Intelligent power distribution equipment fault diagnosis system for data center scenarios

Abstract

The application discloses a data center scene-oriented intelligent power distribution equipment fault diagnosis system and belongs to the technical field of data center power distribution fault diagnosis. The system comprises a data acquisition module, which acquires the instantaneous current values of each feeder circuit in a direct-current bus power distribution cabinet and the breaking state values of solid-state circuit breakers; a vector construction module, which constructs real-time current dynamic vectors of each feeder circuit by taking the rate of change of the instantaneous current values with time as the first dimension and the cumulative amount of the amplitude fluctuation range of the instantaneous current values within a preset time window as the second dimension; an offset quantification module, which obtains a current abnormal offset degree according to the difference between the real-time current dynamic vector and a reference current dynamic vector; and a distortion aggregation module, which aggregates the current abnormal offset degrees by using the spatial distribution weights of the topological connection relationship between the feeder circuits to generate an overall current distortion factor. The system can improve the early fault detection sensitivity of the power distribution cabinet and the rationality of the diagnosis priority.

Description

Technical Field

[0001] This invention relates to the field of data center power distribution fault diagnosis technology, specifically to an intelligent power distribution equipment fault diagnosis system for data center scenarios. Background Technology

[0002] Data center DC bus distribution cabinets carry a large number of feeder circuits, and their operating status directly affects the power supply continuity of server loads. Current technologies for fault diagnosis of feeder circuits within distribution cabinets often employ single current threshold comparisons or simple current change rate analysis. This involves comparing real-time collected branch currents with preset protection settings; once the threshold is exceeded, a fault is identified and protection is triggered. This fixed-boundary-based approach cannot accurately depict the dynamic evolution of current signals across multiple dimensions. When feeder circuits exhibit significant transient anomalies such as poor contact, partial discharge, or early arcing, the current amplitude may not yet exceed limits, but its trend and short-term fluctuations show identifiable distortions. Existing methods are prone to missing or delaying the detection of such potential problems. Furthermore, multiple feeder circuits within the DC bus distribution cabinet are physically adjacent to each other, and these circuits influence each other through the common bus and electromagnetic coupling. Local anomalies in a single circuit often propagate interference to adjacent circuits along the topology. Existing diagnostic methods do not incorporate the actual spatial topological relationships between circuits into the analysis, only making isolated judgments on each circuit. This makes it impossible to accurately assess the risk of anomaly propagation and the overall distortion level at the entire distribution cabinet level. It also lacks a reasonable prioritization mechanism for troubleshooting multiple circuit faults, causing maintenance personnel to rely on experience for response order, potentially delaying the handling of high-risk circuits. In fault risk quantification, the instantaneous value of anomaly severity often fails to reflect the true hazard of the fault mode. Some intermittent anomalies or continuously worsening fault modes exhibit drastically different oscillation characteristics in their temporal sequence. However, existing technologies do not delve into the time series of anomaly offsets, failing to distinguish between occasional disturbances and persistent fault precursors with high-frequency fluctuations. This makes the diagnostic results unable to effectively differentiate the urgency of circuits. The problem to be solved by this invention is how to extract characteristic quantities that can sensitively reflect early anomalies from the multidimensional dynamic information of current, and combine them with the spatial topological relationship between circuits in the distribution cabinet to form a unified index that characterizes the overall distortion state; how to use the frequency of change of abnormal offset in the time dimension to quantify the risk level of fault mode, so as to realize objective and orderly fault diagnosis output for each feeder circuit. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent power distribution equipment fault diagnosis system for data center scenarios. This system obtains the circuit abnormal offset by constructing a real-time current dynamic vector that integrates the current change rate and amplitude fluctuation range, and introduces spatial distribution weights based on the physical topology connection relationship between feeder circuits to generate an overall current distortion factor. At the same time, it determines the fault risk priority coefficient by combining the time sequence sign change frequency of the current abnormal offset, thereby realizing a comprehensive diagnostic capability from single circuit abnormal perception to whole cabinet risk aggregation and diagnostic priority ranking.

[0004] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides an intelligent power distribution equipment fault diagnosis system for data center scenarios, including a data acquisition module, a vector construction module, an offset quantization module, a distortion aggregation module, a risk ranking module, and a diagnostic output module. The data acquisition module acquires the instantaneous current values ​​of each feeder circuit in the DC bus distribution cabinet and the breaking status values ​​of the corresponding solid-state circuit breakers for each feeder circuit. Preferably, the instantaneous current values ​​of each feeder circuit are synchronously acquired at a preset sampling frequency by a current acquisition unit in the control and monitoring terminal room within the DC bus distribution cabinet, and the breaking status values ​​of the corresponding solid-state circuit breakers for each feeder circuit are read by a status reading unit. The breaking status values ​​include both broken and closed states. The vector construction module constructs a real-time current dynamic vector for each feeder circuit, using the rate of change of the instantaneous current value over time as the first dimension and the cumulative amount of the amplitude fluctuation range of the instantaneous current value within a preset time window as the second dimension. In a preferred embodiment, for any feeder loop, the instantaneous current value is linearly fitted within a preset time window to obtain the rate of change of the instantaneous current value over time, which is used as a component of the first dimension. The difference between the maximum and minimum values ​​of the instantaneous current value within the preset time window is used as the cumulative amount of the second dimension. The rate of change and the cumulative amount are combined to form the real-time current dynamic vector. The vector constructed in this way can compactly represent the dynamic trend and fluctuation intensity of the current. The offset quantization module obtains the current anomaly offset of each feeder loop based on the degree of difference between the real-time current dynamic vector of each feeder loop and the reference current dynamic vector of the corresponding feeder loop. Specifically, for any feeder loop, the Euclidean distance between the real-time current dynamic vector and the reference current dynamic vector is calculated, and the Euclidean distance is used as the current anomaly offset; wherein, the reference current dynamic vector is the average value of the historical current dynamic vectors constructed in the same manner under normal operating conditions for that feeder loop. As an improvement, the reference current dynamic vector is grouped according to the rated current level of the feeder circuits, and the average value of the historical current dynamic vectors within each group is calculated. This eliminates misjudgments caused by differences in normal behavior under different rated current levels, making the abnormal offset more sensitive and reliable. The distortion aggregation module, based on the current abnormal offset of each feeder circuit and the disconnection state value, uses the spatial distribution weights based on the topological connection relationship between each feeder circuit to perform weighted aggregation of the current abnormal offset of each feeder circuit, generating the overall current distortion factor of the DC bus distribution cabinet.Preferably, for any two feeder circuits, the spatial distribution weight between the two feeder circuits is calculated based on the physical connection distance between them within the DC bus distribution cabinet. This spatial distribution weight is inversely proportional to the physical connection distance. The current anomaly offset of each feeder circuit is multiplied by the complement of the corresponding feeder circuit's disconnection state value to obtain the effective anomaly offset of each feeder circuit. Then, the effective anomaly offsets of all feeder circuits are multiplied by the sum of the spatial distribution weights of the corresponding feeder circuit relative to all other feeder circuits, and summed to obtain the overall current distortion factor. By introducing spatial distribution weights on the topology, this system can consider the propagation effect of faults between adjacent circuits, obtaining a distortion factor reflecting the overall health status of the distribution cabinet with minimal computation. The risk ranking module calculates the sign change frequency of the first-order difference sequence of the current anomaly offset of each feeder circuit within the preset time window, and, combined with the overall current distortion factor, determines the fault risk priority coefficient of each feeder circuit. As a preferred method, for any feeder loop, a first-order difference operation is performed on the time series of the current anomaly offset within the preset time window to obtain the first-order difference sequence; the number of times the product of two adjacent difference values ​​in the first-order difference sequence is negative is counted, and the ratio of the number of times to the total length of the first-order difference sequence is used as the sign change frequency; the product of the sign change frequency and the overall current distortion factor is used as the fault risk priority coefficient of the feeder loop. This integrates the fluctuation instability of local anomalies with the global distortion degree, so that the fault risk priority coefficient can reflect both the abnormal fluctuation characteristics of the loop itself and its risk weight in the overall power distribution environment. The diagnostic output module outputs the fault diagnosis results corresponding to each feeder loop in descending order of the fault risk priority coefficients. Specifically, the fault risk priority coefficients of each feeder circuit are sorted in descending order to obtain a feeder circuit diagnosis queue. Feeder circuits are then sequentially selected from this queue. For each selected feeder circuit, the instantaneous current value is compared with the rated current value, and the disconnection state value is compared with the expected closure state value. The current comparison results and the state comparison results are combined to form the fault diagnosis result for the feeder circuit and output. This allows maintenance personnel to quickly identify the circuit most likely to fail based on risk priority and obtain intuitive comparative information to aid decision-making.As a further optimization of the present invention, after the offset quantization module obtains the current abnormal offset of each feeder circuit, it is further used to perform a difference calculation on the current abnormal offset of any feeder circuit and the current abnormal offset of the adjacent feeder circuit to obtain the offset difference degree of the feeder circuit; when the offset difference degree exceeds a preset difference threshold, the feeder circuit is marked as an arcing suspected circuit, and the solid-state circuit breaker corresponding to the feeder circuit is driven to perform a disconnection operation. The process of driving the disconnection operation includes: sending a disconnection trigger pulse signal to the drive circuit of the solid-state circuit breaker corresponding to the feeder circuit, the pulse width of the disconnection trigger pulse signal being proportional to the current abnormal offset degree; while sending the disconnection trigger pulse signal, locking the closing control signal of the solid-state circuit breaker corresponding to the feeder circuit until an external reset command is received. This scheme can promptly identify arcing risks and quickly perform protection disconnection when a drastic difference in abnormal offset degree is detected between adjacent circuits. At the same time, by using a pulse width proportional to the abnormal offset degree and a closing lockout mechanism, it balances the reliability of disconnection with the safety of preventing accidental reclosing.

[0005] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0006] By combining the rate of change of instantaneous current value over time and the cumulative amount of amplitude fluctuation within a preset time window into a two-dimensional real-time current dynamic vector, and comparing the difference with the reference current dynamic vector under normal operating conditions of the corresponding feeder circuit, the abnormal current offset is obtained. This allows for the simultaneous capture of abnormal offsets in both the trend and short-term energy fluctuation dimensions of the current signal. The current rate of change dimension is sensitive to rapid transient processes and can promptly reflect the current rise or fall at the initial stage of an arc or short circuit; the cumulative amount of short-term amplitude fluctuation has a good indicative ability for persistent irregular disturbances. When combined, under the Euclidean distance metric relative to the reference vector, even if the current amplitude has not yet reached the protection threshold, as long as its shape deviates from the historical normal dynamic pattern, a significant abnormal offset output will be generated. Based on this, the effective abnormal offsets of each circuit are weighted and aggregated using the spatial distribution weights based on the physical connection distance between each feeder circuit to generate an overall current distortion factor. This weight is inversely proportional to the actual distance between circuits, ensuring that abnormalities are mutually amplified between circuits physically adjacent within the same distribution cabinet, while mutual influence between circuits that are farther apart is suppressed. The resulting overall distortion factor not only reflects the degree of current anomaly in a single feeder circuit but also incorporates the spatial correlation effect of anomalies within the distribution cabinet's internal topology. When an arc or strong interference occurs in a circuit, even minor disturbances in neighboring circuits will increase the overall factor through weighted aggregation, thus forming a macroscopic measure of the risk of early fault propagation at the cabinet level. This avoids the perception blind spots caused by relying solely on isolated circuit thresholds. When ranking feeder circuits for fault risk, a quantitative index describing the oscillation characteristics of the anomaly offset is obtained by performing a first-order difference operation on the time series of current anomaly offset within a preset time window and statistically analyzing the sign change frequency. The sign change frequency reflects the frequency of alternating increases and decreases in the anomaly offset between adjacent time points. Fault modes such as intermittent faults, unstable contact resistance, or repeated arc reignition typically produce higher sign change frequencies, while smooth, continuous drift anomalies have lower sign change frequencies. The fault risk priority coefficient is constructed by multiplying the sign change frequency of each circuit by the overall current distortion factor. This allows circuits with high-frequency oscillation characteristics to have their priority amplified when the overall power distribution cabinet distortion level increases, while circuits with stable offset types receive lower priority coefficients for the same offset amplitude. Fault diagnosis results are output in descending order of this coefficient, prioritizing circuits with more urgent fault modes and those more likely to endanger adjacent equipment to maintenance personnel. This automates and rationally orchestrates diagnostic resources and processing sequences, improving the response speed and targeted handling of potential faults in the data center power distribution system. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0008] Figure 1 This is a schematic diagram of the structure of an intelligent power distribution equipment fault diagnosis system for data center scenarios;

[0009] Figure 2 This is a flowchart of the risk ranking module calculating the fault risk priority coefficient;

[0010] Figure 3 This is a schematic diagram comparing the real-time current dynamic vector of the feeder circuit with the reference current dynamic vector;

[0011] Figure 4 This is a graph showing the relationship between the weighted anomaly contribution of the feeder circuit and the interruption status value;

[0012] Figure 5 This is a priority ranking chart of feeder circuit fault risks;

[0013] Figure 6 It is a diagram showing the ratio of instantaneous current to rated current in the feeder circuit and the distribution of fault conditions. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] See Figure 1This invention provides an intelligent power distribution equipment fault diagnosis system for data center scenarios, comprising: a data acquisition module, a vector construction module, an offset quantization module, a distortion aggregation module, a risk ranking module, and a diagnostic output module. The data acquisition module acquires the instantaneous current values ​​of each feeder circuit within the DC bus distribution cabinet and the breaking status values ​​of the corresponding solid-state circuit breakers for each feeder circuit. The vector construction module constructs a real-time current dynamic vector for each feeder circuit, using the rate of change of the instantaneous current value over time as the first dimension and the cumulative amount of the amplitude fluctuation range of the instantaneous current value within a preset time window as the second dimension. The offset quantization module obtains the current anomaly offset of each feeder circuit based on the degree of difference between the real-time current dynamic vector of each feeder circuit and the reference current dynamic vector of the corresponding feeder circuit. The distortion aggregation module, based on the current anomaly offset and breaking status values ​​of each feeder circuit, utilizes the spatial distribution weights based on the topological connection relationship between each feeder circuit to perform weighted aggregation of the current anomaly offset of each feeder circuit, generating an overall current distortion factor for the DC bus distribution cabinet. The risk ranking module calculates the sign change frequency of the first-order difference sequence based on the current anomaly offset of each feeder circuit within a preset time window, and determines the fault risk priority coefficient of each feeder circuit by combining it with the overall current distortion factor. The diagnostic output module outputs the fault diagnosis results for each feeder circuit in descending order of its fault risk priority coefficient.

[0016] In practical implementation, the data acquisition module obtains the instantaneous current values ​​of each feeder circuit within the DC bus distribution cabinet. These instantaneous current values ​​are time-series data continuously collected at a preset sampling frequency. For any given feeder circuit, the vector construction module extracts data points within a preset time window from the time series of instantaneous current values. The width of the preset time window is pre-set based on the response characteristics of the solid-state circuit breaker. In one implementation, the preset time window is set to cover at least one power frequency cycle and is not greater than the duration of the fault arc characteristic. The vector construction module uses the instantaneous current values ​​within the preset time window to perform linear fitting, obtaining the rate of change of the instantaneous current value over time. The linear fitting employs the least squares method, using time as the independent variable and the instantaneous current value as the dependent variable. The coefficient of the first-order term obtained from the fitting is the rate of change, which serves as the first dimension component of the real-time current dynamic vector.

[0017] Simultaneously, the vector construction module iterates through all instantaneous current values ​​within a preset time window, determines the maximum and minimum values ​​of the instantaneous current within the preset time window, and uses the difference between the maximum and minimum values ​​as the cumulative amount of the amplitude fluctuation range of the instantaneous current within the preset time window. This cumulative amount serves as the second dimension component of the real-time current dynamic vector. The vector construction module combines the rate of change and the cumulative amount into a two-dimensional vector to obtain the real-time current dynamic vector of the feeder loop.

[0018] The offset quantization module stores the reference current dynamic vector for each feeder loop. This reference current dynamic vector is obtained by arithmetically averaging the historical current dynamic vectors corresponding to multiple historical preset time windows, using the same construction method as the real-time current dynamic vector, under normal operating conditions. In one implementation, the historical preset time window is selected from the operating period when no fault alarms occur in the feeder loop and the load rate is stable within the rated range. The offset quantization module calculates the Euclidean distance between the real-time current dynamic vector of the feeder loop and the reference current dynamic vector of the feeder loop, using this Euclidean distance as the current anomaly offset of the feeder loop.

[0019] The Euclidean distance is calculated as follows:

[0020]

[0021] in, This is the width of the preset time window, measured in seconds. This indicates the abnormal current deviation in the feeder circuit. This represents the rate of change component in the real-time current dynamic vector. This represents the cumulative component in the real-time current dynamic vector. This represents the rate of change component in the dynamic vector of the reference current. This represents the cumulative component in the dynamic vector of the reference current. and It is obtained directly through the aforementioned linear fitting and range calculation. and The two corresponding dimensional components in the reference current dynamic vector are read from the pre-stored vector mean record.

[0022] In some implementations, the reference current dynamic vector is maintained in groups based on the rated current level of the feeder circuits. That is, a group of feeder circuits with the same rated current level share the same set of reference current dynamic vectors, which is obtained by averaging the historical current dynamic vectors of all feeder circuits within that group. When calculating the current anomaly offset, the offset quantization module uses the corresponding reference current dynamic vector to calculate the Euclidean distance based on the rated current level of the feeder circuit.

[0023] See Figure 3The figure uses a two-dimensional coordinate system to display the two-dimensional components of the feeder loop current dynamic vector. The horizontal axis represents the rate of change of the instantaneous current value over time (unit: A / s), and the vertical axis represents the cumulative amount of the instantaneous current value's amplitude fluctuation within a preset time window (unit: A). Blue dots represent the distribution of historical current dynamic vectors. These blue dots show a clustering trend centered around approximately 0.05 A / s on the horizontal axis and approximately 0.5 A on the vertical axis, reflecting the statistical distribution range of the feeder loop's current dynamic characteristics under normal operating conditions. The black cross represents the reference current dynamic vector, with coordinates showing a rate of change of approximately 0.05 A / s and a cumulative amplitude fluctuation of approximately 0.5 A. It is located at the cluster center of the historical current dynamic vectors, reflecting the reference vector obtained based on the arithmetic mean of multiple sets of historical normal operating data.

[0024] The red pentagram represents the current real-time dynamic current vector, with a rate of change of approximately 0.21 A / s and a cumulative amplitude fluctuation of approximately 1.15 A, significantly deviating from the normal region where the reference current dynamic vector is located. The black dashed line between the red pentagram and the black cross represents the current anomaly deviation between the real-time current dynamic vector and the reference current dynamic vector. The value 0.67 marked on the dashed line is the calculated Euclidean distance between the two vectors, reflecting the magnitude of the current anomaly deviation.

[0025] In practical implementation, the distortion aggregation module acquires the current anomaly offset of each feeder circuit output by the offset quantization module, and the tripping status value of the corresponding solid-state circuit breaker for each feeder circuit output by the data acquisition module. For all feeder circuits within the DC bus distribution cabinet, a total of... One feeder loop, of which It is a positive integer. For the ... The feeder loop is defined as the first... The current anomaly deviation of each feeder loop is , define the first The disconnection status value of each feeder loop is ,in The value can be 0 or 1. This indicates that the solid-state circuit breaker is in the open state. This indicates that the solid-state circuit breaker is in the closed state. The distortion aggregation module calculates the first... The effective abnormal offset of the feeder loop is equal to the effective abnormal offset of the first feeder loop. Current deviation of each feeder loop With the The product of the complementary values ​​of the disconnection state values ​​of each feeder loop, where the complementary value of the disconnection state value is... .

[0026] The distortion aggregation module pre-stores the physical location coordinates of each feeder loop within the DC bus distribution cabinet. These physical location coordinates are three-dimensional coordinates with a preset reference point within the distribution cabinet as the origin. For any two different feeder loops, i.e., the first... The feeder loop and the first One feeder loop The distortion aggregation module is based on the first The physical location coordinates of the first feeder loop and the first Calculate the Euclidean distance from the physical location coordinates of the first feeder loop, and use the calculated Euclidean distance as the first... The feeder loop and the first Physical connection distance between each feeder loop The distortion aggregation module is based on the physical connection distance. Determine the first The feeder loop and the first The spatial distribution weights between feeder loops are inversely proportional to the physical connection distance. In one embodiment, the spatial distribution weights... Calculate as follows: ,in This is the preset minimum distance threshold. The value is determined based on the minimum center-to-center distance between adjacent feeder circuits within the DC bus distribution cabinet. In one implementation, the minimum distance threshold is... Set to 50 mm. By taking the larger value between the physical connection distance and the minimum distance threshold as the denominator, the spatial distribution weight is prevented from tending to infinity when the physical connection distance between the two feeder loops is too small.

[0027] The distortion aggregation module for the first Calculate the nth feeder loop. The sum of the spatial distribution weights of each feeder loop relative to all other feeder loops , The distortion aggregation module will... The effective abnormal offset of the feeder loop and the first The sum of the spatial distribution weights corresponding to each feeder loop Multiply to get the first... The weighted anomaly contribution of each feeder loop. The distortion aggregation module applies to all... The weighted anomaly contributions of each feeder circuit are summed to obtain the overall current distortion factor of the DC bus distribution cabinet. Overall current distortion factor The calculation method is shown in the following formula:

[0028]

[0029] in, This represents the equivalent amplitude of the overall current distortion in the distribution cabinet, with dimensions in amperes (A). This indicates the total number of feeder circuits within the DC bus distribution cabinet. Indicates the first The current deviation of each feeder loop, measured in amperes (A). Indicates the first The disconnection status value of each feeder loop. Indicates the first The feeder loop and the first The physical connection distance between each feeder loop This represents the minimum distance threshold. Indicates taking and The larger value in the range.

[0030] In some implementations, the distortion aggregation module can also obtain the physical connection distance by: pre-establishing a connection distance table between feeder circuit terminals, and directly obtaining the physical connection distance between any two feeder circuits by looking up the table. When the solid-state circuit breaker of a feeder circuit is in the open state, i.e., the open state value... At that time, the complement value of the discontinuity state value The effective abnormal offset is equal to the current abnormal offset, and the abnormal state of this feeder circuit can participate in aggregation normally; when the solid-state circuit breaker of a feeder circuit is in the open state, i.e., the open state value At that time, the complement value of the discontinuity state value The effective abnormal offset is zero, and the current abnormal offset of the feeder circuit does not participate in the calculation of the overall current distortion factor, thereby eliminating the interference of the disconnected circuit on the overall status judgment of the distribution cabinet.

[0031] See Figure 4 In the figure, the horizontal axis represents the feeder loop number, ranging from 0 to 500, and the vertical axis represents the weighted anomaly contribution. The legend distinguishes between two disconnection state values: green dots represent a disconnection state value of 1 (closed state), and red crosses represent a disconnection state value of 0 (disconnected state). According to this embodiment, the distortion aggregation module calculates the weighted anomaly contribution of each feeder loop based on the current anomaly offset of each feeder loop and the disconnection state value of the corresponding solid-state circuit breaker. The effective anomaly offset is not zero only when the disconnection state value is 0 (disconnected state), and participates in the aggregation of the overall current distortion factor.

[0032] The weighted anomaly contributions of the green dots in the figure are all close to zero, indicating that the current anomaly offset corresponding to the feeder loop in the closed state multiplied by the complement of the interrupted state value is zero, meaning that the anomaly offset of these feeder loops does not contribute to the overall current distortion factor. The interrupted feeder loops corresponding to the red crosses show a more dispersed and significantly higher weighted anomaly contribution than zero, with some feeder loops having a weighted anomaly contribution as high as close to 0.7, indicating that the current anomaly offset of these interrupted feeder loops is relatively high, and their contribution to the overall current distortion factor is significant when combined with their spatial distribution weights with other feeder loops.

[0033] Furthermore, judging from the distribution of the horizontal axis and the changing trend of the vertical axis, the weighted anomaly contribution of the discontinuous feeder circuit has no obvious clustering area and exhibits discrete distribution characteristics, reflecting the spatial characteristics of the physical location coordinates of the feeder circuit and its current anomaly offset.

[0034] In specific implementation, please refer to Figure 2 The risk ranking module obtains the current anomaly offset of each feeder loop from the offset quantization module, and the overall current distortion factor of the DC bus distribution cabinet from the distortion aggregation module. For any feeder loop, the risk ranking module extracts the time series of the feeder loop's current anomaly offset within a preset time window. The time series consists of multiple current anomaly offset values ​​arranged in order of sampling time. Let the th... The current anomaly offset time series of each feeder loop includes There are 1 sampling points, among which are positive integers and , No. The window width of the preset time window for each feeder loop is the same as the preset time window used in the vector construction module.

[0035] The risk ranking module for the first The current anomaly offset time series of the feeder loop is subjected to first-order difference operation to obtain a first-order difference series. The first-order difference operation is performed as follows: The time series of the feeder loop is processed by performing a first-order difference operation on the current anomaly offset time series. Current anomaly offset value at each sampling point With the Current anomaly offset value at each sampling point Find the difference to get the first... Each first-order difference value ,in The range of values ​​is to From all first-order difference values Constituting the first The first-order difference sequence of each feeder loop has a total length of [number] first-order difference sequences. .

[0036] The risk ranking module counts the number of times the product of two adjacent difference values ​​in a first-order difference sequence is negative. Two adjacent difference values ​​refer to the nth difference value in the first-order difference sequence. Each first-order difference value With the Each first-order difference value ,in The range of values ​​is to .when and When the product is negative, it indicates a sign change in the trend of abnormal current deviation, i.e., from increasing to decreasing or from decreasing to increasing. The risk ranking module counts the number of adjacent difference pairs with negative products, obtaining the first... Number of sign changes per feeder loop .

[0037] The risk ranking module will be the first Number of sign changes per feeder loop Divide by the The total length of the first-order difference sequence of each feeder loop , obtained the The symbol change frequency of the feeder loop. The risk ranking module will... Multiplying the sign change frequency of the first feeder loop by the overall current distortion factor yields the first... The fault risk priority coefficient of each feeder circuit. Fault risk priority coefficient for each feeder circuit The calculation method is shown in the following formula:

[0038]

[0039] in, Indicates the first Fault risk priority coefficient for each feeder circuit Indicates the first The number of times the product of adjacent difference values ​​is negative in the first-order difference sequence of a feeder loop. Indicates the first The total number of sampling points for the time series of current anomaly offset of each feeder loop. Indicates the first The total length of the first-order difference sequence of each feeder loop. This represents the overall current distortion factor of the DC bus distribution cabinet generated by the distortion aggregation module.

[0040] In some implementations, when the first Length of current anomaly offset time series for each feeder loop When the difference is less than or equal to 2, the length of the first-order difference sequence is 1, which cannot form adjacent difference value pairs. The risk ranking module will then... The sign change frequency of the first feeder loop is directly set to zero. The fault risk priority factor for each feeder circuit is zero.

[0041] The risk prioritization module calculates the fault risk priority coefficient for each feeder circuit in the DC bus distribution cabinet in the above manner, and outputs the fault risk priority coefficient of all feeder circuits to the diagnostic output module.

[0042] See Figure 5 In the graph, the vertical axis represents the fault risk priority coefficient of each feeder circuit, ranging from 0 to 0.7. The horizontal axis represents the feeder circuit numbers arranged in descending order of fault risk priority coefficient, ranging from 0 to 500, indicating that a total of 500 feeder circuits were sorted. The purple bars in the graph visually illustrate the distribution trend of the fault risk priority coefficients of all feeder circuits.

[0043] As shown in the graph, the fault risk priority coefficient decreases with increasing feeder loop serial number, exhibiting a smooth downward trend overall. Feeder loops with serial numbers from approximately 0 to 50 have relatively high fault risk priority coefficients, with a maximum value close to 0.65, indicating that these feeder loops have a high priority in risk ranking and significant fault concern value. As the serial number increases, the fault risk priority coefficient gradually decreases. The fault risk priority coefficient for feeder loops from 50 to 200 ranges from approximately 0.4 to 0.5, showing a slow downward trend. From 200 to 500, the fault risk priority coefficient decreases significantly, approaching zero, indicating that these feeder loops have a low fault risk.

[0044] In practice, the diagnostic output module obtains the fault risk priority coefficient of each feeder circuit from the risk ranking module and sorts all fault risk priority coefficients in descending order. This descending order arrangement forms a feeder circuit diagnostic queue, with the first element in the queue representing the feeder circuit with the highest fault risk priority coefficient and the last element representing the feeder circuit with the lowest. The diagnostic output module then sequentially retrieves feeder circuits from the queue, from the first element to the last.

[0045] For each feeder circuit retrieved from the feeder circuit diagnostic queue, the diagnostic output module obtains the comparison result between the instantaneous current value and the rated current value of the feeder circuit. The instantaneous current value is the instantaneous current value of the feeder circuit at the current moment, synchronously acquired by the data acquisition module at a preset sampling frequency. The rated current value is the pre-stored nominal rated current value of the feeder circuit. The diagnostic output module calculates the ratio of the instantaneous current value to the rated current value. When the ratio exceeds a preset overcurrent threshold, the current comparison result is an overcurrent state; when the ratio is lower than a preset undercurrent threshold, the current comparison result is an undercurrent state; when the ratio is between the preset undercurrent threshold and the preset overcurrent threshold, the current comparison result is a normal state. The preset overcurrent threshold is determined based on the overload protection characteristic setting value of the solid-state circuit breaker. In one embodiment, the preset overcurrent threshold is set to 1.2. The preset undercurrent threshold is determined based on the minimum operating current of the load carried by the feeder circuit. In one embodiment, the preset undercurrent threshold is set to 0.1.

[0046] The diagnostic output module also acquires the comparison results between the open state value and the expected closed state value of the feeder circuit. The open state value is the open state value of the solid-state circuit breaker corresponding to the feeder circuit, read by the data acquisition module through the state reading unit in the control monitoring terminal room. The open state value includes an open state and a closed state, represented by the value 0 and the value 1. The expected closed state value is the pre-configured state value that the solid-state circuit breaker should maintain under normal operating conditions. Under normal operating conditions, the expected closed state value is 1, indicating a closed state. The diagnostic output module compares the open state value with the expected closed state value. When the open state value equals the expected closed state value, the comparison result is a normal state; when the open state value does not equal the expected closed state value, the comparison result is an abnormal state, specifically, expected closure but actual open.

[0047] The diagnostic output module combines the current comparison results and status comparison results of the feeder circuits into a fault diagnosis result for each feeder circuit. The combined fault diagnosis result is a data record containing both current comparison result and status comparison result fields. The data record also includes the feeder circuit's identifier and fault risk priority coefficient. The diagnostic output module outputs the fault diagnosis results for each feeder circuit sequentially, according to the order in which they are retrieved from the feeder circuit diagnostic queue.

[0048] In some implementations, the data acquisition module synchronously acquires the instantaneous current values ​​of each feeder circuit at a preset sampling frequency via a current acquisition unit located in the control and monitoring terminal room within the DC bus distribution cabinet. The control and monitoring terminal room is an independent compartment within the DC bus distribution cabinet, and the current acquisition unit is a multi-channel synchronous sampling module installed within it. Each channel of the multi-channel synchronous sampling module is connected to each feeder circuit via a current sensor. The preset sampling frequency is determined based on the characteristic frequency of fault arcs that may occur in the feeder circuits; in one implementation, the preset sampling frequency is set to 10000 Hz. The current acquisition unit outputs the instantaneous current values ​​acquired from all channels to the buffer of the data acquisition module in a timestamp-aligned manner.

[0049] The data acquisition module also reads the breaking status values ​​of the corresponding solid-state circuit breakers for each feeder circuit through a status reading unit in the control and monitoring terminal room. The status reading unit is a digital input module installed in the control and monitoring terminal room. The input terminal of the digital input module is connected to the auxiliary contact signal line of each solid-state circuit breaker. The digital input module reads the on / off state of the auxiliary contacts and converts it into a breaking status value. When the auxiliary contact is open, the corresponding breaking status value is in the breaking state; when the auxiliary contact is closed, the corresponding breaking status value is in the closed state. The status reading unit and the current acquisition unit are synchronized using the same clock source, and the time when the status reading unit reads the breaking status value is aligned with the time when the current acquisition unit acquires the instantaneous current value.

[0050] See Figure 6 In the graph, the horizontal axis represents the feeder loop identifier, and the vertical axis represents the ratio of the instantaneous current value to the rated current value. There are six types of data points in the graph, corresponding to different combinations of current comparison results and status comparison results: overcurrent and normal status, overcurrent and abnormal status, normal current and normal status, normal current and abnormal status, undercurrent and normal status, and undercurrent and abnormal status. The overcurrent threshold is set to 1.2, and the undercurrent threshold is set to 0.1. These two key numerical ranges are marked by two horizontal dashed lines in the graph.

[0051] As shown in the figure, feeder circuits with overcurrent but normal status are indicated by red dots. The ratio of their instantaneous current to rated current is greater than the overcurrent threshold of 1.2, and the corresponding status comparison result is normal, indicating that although the instantaneous current value of this type of feeder circuit is high, the solid-state circuit breaker status meets expectations. Feeder circuits with overcurrent and abnormal status are indicated by red crosses. The current ratio also exceeds 1.2, but the status comparison result is abnormal, suggesting that the solid-state circuit breaker may not be closed or the expected closing state has not been achieved, indicating a potential anomaly.

[0052] Feeder circuits with normal current and normal status are indicated by blue solid dots. Their instantaneous current values ​​are between the undercurrent threshold of 0.1 and the overcurrent threshold of 1.2, and the status comparison results are normal, indicating that this type of feeder circuit is operating stably and its status meets expectations. Feeder circuits with normal current but abnormal status are indicated by blue crosses. The current ratio is within the normal range, but the status comparison is abnormal, suggesting that there is an abnormality in the circuit breaker status, and further diagnosis and confirmation are required.

[0053] Feeder circuits with undercurrent but normal status are indicated by light blue solid dots. Their current ratio is below the undercurrent threshold of 0.1 and their status is normal, possibly indicating a light load or insufficient load. Feeder circuits with undercurrent and abnormal status are indicated by light blue crosses. Their current ratio is below the undercurrent threshold and their status is abnormal, indicating a significant risk of abnormality in the feeder circuit.

[0054] In practice, the reference current dynamic vector is obtained by grouping feeder circuits according to their rated current levels, with feeder circuits of the same rated current level grouped together. For each group, the offset quantization module collects historical current dynamic vectors constructed in the same manner for all feeder circuits within that group under normal operating conditions. These historical current dynamic vectors consist of a set of two-dimensional vectors corresponding to multiple preset historical time windows. The offset quantization module then performs arithmetic mean calculations on both dimensions of all historical current dynamic vectors within the group, obtaining the average value of the rate of change component and the average value of the cumulative component. These two average values ​​are combined to form the reference current dynamic vector for the corresponding rated current level group. All feeder circuits within the same rated current level group share the same set of reference current dynamic vectors.

[0055] After obtaining the current anomaly offset of each feeder circuit, the offset quantization module is also used to perform arcing suspicion determination on any one feeder circuit. (This corresponds to the DC bus distribution cabinet...) The offset quantization module obtains the first feeder loop. The current anomaly offset of the feeder loop is obtained, and the current anomaly offset of the feeder loop is compared with that of the feeder loop. The current anomaly offset of a feeder circuit is measured in terms of its physical layout relative to adjacent feeder circuits. The physical adjacency is determined by the terminal arrangement order of the feeder circuits within the distribution cabinet; adjacent feeder circuits are one or two feeder circuits with adjacent terminal numbers. When two adjacent feeder circuits exist, the offset quantization module takes the average of the current anomaly offsets of the two adjacent feeder circuits as the current anomaly offset of the adjacent feeder circuit. Offset difference of individual feeder loops Calculate according to the following formula:

[0056]

[0057] in, Indicates the first Offset difference of each feeder loop Indicates the first Current deviation of each feeder loop Indicates the relationship with the first Current anomaly deviation of adjacent feeder loops, symbol This indicates the operation of taking the absolute value.

[0058] The offset quantization module will the first Offset difference of individual feeder loops The difference is compared with a preset difference threshold. The preset difference threshold is determined based on the statistical distribution of the current abnormal offset difference between adjacent feeder circuits in the DC bus distribution cabinet under normal operating conditions. In one embodiment, the preset difference threshold is set to 3.0, which is obtained by multiplying the maximum value of the current abnormal offset difference between adjacent feeder circuits in the normal operating history data by 1.5 times the safety margin.

[0059] When the Offset difference of individual feeder loops When the difference exceeds the preset threshold, the offset quantization module will... The feeder loop is marked as a suspected arcing loop, and a target is generated for the first... The disconnect drive command for the first feeder circuit is sent to the first... The drive circuit of the solid-state circuit breaker corresponding to each feeder circuit sends a disconnection trigger pulse signal to perform the disconnection operation.

[0060] In the specific implementation of the drive segmentation operation, the offset quantization module determines the data to be sent to the first... The pulse width of the breaking trigger pulse signal of the solid-state circuit breaker drive circuit corresponding to the feeder circuit is the same as that of the first feeder circuit. Current deviation of each feeder loop Proportional. The value of the proportionality coefficient is determined based on the gate drive characteristics of the power semiconductor devices in the solid-state circuit breaker. In one embodiment, the proportionality coefficient is set to 2 microseconds per unit current anomaly offset, ensuring that the minimum pulse width of the disconnection trigger pulse signal can reliably trigger the IGBT or MOSFET devices within the solid-state circuit breaker to turn on and off, while a larger pulse width provides stronger drive energy to cope with potentially increased bus voltage stress when the current anomaly offset is high. The offset quantization module sends the disconnection trigger pulse signal to the first... The closing control circuit of the solid-state circuit breaker corresponding to each feeder circuit outputs a lockout signal, forcing the closing control signal to an inactive level. This lockout remains until the offset quantization module receives an external reset command from an external operating terminal. The external reset command is manually triggered by maintenance personnel through the human-machine interface in the control and monitoring terminal room after confirming that the arc flash hazard has been eliminated. Upon receiving the external reset command, the offset quantization module removes the lockout on the closing control signal, restoring the solid-state circuit breaker to its normal closing response capability.

[0061] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A fault diagnosis system for intelligent power distribution equipment in data center scenarios, characterized in that, include: The data acquisition module acquires the instantaneous current value of each feeder circuit in the DC bus distribution cabinet and the breaking status value of the corresponding solid-state circuit breaker of each feeder circuit. The vector construction module is used to construct a real-time current dynamic vector for each feeder circuit, with the rate of change of the instantaneous current value over time as the first dimension and the cumulative amount of the amplitude fluctuation range of the instantaneous current value within a preset time window as the second dimension. The offset quantization module obtains the current anomaly offset of each feeder circuit based on the degree of difference between the real-time current dynamic vector of each feeder circuit and the reference current dynamic vector of the corresponding feeder circuit. The distortion aggregation module, based on the current abnormal offset of each feeder circuit and the disconnection state value, uses the spatial distribution weights based on the topological connection relationship between each feeder circuit to perform weighted aggregation of the current abnormal offset of each feeder circuit, and generates the overall current distortion factor of the DC bus distribution cabinet. The risk ranking module calculates the sign change frequency of the first-order difference sequence based on the current abnormal offset of each feeder circuit within the preset time window, and determines the fault risk priority coefficient of each feeder circuit in conjunction with the overall current distortion factor. The diagnostic output module outputs the fault diagnosis results for each feeder circuit in descending order of the fault risk priority coefficient.

2. The intelligent power distribution equipment fault diagnosis system for data center scenarios according to claim 1, characterized in that, Using the rate of change of the instantaneous current value over time as the first dimension, and the cumulative amount of the amplitude fluctuation range of the instantaneous current value within a preset time window as the second dimension, a real-time current dynamic vector for each feeder loop is constructed, including: For any feeder loop, the instantaneous current value is linearly fitted within the preset time window to obtain the rate of change of the instantaneous current value with time, which is used as a component of the first dimension. The difference between the maximum and minimum values ​​of the instantaneous current within the preset time window is used as the cumulative amount of the second dimension, and the rate of change and the cumulative amount are combined to form the real-time current dynamic vector.

3. The intelligent power distribution equipment fault diagnosis system for data center scenarios according to claim 2, characterized in that, Based on the degree of difference between the real-time current dynamic vector of each feeder loop and the reference current dynamic vector of the corresponding feeder loop, the current anomaly offset of each feeder loop is obtained, including: For any feeder loop, calculate the Euclidean distance between the real-time current dynamic vector and the reference current dynamic vector, and use the Euclidean distance as the current anomaly offset. The reference current dynamic vector is the average value of the historical current dynamic vectors constructed in the same way under normal operating conditions of the feeder circuit.

4. The intelligent power distribution equipment fault diagnosis system for data center scenarios according to claim 3, characterized in that, Based on the current anomaly deviation and the disconnection status value of each feeder circuit, and utilizing the spatial distribution weights based on the topological connection relationship between each feeder circuit, the current anomaly deviation of each feeder circuit is weighted and aggregated to generate the overall current distortion factor of the DC bus distribution cabinet, including: For any two feeder circuits, the spatial distribution weight between the two feeder circuits is calculated based on the physical connection distance between the two feeder circuits in the DC bus distribution cabinet. The spatial distribution weight is inversely proportional to the physical connection distance. The effective abnormal offset of each feeder loop is obtained by multiplying the current abnormal offset of each feeder loop by the complement of the interruption state value of the corresponding feeder loop. The overall current distortion factor is obtained by multiplying the effective abnormal offset of all feeder loops by the sum of the spatial distribution weights of the corresponding feeder loops relative to all other feeder loops.

5. The intelligent power distribution equipment fault diagnosis system for data center scenarios according to claim 4, characterized in that, Based on the first-order differential sequence of the current anomaly offset of each feeder circuit within the preset time window, the sign change frequency of the first-order differential sequence is calculated, and combined with the overall current distortion factor, the fault risk priority coefficient of each feeder circuit is determined, including: For any feeder loop, the first-order difference operation is performed on the time series of the current anomaly offset within the preset time window to obtain the first-order difference sequence. The frequency of negative product of two adjacent difference values ​​in the first-order difference sequence is counted, and the ratio of the number of times the product of two adjacent difference values ​​in the first-order difference sequence is used as the sign change frequency. The product of the symbol change frequency and the overall current distortion factor is used as the fault risk priority coefficient of the feeder circuit.

6. The intelligent power distribution equipment fault diagnosis system for data center scenarios according to claim 5, characterized in that, Based on the descending order of the fault risk priority coefficients for each feeder circuit, the fault diagnosis results for each feeder circuit are output sequentially, including: The fault risk priority coefficients of each feeder circuit are sorted in descending order to obtain the feeder circuit diagnosis queue; Feeder circuits are sequentially retrieved from the feeder circuit diagnostic queue. For each retrieved feeder circuit, the current comparison result between the instantaneous current value and the rated current value of the feeder circuit, as well as the state comparison result between the disconnection state value and the expected closure state value, are obtained. The current comparison result and the state comparison result are combined to form the fault diagnosis result of the feeder circuit and then output.

7. The intelligent power distribution equipment fault diagnosis system for data center scenarios according to claim 1, characterized in that, Obtain the instantaneous current values ​​of each feeder circuit in the DC bus distribution cabinet and the breaking status values ​​of the corresponding solid-state circuit breakers for each feeder circuit, including: The instantaneous current values ​​of each feeder circuit are synchronously collected by the current acquisition unit in the control and monitoring terminal room of the DC bus distribution cabinet at a preset sampling frequency. The status reading unit in the control and monitoring terminal room reads the disconnection status value of the solid-state circuit breaker corresponding to each feeder circuit. The disconnection status value includes the disconnection state and the closed state.

8. The intelligent power distribution equipment fault diagnosis system for data center scenarios according to claim 3, characterized in that, The reference current dynamic vector is obtained by grouping the feeder circuits according to their rated current levels and then calculating the average value of the historical current dynamic vector within each group.

9. The intelligent power distribution equipment fault diagnosis system for data center scenarios according to claim 1, characterized in that, The system further includes: an offset quantization module, which, after obtaining the current anomaly offset of each feeder loop, is also used for: For any feeder loop, the current abnormality offset of the feeder loop is calculated by the difference between the current abnormality offset of the feeder loop and the current abnormality offset of the adjacent feeder loop to obtain the offset difference of the feeder loop. When the offset difference exceeds a preset difference threshold, the feeder circuit is marked as an arc suspected circuit, and the solid-state circuit breaker corresponding to the feeder circuit is driven to perform a disconnection operation.

10. The intelligent power distribution equipment fault diagnosis system for data center scenarios according to claim 9, characterized in that, Driving the solid-state circuit breaker corresponding to the feeder circuit to perform a disconnection operation includes: Send a disconnection trigger pulse signal to the drive circuit of the solid-state circuit breaker corresponding to the feeder circuit, wherein the pulse width of the disconnection trigger pulse signal is proportional to the current abnormality offset. While sending the disconnection trigger pulse signal, the closing control signal of the solid-state circuit breaker corresponding to the feeder circuit is locked until an external reset command is received.

Images