the average temperature are marked as outliers. When the spatial span of the consecutive abnormal points is not less than the actual coverage length of the crimped section of the bus connector, the high-temperature suspected section is finally obtained.

3. The intelligent monitoring method for the operating status of a data center busbar trunking according to claim 1, characterized in that, The step of arranging sensors in the suspected high-temperature section to collect joint lap pressure includes: For the identified high-temperature suspected area, a multi-channel infrared temperature sensor is arranged at the connector joint; The multi-channel infrared temperature sensor was used to measure the temperature at the interface of the multiphase conductor and the neutral conductor, respectively. A thin-film contact pressure sensor is synchronously placed on the bolt clamping surface of the joint to obtain the joint lap pressure.

4. The intelligent monitoring method for the operating status of a data center busbar trunking according to claim 1, characterized in that, The trigger threshold based on the statistical characteristics of the silent segment includes: During periods of low business activity, the baseline sequence of total harmonic current distortion rate is continuously collected. Data aligned with known abrupt change segments are removed, while silent segments related to normal switching noise are retained. Calculate the mean and standard deviation of the absolute values ​​of the first-order differences of the silent segment; The sum of the mean and the standard deviation by a preset multiple is finally used to form the trigger threshold.

5. The intelligent monitoring method for the operating status of a data center busbar trunking according to claim 1, characterized in that, The triggered high-density sampling event acquires multi-parameter synchronous frozen data, including the joint overlap pressure, to form a harmonic pulse event slice, including: After the high-density sampling event is triggered, a preset duration is extended before and after the event occurs to form a high-density sampling window; Within the high-density sampling window, temperature, current, voltage, joint lap pressure, contact resistance, and vibration value are simultaneously frozen and collected to obtain the multi-parameter synchronous frozen data. The multi-parameter synchronous frozen data are integrated to form the harmonic pulse event slice.

6. The intelligent monitoring method for the operating status of a data center busbar trunking according to claim 1, characterized in that, The combination of the joint thermal time constant calibrated by the step current injection test includes: With the target bus joint in a calibration state isolated from the data center operating circuit, a step current injection test is performed on the target bus joint, and the step current is removed after the temperature rise of the target bus joint reaches a stable state. The temperature rise curve of the connector under natural cooling was collected by the infrared temperature sensor after the step current was removed. The time taken for the joint's natural cooling temperature rise curve to decay from the steady-state temperature rise when the step current is removed to a preset proportion of the steady-state temperature rise is recorded, and the joint's thermal time constant is finally obtained.

7. The intelligent monitoring method for the operating status of a data center busbar trunking according to claim 1, characterized in that, The calculation of the cumulative thermal shock index of the joint with an exponentially decaying core includes: The Joule heating of a single event is obtained by multiplying the square of the root mean square value of the harmonic current by the contact resistance and the equivalent duration of the event. Calculate the time difference between the current moment and the moment the event occurred, divide the time difference by the joint thermal time constant and take the negative value, and use it as the exponential term to calculate the exponential decay kernel; The cumulative thermal shock index of the joint is obtained by multiplying the single-event Joule heat by the exponential decay kernel and summing the results over multiple historical events.

8. The intelligent monitoring method for the operating status of a data center busbar trunking according to claim 1, characterized in that, The grading thresholds determined based on the statistical distribution of historical monitoring data include: Samples from historical monitoring data that have been confirmed to have experienced a decrease in insulation resistance; The distribution range of insulation aging scores within a preset time period before the occurrence of the sample is statistically analyzed; The lower quartile, upper quartile, and statistical upper bound of the compliance segment score sequence of the distribution interval are extracted to finally obtain the classification threshold.

9. The intelligent monitoring method for the operating status of a data center busbar trunking according to claim 1, characterized in that, The generation of early warning signals and the execution of corresponding intervention actions include: When the insulation aging score exceeds the grading threshold, the warning signal is generated according to different levels. Based on the root mean square values ​​of different harmonic currents in the harmonic pulse event slice, the harmonic number corresponding to the largest root mean square value of the harmonic current is determined as the dominant harmonic number. The warning signal, along with the corresponding spatial coordinates of the high-temperature suspected section, the timestamp of the harmonic pulse event slice, and the dominant harmonic number, is pushed to the operation and maintenance platform. Use phase comparison method to locate the harmonic source branch; The system sends a compensation switching signal or load migration suggestion to the harmonic source branch, generates a bolt tightening work order for the abnormal joint lap pressure section, and finally executes the corresponding intervention action.

10. The intelligent monitoring method for the operating status of a data center busbar trunking according to claim 9, characterized in that, The method of using phase comparison to lock the harmonic source branch includes: Obtain the current transformer data of each outgoing branch on the low-voltage side of the distribution cabinet; Find the branch in the current transformer data that best matches the phase of the harmonic current of the main bus at the dominant harmonic order. The cabinet group corresponding to the most matching branch is ultimately identified as the harmonic source branch.

Images