A power distribution cabinet intelligent monitoring method and system

Abstract

The present application relates to power distribution cabinet safety monitoring technical field, especially in kind of power distribution cabinet intelligent monitoring method and system. The method comprises: collecting the multi-dimensional data signal of bus area, the degree of deviation of the current signal characteristics and the health vector learned by clustering is quantified through Mahalanobis distance, and the original fault risk score is obtained; the signal quality is evaluated by introducing spectrum analysis, and the signal data quality score is obtained; the original fault risk score and the signal quality score are weighted to obtain the weighted fault risk score. By clustering the historical weighted fault risk score, a dynamic threshold that can adapt to the normal working condition drift of the equipment is constructed. The weighted fault risk score is compared with the dynamic threshold, and the fault condition of the bus area is evaluated according to the comparison result, so as to realize the intelligent monitoring of power distribution cabinet. The present application realizes the accurate early warning of early hidden faults such as partial discharge, and significantly reduces the false positive rate.

Description

Technical Field

[0001] This invention relates to the field of power distribution cabinet safety monitoring technology, and in particular to a method and system for intelligent monitoring of power distribution cabinets. Background Technology

[0002] As a core hub device in the power system, the safe and stable operation of the switchgear is of paramount importance. Due to long-term exposure to electrical, thermal, and mechanical stresses, the insulation performance of its internal busbar connection areas is prone to premature degradation, leading to partial discharge. Partial discharge is a precursor to serious faults, characterized by weak signals and rapid development; therefore, accurate early intelligent monitoring is a key technology for ensuring power grid safety.

[0003] Traditional power distribution cabinet monitoring mainly relies on manual inspections or fixed threshold alarms based on a single signal. These methods suffer from poor real-time performance and high subjectivity, and cannot adapt to the dynamic drift of normal operating parameters of the power distribution cabinet due to factors such as power grid load and seasonal environment. They are also prone to generating a large number of false alarms under normal operating conditions, affecting the effectiveness of monitoring.

[0004] To address this, the industry has begun employing multidimensional signal fusion analysis technology, collecting various signals such as temperature, current, and sound waves to more comprehensively assess equipment status. Among these, the Mahalanobis distance algorithm has gained application due to its ability to effectively measure changes in the synergistic relationships between multidimensional data. This algorithm assesses fault risk by constructing a data model under normal conditions and calculating the deviation of the current signal vector from this model, rather than relying on the absolute value of a single signal, theoretically improving sensitivity to minor faults.

[0005] However, in the complex electromagnetic environment of distribution cabinets, the Mahalanobis distance algorithm has inherent limitations: while extremely sensitive to all signals deviating from the normal model, it cannot effectively distinguish whether the deviation is caused by a genuine early fault or by false fault signals such as external electromagnetic interference or sensor momentary jitter. This makes monitoring methods relying solely on Mahalanobis distance still face a high risk of false alarms, making it difficult to ensure both high sensitivity and high reliability. Summary of the Invention

[0006] To address the technical problem that the aforementioned multi-dimensional signal fusion combined with Mahalanobis distance monitoring method, while improving sensitivity to weak faults, still faces a high risk of false alarms due to its inability to distinguish between signal deviations originating from real early faults and spurious fault signals in complex electromagnetic environments, this invention provides solutions in the following aspects.

[0007] In a first aspect, the present invention provides an intelligent monitoring method for a power distribution cabinet, the method comprising the steps of:

[0008] Based on the multidimensional signal data, the feature vector at the current moment is obtained. Clustering is performed according to the distance between all feature vectors within the set time period to which the current moment belongs. The cluster center of the densest cluster is taken as the health vector at the current moment. The Mahalanobis distance between the feature vector and the health vector at the current moment is calculated to obtain the original fault risk score at the current moment. Based on the spectrum of the multidimensional signal data of the bus region within the set time period, the amplitude stability and time continuity of the spectrum are obtained, and the signal data quality score corresponding to the current moment is determined accordingly. The signal data quality score is positively correlated with both amplitude stability and time continuity. The signal data quality score and the original fault risk score are weighted and fused to obtain the weighted fault risk score at the current moment. Based on the statistical distribution characteristics of multiple weighted fault risk scores within the set time period to which the current moment belongs, the dynamic threshold at the current moment is determined. The weighted fault risk score at the current moment is compared with the dynamic threshold, and the fault status of the bus region is evaluated based on the comparison result to achieve intelligent monitoring of the distribution cabinet.

[0009] This invention first addresses the problem of traditional methods, which rely on fixed baseline models and cannot adapt to fluctuations in normal equipment operation, by clustering recent data to dynamically generate health vectors. Second, it introduces signal data quality scores based on spectral characteristics to weight the original risk scores, resolving the core issue of the Mahalanobis distance algorithm's high false alarm rate in industrial settings due to its oversensitivity to noise, interference, and other pseudo-fault signals. Finally, it utilizes statistical distribution characteristics to generate dynamic thresholds, enabling alarm limits to follow the normal fluctuations of the equipment in real time, avoiding missed or false alarms caused by fixed thresholds. Overall, this method achieves high sensitivity in detecting early, hidden faults and high robustness in suppressing interference signals.

[0010] Preferably, the original fault risk score at the current moment Satisfying the relation:

[0011] ;

[0012] in, It is the first The feature vector at time step; It is the first Health vector at any given time; It is a transpose operation; It is the first The inverse of the covariance matrix at time t; It is the standard normalization function.

[0013] This invention defines the specific mathematical expression of Mahalanobis distance, thus limiting the method for calculating the original fault risk score. This invention provides a repeatable and verifiable standardized calculation basis, ensuring the rigor and consistency of the original fault risk score quantification process, and facilitating engineering implementation and algorithm verification.

[0014] Preferably, the step of obtaining the amplitude stability and temporal continuity of the spectrum based on the spectrum of multidimensional signal data of the bus region within the set time period, and determining the signal data quality score corresponding to the current time, includes: performing a Fourier transform on the multidimensional signal data within the set time period to obtain the spectrum of each dimension; extracting the amplitude and timestamp of multiple peaks in the spectrum of each dimension; and calculating the signal data quality score based on the standard deviation of the amplitude of the multiple peaks and the maximum interval of the timestamps of the multiple peaks.

[0015] This invention extracts peak amplitude and timestamp features from the spectrum through Fourier transform, which can provide a deeper understanding of the stability and continuity of the signal, thereby more effectively identifying signal anomalies caused by sensor jitter or external electromagnetic interference, and providing a more reliable basis for quality judgment for subsequent weighted fusion.

[0016] Preferably, the signal data quality score Satisfying the relation:

[0017] ;

[0018] in, It is the first Within the set time period, the first The standard deviation of the peak amplitude in the spectrum of 3D signal data; Standard deviation Normalization Preset reference standard deviation for dimensional signal data; It is a preset micro value; It is the total number of dimensions of the signal data; It is the first Within the set time period, the first The maximum time interval in the timestamp of the peak in the spectrum of 3D signal data; It is the first Set the total duration of the time period.

[0019] This invention normalizes and integrates the fluctuation of signal amplitude and the sparsity of time into a single score. Compared with relying on a single indicator for judgment, this multi-dimensional quantification method can more comprehensively and accurately evaluate signal quality.

[0020] Preferably, the weighted fusion of the signal data quality score and the original fault risk score is performed by multiplying the original fault risk score by the signal data quality score.

[0021] Preferably, determining the dynamic threshold at the current moment includes: performing distance-based clustering on multiple weighted fault risk scores within the set time period to obtain the highest density cluster; and determining the maximum deviation between each data point in the highest density cluster and the cluster center as the dynamic threshold at the current moment.

[0022] This invention constructs a boundary that closely matches the range of risk value fluctuations under normal equipment operating conditions. Compared to traditional statistical thresholds based on the mean and fixed standard deviation multiples, this method is insensitive to outliers and can more accurately define the boundary between normal and abnormal values.

[0023] Preferably, the densest cluster refers to the cluster with the smallest intra-cluster dispersion; the intra-cluster dispersion is calculated based on the average distance between all feature vectors in each cluster and the corresponding cluster center.

[0024] Preferably, the multiple acquisition dimensions include temperature, current, and acoustic signal dimensions.

[0025] Preferably, comparing the weighted fault risk score at the current moment with the dynamic threshold and assessing the fault status of the bus region based on the comparison result includes: calculating the difference between the weighted fault risk score at the current moment and the weighted fault risk score at the previous moment; when the difference is greater than the dynamic threshold, determining that a fault has occurred in the bus region, and triggering an alarm or controlling the switch of the bus region to disconnect.

[0026] In a second aspect, the present invention provides an intelligent monitoring system for a power distribution cabinet, the system comprising a memory and a processor, the memory storing computer program instructions, which, when executed by the processor, implement the intelligent monitoring method for a power distribution cabinet according to the first aspect of the present invention.

[0027] By adopting the above technical solution, a computer program for intelligent monitoring of power distribution cabinets according to the first aspect of the present invention is generated and stored in a memory so that it can be loaded and executed by a processor, thereby creating a terminal device based on the memory and the processor for convenient use.

[0028] The beneficial effects of this invention are as follows: The core innovation of this invention lies in the construction of a dual dynamic adaptive intelligent monitoring framework. It abandons the traditional fixed benchmarks and fixed thresholds, dynamically learning the health status model of the equipment through clustering algorithms, and adaptively generating alarm thresholds based on the statistical distribution of recent risk values. Compared with traditional statistical thresholds based on mean and fixed standard deviation multiples, it can more accurately define the boundary between normal and abnormal conditions. This invention introduces a signal quality assessment mechanism based on spectrum analysis to weight and correct the original risk scores, thereby accurately solving the core technical problem of high-sensitivity monitoring algorithms being susceptible to interference and generating a large number of false alarms in complex industrial environments, achieving highly sensitive and reliable monitoring of early faults in distribution cabinets. Attached Figure Description

[0029] Figure 1 A flowchart illustrating an intelligent monitoring method for a power distribution cabinet provided in an embodiment of the present invention;

[0030] Figure 2 This is a structural block diagram of an intelligent monitoring system for a power distribution cabinet provided in an embodiment of the present invention. Detailed Implementation

[0031] The first aspect of this invention provides a method for intelligent monitoring of power distribution cabinets, such as... Figure 1 As shown, the method includes steps S100-S600:

[0032] Step S100: Acquire multi-dimensional signal data collected along multiple acquisition dimensions in the busbar area within the distribution cabinet.

[0033] It should be noted that the busbar area of ​​the distribution cabinet is a frequent site of insulation degradation due to the long-term coupled effects of electrical, thermal, and mechanical stresses. The most direct physical manifestation of early insulation degradation is partial discharge. Partial discharge simultaneously generates observable signals in multiple physical dimensions: firstly, repeated breakdown at the discharge point generates Joule heating, leading to a slight but continuous increase in local temperature; secondly, partial discharge is essentially a brief pulse current, which superimposes a high-frequency component onto the power frequency current, altering the characteristics of the total current; and thirdly, the rapid movement of charges during the discharge process excites ultrasonic waves in the surrounding medium. Therefore, by simultaneously monitoring the temperature, current, and acoustic wave signals in the main line area of ​​each busbar, a multi-dimensional fault characteristic space can be constructed, significantly improving the accuracy and reliability of fault early warning and identification.

[0034] Specifically, infrared temperature sensors, high-frequency current transformers, and ultrasonic sensors are deployed in the main line area of ​​each busbar row to collect temperature, current, and acoustic signals in real time and synchronously. Regarding sensor deployment, a preferred implementation is to focus on key connection points prone to faults, such as busbar lap joints and important bolt connections, for each phase (A, B, C) of the distribution cabinet. Each key monitoring point is equipped with a monitoring unit integrating an infrared temperature sensor, a high-frequency current transformer, and an ultrasonic sensor.

[0035] As a preferred implementation, the acquisition frequency of each sensor can be set to once per second to ensure the real-time and continuous nature of the data. To eliminate the potential impact of differences in physical dimensions and numerical ranges between the acquisition dimensions on subsequent calculations, the raw multidimensional signal data acquired in real time needs to undergo max-min normalization processing within each dimension, mapping the data to a unified... The normalization process can be performed based on the extreme values ​​of historical data accumulated from the long-term operation of the distribution cabinet under normal working conditions.

[0036] Thus, multidimensional signal data were acquired from the busbar area within the distribution cabinet, collected along multiple acquisition dimensions, and normalized.

[0037] Step S200: Obtain the feature vector at the current moment based on the multidimensional signal data; perform clustering according to the distance between all feature vectors within the set time period to which the current moment belongs; take the cluster center of the densest cluster as the health vector at the current moment; calculate the Mahalanobis distance between the feature vector at the current moment and the health vector to obtain the original fault risk score at the current moment.

[0038] It should be noted that during healthy operation, the multidimensional signal data of the distribution cabinet busbar area exhibit a stable and predictable intrinsic correlation pattern; for example, an increase in load current is accompanied by a proportional increase in temperature. Early faults such as partial discharge disrupt this inherent physical correlation, introducing new, weak abnormal data patterns. Relying solely on the absolute value of a single signal for threshold judgment can easily lead to missed detections due to subtle signal changes. Therefore, this invention introduces Mahalanobis distance, the core of which lies in using the covariance matrix to measure whether the cooperative relationship between multidimensional signal data points has been disrupted. Even if the signal values ​​themselves do not change significantly, as long as their combination pattern violates the learned healthy correlation pattern, the Mahalanobis distance will be significantly amplified, thereby enabling accurate identification of early, weak faults that are difficult to detect using conventional methods.

[0039] Specifically, taking a specific busbar row as an example, the processing procedure for other busbar rows is the same. First, based on the normalized multidimensional signal data at the current moment, a feature vector is constructed. This feature vector contains the normalized data for each dimension at that moment. Second, to define the health status of the distribution cabinet busbar area, a time window needs to be set as a set time period. The length of this set time period aims to achieve a balance between statistical robustness and real-time monitoring. For example, this time period can preferably include the most recent 50 sampling points, including the current moment. If the time period is too short, such as less than 30 points, insufficient sample size may lead to an unstable covariance matrix, affecting the accuracy of the Mahalanobis distance; if the time period is too long, such as more than 100 points, it will reduce the sensitivity to sudden faults, leading to response delays. Therefore, 50 sampling points is an optimal value, ensuring that clustering and covariance calculation have sufficient statistical significance while also ensuring that the system can quickly respond to changes in the current state. Implementers can set this value according to their needs.

[0040] Then, clustering is performed on all feature vectors within the set time period. In this embodiment, the k-means clustering algorithm can be used. The number of clusters... Determining the optimal clustering method is crucial to the clustering results. Preferably, the optimal method can be... The value is preset to a small, fixed integer, for example This setting is based on the fact that, within a relatively short set period, the operating status of the bus can usually be categorized into a few modes, such as a stable and healthy state, a normal operating condition with fluctuations or slight disturbances, and a potentially abnormal state. Therefore, the setting... It can effectively distinguish these core states. Implementers can configure it according to their needs. Value. The k-means clustering algorithm is a current technology and will not be discussed in detail here.

[0041] After obtaining the clusters, to identify the cluster representing the most stable operating state within the given time period, it is necessary to determine the densest cluster. In this embodiment, the densest cluster refers to the cluster with the smallest intra-cluster dispersion, which is calculated based on the average distance between all feature vectors within each cluster and their corresponding cluster centers. The smallest dispersion of a cluster means that the data points within that cluster are most concentrated and best represent the operating state within the given time period. The cluster center of the densest cluster is used as the health vector at the current moment. Simultaneously, based on all feature vectors within the given time period, the covariance matrix is ​​calculated, and its inverse matrix also needs to be calculated. Calculating the covariance matrix and its inverse matrix is ​​existing technology and will not be elaborated upon here.

[0042] Based on the above logic, the original fault risk score at the current moment is constructed using Mahalanobis distance. This original fault risk score satisfies the following relationship:

[0043] ;

[0044] in, It is the first The original fault risk score at any given time. It is the first The feature vector at time step; It is the first Health vector at any given time; It is a transpose operation; It is the first The inverse of the covariance matrix at time t; It is a standard normalization function used to map the calculation results to... Interval.

[0045] In this relation, the difference vector This intuitively reflects the original deviation of the current operating state from a healthy baseline. Because early fault signals are weak, the values ​​of each component within this difference vector may be very small. The core lies in introducing the inverse covariance matrix. Weighting the original bias vector is mathematically equivalent to projecting the data into a feature space where linear correlations between data dimensions are eliminated and variance is normalized. Therefore, the transformed vector... The measurement focuses on the degree of pure anomaly after removing normal correlation effects and undergoing volatility correction. If, at the current moment, the signal deviation in one dimension of the original deviation vector perfectly conforms to the normal correlation between the signal deviations in other dimensions, then after transformation, the value of that corresponding dimension will become very small, indicating a low degree of anomaly in that dimension. Conversely, if the deviation pattern violates normal correlation, the transformed value will become very large, indicating a high degree of anomaly in that dimension. Finally, through inner and outer product operations, the multidimensional anomaly measures are integrated into a single scalar risk value. This enables sensitive detection of early, weak fault signals.

[0046] At this point, the original fault risk score for the current moment has been obtained.

[0047] Step S300: Based on the spectrum of multidimensional signal data of the bus region within the set time period, obtain the amplitude stability and time continuity of the spectrum, and determine the signal data quality score corresponding to the current moment.

[0048] It should be noted that a high initial fault risk score calculated in step S200 may originate from a genuine partial discharge fault, but it could also be caused by external electromagnetic interference, sensor momentary jitter, or other factors, resulting in a false fault signal. To reduce false alarms caused by such false faults, the quality of the signal source must be assessed before adopting the risk score. Genuine partial discharge signals typically exhibit a certain degree of energy concentration and periodicity in the frequency domain, while noise signals such as electromagnetic interference exhibit chaotic energy distribution and no periodicity in the frequency domain. Therefore, by analyzing the spectral characteristics of multidimensional signal data within the current set time period, signal quality can be effectively quantified, distinguishing between genuine and false faults.

[0049] Specifically, first, Fourier transform is performed on the multidimensional signal data within a set time period to obtain the spectrum of each dimension; then, the amplitude and timestamp information of multiple peaks are extracted from the spectrum of each dimension; finally, the signal data quality score is calculated based on the standard deviation of the amplitude of multiple peaks and the maximum interval of the timestamps of multiple peaks.

[0050] Based on the above logic, the signal data quality score satisfies the following relationship:

[0051] ;

[0052] in, It is the first Signal data quality score at any given time. It is the first Within the set time period, the first The standard deviation of the peak amplitude in the spectrum of 3D signal data; Standard deviation Normalization Preset reference standard deviation for dimensional signal data; This is a preset microvalue used to prevent the denominator of the first ratio from being 0. It can be set to 0.001 or as needed. It is the total number of dimensions of the signal data; It is the first Within the set time period, the first The maximum time interval in the timestamp of the peak in the spectrum of 3D signal data; It is the first The total duration of the time period is set, and since the aforementioned sampling frequency is once per second and the time period has 50 sampling points, therefore... Second.

[0053] In this relationship, the first term is the ratio. Reflects the number of times within the set time period at the current moment. This refers to the dispersion of signal energy distribution in the frequency domain. A larger value indicates more irregular peak amplitudes and a higher likelihood that the signal is noise. Furthermore, since real fault signals exhibit periodicity, their spectra will show continuous peaks. Therefore, a central tendency index for the time distribution of these peaks, i.e., the second ratio, is introduced. This ratio reflects the temporal dispersion of the spectral peaks. A larger value indicates a more irregular time interval between peak occurrences and a poorer periodicity of the signal. When the... When the dimensional signal is a pseudo-fault signal, both ratios usually increase simultaneously, causing the product term within the parentheses to increase. Finally, subtracting the average of this product term over all dimensions from 1 ensures that higher signal quality and more regular spectral characteristics in each dimension are achieved. The closer the value is to 1, the worse the signal quality in each dimension. The closer the value is to 0.

[0054] It should be added that, Standard deviation A preset reference standard deviation is used for normalization. This reference standard deviation aims to provide a reasonable benchmark value representing the upper limit of fluctuation under historical operating conditions. In different embodiments of the present invention, this reference standard deviation can be determined in various ways:

[0055] As a preferred implementation method, This can be the maximum value of the standard deviation calculated over all historical time periods, dynamically recorded and continuously updated during system operation. This method is simple to implement and can adapt to the long-term operating characteristics of the system.

[0056] As another alternative implementation method, This can be done during the debugging phase before system deployment by performing offline analysis on a large amount of historical data under normal operating conditions, with a preset fixed statistical value such as the historical maximum value or percentile value.

[0057] As a specific, but computationally complex, alternative implementation method, The method for determining is as follows: based on the first Taking time as an example, suppose the first time... The time period is set to the [number]th [time period]. The 50 seconds immediately preceding the specified time, combined with the previous sampling frequency of once per second, correspond to 50 sampling points, i.e., 50 data points at each of the 50 time points. Each data point contains multi-dimensional data. First, the local standard deviation is calculated for the ... For each of the 50 sampling points within a given time period, an analysis time period is constructed, with each sampling point serving as the analysis time. Then, the analysis is performed on the 1st sampling point within each time period. The signal data undergoes a Fourier transform to extract the amplitude information of all peaks in the spectrum and calculates the standard deviation of the peak amplitude within that time period. This operation yields 50 local standard deviations corresponding to 50 sampling points. Then, the maximum value is extracted from these local standard deviations as... .

[0058] As a preferred implementation, in the initial stage of system startup or when data is insufficient, or when there are not enough data points within a set time period, an auxiliary judgment mechanism based on a historical database can be introduced. Specifically, this may include: extracting basic operating condition information such as total current and voltage at the current moment; filtering out stable operating periods similar to the current operating conditions from the equipment's historical database; and using the spectral characteristic parameters of these historically similar time periods, such as standard deviation and statistical values ​​of time intervals, to assist in calculating or correcting the signal data quality score.

[0059] At this point, the signal data quality score corresponding to the current moment has been determined.

[0060] Step S400: The signal data quality score and the original fault risk score are weighted and fused to obtain the weighted fault risk score at the current moment.

[0061] It should be noted that since the original fault risk score at each moment is quantified by analyzing the signal data change characteristics within a time period, after considering the false fault situation in step S300 and quantifying the signal data quality score, the original risk score needs to be weighted and corrected to effectively suppress false alarms caused by false fault signals.

[0062] Specifically, for any given time, the original fault risk score for that time is weighted using the signal data quality score over a preset time period to calculate a weighted fault risk score, which satisfies the following relationship:

[0063] ;

[0064] in, It is the first Fault risk score at any given moment; It is the first The original fault risk score at any given time; It is the first Signal data quality score at any given time.

[0065] In this relation, As a weighting coefficient, its value reflects the reliability of the original fault risk score at the current moment. A value approaching 1 indicates good signal data quality; in this case, the original fault risk score should be maintained. Approaching Conversely, when A value approaching 0 indicates poor signal data quality, which may be a false fault. In this case, the original fault risk score should be significantly reduced. Less than This reduces false alarms.

[0066] At this point, the weighted fault risk score for the current moment has been obtained.

[0067] Step S500: Based on the statistical distribution characteristics of multiple weighted fault risk scores within the set time period to which the current time belongs, determine the dynamic threshold for the current time.

[0068] It should be noted that the normal operating conditions of the distribution cabinet can drift due to factors such as external power grid load and seasonal environmental conditions, causing the normal baseline of the weighted fault risk score to change dynamically. Using a fixed warning threshold is difficult to adapt to this change and is prone to false alarms or missed alarms. Therefore, by analyzing the statistical distribution characteristics of recent historical risk scores, an adaptive warning threshold can be dynamically constructed to fit the current normal operating state boundary in real time.

[0069] Specifically, a clustering algorithm is introduced to calculate the corresponding weighted fault risk scores for all times within a given time period, using these scores as clustering samples. Clustering processing is then performed, for example, using the k-means clustering algorithm, to obtain risk clusters representing different operating states. The logic for selecting the number of clusters is consistent with step S200 and will not be elaborated further. Subsequently, the highest density cluster is determined, representing the most concentrated, and therefore the most normal, risk score distribution within that time period. Finally, the maximum deviation between each data point within the highest density cluster and the cluster center is determined as the dynamic threshold for the current time period. This maximum deviation represents the maximum allowable fluctuation range of the risk score under the current normal operating conditions.

[0070] At this point, the dynamic threshold for the current moment has been determined.

[0071] Step S600: Compare the weighted fault risk score at the current moment with the dynamic threshold, and evaluate the fault status of the bus area based on the comparison result, so as to realize intelligent monitoring of the distribution cabinet.

[0072] It should be noted that the occurrence of a fault usually manifests as a sudden change in the risk state, that is, a significant jump in the weighted fault risk score within a short period of time. Therefore, by comparing the instantaneous increment of the risk score with a dynamic threshold, the moment of fault occurrence can be captured more sensitively and accurately.

[0073] Specifically, firstly, the weighted fault risk score at the current moment is calculated as the difference between the weighted fault risk score at the previous moment (i.e., after a 1-second interval) and the weighted fault risk score at the same sampling frequency as described above. Then, this difference is compared with the dynamic threshold determined in step S500 for the current moment. When the difference is greater than the dynamic threshold, it indicates a sudden increase in the risk score beyond the normal fluctuation range, at which point a fault is determined to have occurred in the bus area. Once a fault is determined, the system can immediately trigger an audible and visual alarm and automatically send a command to the control system to control the main circuit switch corresponding to the bus area to perform a disconnection operation, thereby achieving rapid fault isolation and effectively preventing the occurrence of safety accidents.

[0074] The second aspect of this embodiment provides an intelligent monitoring system for power distribution cabinets, such as... Figure 2 As shown, the intelligent monitoring system for power distribution cabinets includes a memory and a processor. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the first aspect of the present invention, an intelligent monitoring method for power distribution cabinets, is implemented.

[0075] The intelligent monitoring system for distribution cabinets also includes other components well known to those skilled in the art, such as communication buses and communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0076] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (DRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (DRAM), high-bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

[0077] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent monitoring of power distribution cabinets, characterized in that, Including the following steps: Acquire multi-dimensional signal data along multiple acquisition dimensions in the busbar area within the distribution cabinet; The feature vector at the current moment is obtained based on multidimensional signal data. Clustering is performed based on the distance between all feature vectors within the set time period to which the current moment belongs. The cluster center of the densest cluster is taken as the health vector at the current moment. The densest cluster is the cluster with the smallest intra-cluster dispersion. The intra-cluster dispersion is calculated based on the average distance between all feature vectors in each cluster and their corresponding cluster centers. Calculate the Mahalanobis distance between the feature vector and the health vector at the current moment to obtain the original fault risk score at the current moment; Based on the spectrum of multidimensional signal data of the bus region within a set time period, the amplitude stability and temporal continuity of the spectrum are obtained, and the signal data quality score corresponding to the current moment is determined accordingly; among them, the signal data quality score is positively correlated with both amplitude stability and temporal continuity; The signal data quality score and the original fault risk score are weighted and fused to obtain the weighted fault risk score at the current moment. Based on the statistical distribution characteristics of multiple weighted fault risk scores within a set time period to which the current time belongs, the dynamic threshold for the current time is determined, including: performing distance-based clustering on multiple weighted fault risk scores within the set time period to obtain the highest density cluster; and determining the maximum deviation between each data point in the highest density cluster and the cluster center as the dynamic threshold for the current time. The system compares the weighted fault risk score at the current moment with the dynamic threshold, and assesses the fault status of the busbar area based on the comparison results, so as to achieve intelligent monitoring of the distribution cabinet.

2. The intelligent monitoring method for power distribution cabinets according to claim 1, characterized in that, The original fault risk score at the current moment Satisfying the relation: ； in, It is the first The feature vector at time step; It is the first Health vector at any given time; It is a transpose operation; It is the first The inverse of the covariance matrix at time t; It is the standard normalization function.

3. The intelligent monitoring method for power distribution cabinets according to claim 1, characterized in that, The method of obtaining the amplitude stability and temporal continuity of the spectrum based on the spectrum of multidimensional signal data of the bus region within the set time period, and determining the signal data quality score corresponding to the current moment accordingly, includes: Perform Fourier transform on the multidimensional signal data within the set time period to obtain the spectrum of each dimension; Extract the amplitude and timestamp of multiple peaks from the spectrum of each dimension; The signal data quality score is calculated based on the standard deviation of the amplitudes of the multiple peaks and the maximum interval of the timestamps of the multiple peaks.

4. The intelligent monitoring method for power distribution cabinets according to claim 3, characterized in that, The signal data quality score Satisfying the relation: ； in, It is the first Within the set time period, the first The standard deviation of the peak amplitude in the spectrum of 3D signal data; Standard deviation Normalization Preset reference standard deviation for dimensional signal data; It is a preset micro value; It is the total number of dimensions of the signal data; It is the first Within the set time period, the first The maximum time interval in the timestamp of the peak in the spectrum of 3D signal data; It is the first Set the total duration of the time period.

5. The intelligent monitoring method for power distribution cabinets according to claim 1, characterized in that, The weighted fusion of the signal data quality score and the original fault risk score is performed by multiplying the original fault risk score by the signal data quality score.

6. The intelligent monitoring method for power distribution cabinets according to claim 1, characterized in that, The multiple acquisition dimensions include temperature, current, and acoustic signal dimensions.

7. The intelligent monitoring method for power distribution cabinets according to claim 1, characterized in that, The step of comparing the weighted fault risk score at the current moment with the dynamic threshold, and assessing the fault status of the bus region based on the comparison result, includes: Calculate the difference between the weighted fault risk score at the current moment and the weighted fault risk score at the previous moment; When the difference is greater than the dynamic threshold, a fault is determined to have occurred in the bus area, and an alarm is triggered or the switch of the bus area is disconnected.

8. A smart monitoring system for power distribution cabinets, characterized in that, The intelligent monitoring system for the distribution cabinet includes a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the intelligent monitoring method for the distribution cabinet according to any one of claims 1-7 is implemented.

Images