An intelligent identification method for abnormal state of a coal mine power supply system

Abstract

The present application relates to the technical field of coal mine power supply system anomaly identification, and particularly relates to a coal mine power supply system abnormal state intelligent identification method. The method comprises: collecting zero sequence current and zero sequence voltage of each feeder branch for a preset length of time to form a zero sequence current sequence and a zero sequence voltage sequence; taking one zero sequence current sequence as a sample and obtaining a frequency domain amplitude sequence of the sample; obtaining a spectral fourth moment deviation factor and a phase space trajectory singularity factor between two samples; then correcting the original Euclidean distance between the two samples to obtain a corrected Euclidean distance between the two samples; clustering according to the corrected Euclidean distance between each two samples to obtain different class clusters; obtaining a fault class cluster according to the representative spectral fourth moment ratio and the representative phase space correlation coefficient square of each class cluster; and identifying whether the state of the coal mine power supply system is abnormal according to the fault class cluster. The present application improves the accuracy of coal mine power supply system anomaly identification.

Description

Technical Field

[0001] This invention relates to the field of anomaly identification technology in coal mine power supply systems, and specifically to an intelligent identification method for abnormal states in coal mine power supply systems. Background Technology

[0002] As the core power source for mine production, the underground power supply system directly impacts the safety of coal mine operations. Modern coal mine power supply networks often employ neutral-point ungrounded or arc-suppression coil-grounded operation. Furthermore, with advancements in coal mining technology, high-power frequency converters are widely used underground to drive heavy equipment such as scraper conveyors and coal mining machines. The system typically collects zero-sequence current data from each branch in real time and analyzes this data using intelligent algorithms deployed on monitoring terminals to identify normal operation, single-phase grounding faults, and various interference conditions.

[0003] In existing data processing solutions, density-based clustering algorithms (DBSCAN) are widely used in unsupervised learning scenarios. These methods typically extract the time-domain amplitude or frequency-domain energy of the zero-sequence current signal as a feature vector and use Euclidean distance as a measure of similarity between samples. The basic logic of the algorithm is to calculate the geometric distance between the sample to be tested and known fault samples in the feature space. If the distance is less than a preset threshold and falls into a high-density region, they are classified as belonging to the same type of fault. This method performs well when processing steady-state signals generated by linear loads, effectively separating fault signals with significant amplitude differences from background noise.

[0004] However, in the complex working conditions of actual underground coal mines, the clustering analysis method using traditional Euclidean distance has shortcomings. Specifically, high-frequency harmonic interference from frequency converters and random electromagnetic noise exhibit severe aliasing with high-resistance grounding fault signals in the Euclidean feature space, preventing the algorithm from correctly distinguishing signal properties based on numerical distance. Specifically, the arc signal generated by a high-resistance grounding fault in an underground coal mine is essentially a broadband random process driven by system voltage, with its energy dispersed across the entire high-frequency band. Meanwhile, the specific harmonic interference generated by a high-power frequency converter during operation has its energy highly concentrated at discrete frequency points. However, after coupling and filtering by the cable's distributed capacitance, the total energy amplitude (i.e., the modulus of the eigenvector) presented in the zero-sequence loop is often very close to that of the high-resistance fault signal. Simultaneously, although the random electromagnetic noise in the underground environment also exhibits broadband characteristics in the spectrum, its generation mechanism is independent of system voltage and lacks the voltage-current causal correlation unique to fault signals. Since Euclidean distance only calculates the difference in the magnitude of eigenvectors, it is completely insensitive to the morphological structure of spectral energy distribution and the correlation between multimodal data. This causes the DBSCAN algorithm to misjudge discrete harmonic interference from frequency converters with similar total energy as broadband fault signals, or to miscluster random noise unrelated to voltage into real fault clusters driven by voltage, thus triggering serious false alarms or missed alarms. This cannot meet the actual needs of coal mine power supply systems for accurate identification of abnormal states. Summary of the Invention

[0005] To address the aforementioned technical problems, the present invention aims to provide an intelligent identification method for abnormal states in coal mine power supply systems. The specific technical solution adopted is as follows:

[0006] One embodiment of the present invention provides an intelligent identification method for abnormal states in a coal mine power supply system, the method comprising:

[0007] Zero-sequence current and zero-sequence voltage of each feeder branch of the coal mine power supply system are collected for a preset time length and formed into zero-sequence current sequence and zero-sequence voltage sequence, respectively; a zero-sequence current sequence is taken as a sample and the frequency domain amplitude sequence of the sample is obtained.

[0008] The spectral fourth-order moment deviation factor between the two samples is obtained from the frequency domain amplitudes of each frequency domain amplitude sequence; the phase space trajectory singularity factor between the two samples is obtained using the zero-sequence current sequence and zero-sequence voltage sequence corresponding to the two samples.

[0009] The original Euclidean distance between two samples is corrected by the spectral fourth-order moment deviation factor and the phase space trajectory singularity factor to obtain the corrected Euclidean distance between the two samples; clustering is then performed based on the corrected Euclidean distance between each pair of samples to obtain different clusters;

[0010] Fault clusters are obtained based on the representative spectral fourth-order moment ratio and the square of the representative phase space correlation coefficient of each cluster; the state of the coal mine power supply system is identified as abnormal based on the fault clusters.

[0011] Preferably, the calculation model for the fourth-order moment deviation factor of the spectrum is as follows:

[0012] ,

[0013] in, denoted as the fourth-order moment deviation factor of the spectrum between the i-th sample and the j-th sample; μ represents the weighting coefficient; tanh() represents the hyperbolic tangent function; ln() represents the logarithm with the natural constant e as the base; N represents the number of frequency domain amplitudes in the frequency domain amplitude sequence; This represents the k-th frequency domain amplitude in the frequency domain amplitude sequence of the i-th sample; This represents the k-th frequency domain amplitude in the frequency domain amplitude sequence of the j-th sample.

[0014] Preferably, the phase space trajectory singularity factor between the two samples is obtained using the zero-sequence current sequence and the zero-sequence voltage sequence corresponding to the two samples, including:

[0015] Multiply the zero-sequence current at a certain moment in the zero-sequence current sequence corresponding to a sample by the zero-sequence voltage at the same moment in the zero-sequence voltage sequence corresponding to the sample to obtain the voltage-current multiplication result at that moment; sum the voltage-current multiplication results at each moment in the zero-sequence current sequence corresponding to the sample and then square them to obtain the first coefficient of the sample; sum the squares of the zero-sequence current at each moment in the zero-sequence current sequence corresponding to the sample to obtain the second coefficient of the sample; sum the squares of the voltage-current at each moment in the zero-sequence voltage sequence corresponding to the sample to obtain the third coefficient of the sample; divide the first coefficient by the product of the second coefficient and the third coefficient to obtain the ratio result; subtract the ratio result from the first preset value and take the square root to obtain the square root result of the sample; map the product of the preset exponential coefficient and the absolute value of the difference between the square root result of the sample and the square root result of another sample using an exponential function with the natural constant as the base to obtain the phase space trajectory singularity factor between the two samples.

[0016] Preferably, before correcting the original Euclidean distance between the two samples based on the spectral fourth-order moment deviation factor and the phase space trajectory singularity factor, the method further includes:

[0017] The original Euclidean distance between two samples is obtained using the Euclidean distance calculation formula based on the zero-sequence current sequences corresponding to the two samples.

[0018] Preferably, the original Euclidean distance between the two samples is corrected based on the spectral fourth-order moment deviation factor and the phase space trajectory singularity factor, resulting in the corrected Euclidean distance between the two samples, including:

[0019] The corrected Euclidean distance between two samples is obtained by multiplying the original Euclidean distance between the two samples, the spectral fourth moment deviation factor, and the phase space trajectory singularity factor.

[0020] Preferably, fault clusters are obtained based on the representative spectral fourth-order moment ratio and the squared representative phase space correlation coefficient of each cluster, including:

[0021] If the ratio of the fourth moment of the representative spectrum of a cluster is less than the spectrum discrimination threshold, and the square of the representative phase space correlation coefficient of the cluster is greater than the correlation discrimination threshold, then the cluster is a fault cluster.

[0022] Preferably, identifying whether the coal mine power supply system is abnormal based on the fault cluster includes:

[0023] The cluster to which the latest sample belongs is determined by the corrected Euclidean distance between the latest sample collected in real time and the cluster centers of each type of cluster. If the latest sample belongs to the fault cluster, then the coal mine power supply system is in an abnormal state.

[0024] The embodiments of the present invention have at least the following beneficial effects: This application collects zero-sequence current and zero-sequence voltage of each feeder branch of the coal mine power supply system for a preset time length, forming zero-sequence current sequences and zero-sequence voltage sequences respectively; a zero-sequence current sequence is taken as a sample, and the frequency domain amplitude sequence of the sample is obtained; then, based on the frequency domain amplitudes of the two samples, the spectral fourth-order moment deviation factor between the two samples is obtained, constructing a differentiated feature reflecting spectral sparsity. This adaptive adjustment mechanism based on data features can solve the problem of low fault signal identification rate under strong inverter interference environment; next, the phase space trajectory singularity between the two samples is obtained using the zero-sequence current sequence and zero-sequence voltage sequence corresponding to the two samples. The algorithm employs several methods to improve the accuracy of identifying anomalies in coal mine power supply systems. First, it corrects the original Euclidean distance between two samples using a spectral fourth-order moment deviation factor and a phase space trajectory singularity factor. Then, it clusters samples based on the corrected Euclidean distance, making the clustering algorithm more adaptable to the application environment and improving clustering accuracy. Finally, it obtains fault clusters based on the representative spectral fourth-order moment ratio and the squared representative phase space correlation coefficient of each cluster, thereby identifying whether the coal mine power supply system is abnormal and improving the accuracy of identifying abnormal states in the coal mine power supply system. Attached Figure Description

[0025] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart of a method for intelligent identification of abnormal states in a coal mine power supply system provided in an embodiment of the present invention. Detailed Implementation

[0027] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an intelligent identification method for abnormal states in a coal mine power supply system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0029] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent identification method for abnormal states in coal mine power supply systems provided by this invention.

[0030] Example:

[0031] The main application scenario of this invention is to analyze the zero-sequence current of each branch collected in a coal mine using a clustering algorithm, and to adaptively improve the clustering algorithm by combining the current data characteristics and scenarios to identify abnormal states in the coal mine power supply system.

[0032] Please see Figure 1 The diagram illustrates a flowchart of an intelligent identification method for abnormal states in a coal mine power supply system according to an embodiment of the present invention. The method includes the following steps:

[0033] Step S1: Collect the zero-sequence current and zero-sequence voltage of each feeder branch of the coal mine power supply system for a preset time length, and form a zero-sequence current sequence and a zero-sequence voltage sequence respectively; take a zero-sequence current sequence as a sample, and obtain the frequency domain amplitude sequence of the sample.

[0034] First, high-precision instrument transformers and data acquisition cards need to be installed on each feeder branch of the coal mine power supply system to collect the zero-sequence current and zero-sequence voltage of each branch in real time. The sampling frequency is set to [missing information]. (1 𝑘𝐻𝑧) is used to meet the requirements for capturing high-frequency transient signals. For subsequent algorithm calculations, a time window of preset length T is selected as the sample sequence obtained through one sampling. For the i-th sampling, the zero-sequence currents at each moment within the preset time length of a sampled branch are combined to form a zero-sequence current sequence, denoted as . , This can represent the zero-sequence current at time t in the i-th zero-sequence current sequence. The zero-sequence voltages at various times within a preset time length of this branch, which are collected synchronously, form a zero-sequence voltage sequence, denoted as . Let represent the zero-sequence voltage at time t in the i-th zero-sequence current sequence, where t = 1, 2, ..., T; for ease of subsequent analysis, a zero-sequence current sequence is analyzed as a single sample. It can represent the zero-sequence current at time t in the zero-sequence current sequence corresponding to the i-th sample.

[0035] After acquiring the raw data, the collected zero-sequence current and zero-sequence voltage need to undergo noise reduction preprocessing to filter out ultra-high frequency random spikes other than power frequency interference, and the data needs to be normalized to eliminate the influence of system fluctuations on the amplitude reference. Subsequently, the zero-sequence current sequence corresponding to the sample is... Perform a Discrete Fourier Transform to obtain the frequency domain amplitude sequence of the sample, denoted as... , represents the k-th frequency domain amplitude in the frequency domain amplitude sequence of the i-th sample, where , This represents the total number of frequency domain sampling points. At this point, all the necessary raw data and basic features for subsequent optimization factor calculations have been prepared.

[0036] Step S2: Obtain the spectral fourth moment deviation factor between the two samples based on the frequency domain amplitudes of each sample in the frequency domain amplitude sequence; obtain the phase space trajectory singularity factor between the two samples using the zero-sequence current sequence and zero-sequence voltage sequence corresponding to the two samples.

[0037] In the actual operation of underground power supply systems in coal mines, inverter interference signals and high-resistance ground fault signals are two of the most easily confused abnormal states. When an inverter performs speed control, its inverter unit generates specific high-order harmonics. These harmonics appear as highly concentrated discrete peaks in the frequency domain, meaning they have extremely large amplitudes at a few harmonic frequencies, while their amplitudes are smaller in other frequency bands. In contrast, the arc signal generated by a high-resistance ground fault is broadband random noise, with its energy exhibiting a relatively flat and uniform distribution across the entire high-frequency band, without prominent isolated peaks. However, traditional distance measurement methods typically only focus on the total energy of the signal in the frequency domain or the absolute difference in amplitude at each frequency point. Under certain operating conditions, the total energy of strong inverter harmonic interference, after attenuation through cables, may be very close to the total energy of a weak high-resistance ground fault signal, resulting in a very small Euclidean distance between the two in the characteristic space. To solve this problem, it is necessary to introduce an index that can quantify the "concentration of frequency domain energy distribution." Because inverter signals exhibit significant spike characteristics, performing high-order power operations (e.g., fourth power) on the frequency domain amplitude will drastically amplify those large-amplitude spikes, causing them to dominate the total energy distribution. Conversely, for flat-distributed fault signals, the lack of prominent large-value points means that the energy distribution remains relatively uniform after the same high-order operations. Utilizing this nonlinear numerical amplification characteristic, differentiated features reflecting spectral sparsity can be constructed, thus distinguishing the two types of signals.

[0038] Based on the above analysis, the spectral fourth-order moment deviation factor between the two samples is obtained from the frequency domain amplitude values ​​in the frequency domain amplitude sequences of the two samples. The specific calculation model for the spectral fourth-order moment deviation factor is as follows:

[0039] ,

[0040] in, denoted as the fourth-order moment deviation factor of the spectrum between the i-th sample and the j-th sample; μ represents the weighting coefficient; tanh() represents the hyperbolic tangent function; ln() represents the logarithm with the natural constant e as the base; N represents the number of frequency domain amplitudes in the frequency domain amplitude sequence; This represents the k-th frequency domain amplitude in the frequency domain amplitude sequence of the i-th sample; This represents the k-th frequency domain amplitude in the frequency domain amplitude sequence of the j-th sample.

[0041] Spectral fourth moment deviation factor The design aims to address the difficulty in distinguishing between inverter interference and high-impedance faults when their total energy is similar by capturing the microscopic morphological differences in spectral energy distribution. The calculation model employs the ratio of the fourth-order spectral moment to the square of the second-order spectral moment. In real-world scenarios, the spectrum of inverter interference signals is dominated by a few high-amplitude harmonic spikes. When these spike amplitudes are raised to the fourth power, their values ​​increase exponentially, resulting in a very large numerator and a ratio significantly greater than that of a flatly distributed signal. Conversely, the spectrum of a high-impedance ground fault signal approximates white noise, with uniform and relatively small amplitudes at each frequency point. After the fourth power calculation, the increase in the numerator is limited, and the ratio remains at a low level.

[0042] To quantify the difference between the i-th and j-th samples in this feature, the formula employs logarithmic differencing. The introduction of the logarithmic function ln() not only smooths the dynamic range of the values ​​but, more importantly, transforms the difference in ratios into a distance on a linear scale, making the algorithm more sensitive to changes in the spectral morphology. When the i-th sample represents inverter interference (larger ratio) and the j-th sample represents a fault signal (smaller ratio), the absolute value of the logarithmic difference between the two will be very significant, indicating that although their total energy is similar, their frequency domain structures are drastically different.

[0043] To translate this structural difference into an effective correction for the clustering distance, the formula uses the hyperbolic tangent function tanh and weighting coefficient μ to map the difference value. This design ensures that when the spectral morphology of the two signals differs significantly, the fourth-order moment deviation factor will output a significant correction value greater than 1, thereby widening the distance between them in subsequent calculations and forcing the clustering algorithm to remove inverter interference from the fault cluster. Conversely, when both are fault signals or interference signals, due to their similar spectral morphology and minimal ratio difference, the fourth-order moment deviation factor will approach 1, thus maintaining the original Euclidean distance and ensuring that similar samples can be correctly clustered. Through this adaptive adjustment mechanism based on data features, the problem of low fault signal identification rate in environments with strong inverter interference is successfully solved.

[0044] Based on eliminating discrete harmonic interference from the frequency converter, a phase space trajectory singularity factor is constructed to further address the issue that random broadband noise and the frequency domain characteristics of real faults are similar.

[0045] After the initial correction using the fourth-order moment deviation factor, the algorithm can effectively identify and eliminate inverter harmonic interference with discrete spectral characteristics. However, in the complex electromagnetic environment of underground coal mines, in addition to inverter interference, there is also widespread random broadband noise generated by the start-up and shutdown of large equipment and the engagement of contactors. This type of noise manifests as a continuous spectrum with uniform energy distribution in the frequency domain, and its "spectral flatness" characteristics are very similar to the arc signal generated by a high-resistance grounding fault. Therefore, it is difficult to completely distinguish it based solely on frequency domain morphological analysis. Without further processing, this random noise is easily misjudged as a fault signal, leading to a persistently high false alarm rate in the system.

[0046] To address this lingering problem, a breakthrough must be sought in the physical mechanism of signal generation. Real high-resistance grounding faults are caused by system insulation failure, and the resulting zero-sequence current is directly driven by the system's zero-sequence voltage; a clear physical causal relationship and strong linear correlation exist between the two. Random broadband noise, on the other hand, is caused by external electromagnetic field induction or stray line parameters; its fluctuations are completely independent of the system's zero-sequence voltage and are uncorrelated. This difference in "correlation" can be mathematically mapped to the geometric trajectory characteristics of signal vectors in phase space: if the two are strongly correlated, their combined trajectory will tend towards a flat straight line or ellipse, enclosing a very small "geometric area"; if they are uncorrelated, their trajectory will exhibit a chaotic distribution, enclosing a larger "geometric area." Based on this physical fact, a metric factor reflecting the difference in correlation between signals can be constructed, thereby achieving precise removal of random noise.

[0047] The phase space trajectory singularity factor between two samples is obtained by using the zero-sequence current sequence and zero-sequence voltage sequence corresponding to the two samples.

[0048] Specifically, the zero-sequence current at a given moment in the zero-sequence current sequence corresponding to a sample is multiplied by the zero-sequence voltage at the same moment in the zero-sequence voltage sequence corresponding to the sample to obtain the voltage-current multiplication result at that moment; the summation of the voltage-current multiplication results at each moment in the zero-sequence current sequence corresponding to the sample is then squared to obtain the first coefficient of the sample; the summation of the squares of the zero-sequence current at each moment in the zero-sequence current sequence corresponding to the sample is obtained to obtain the second coefficient of the sample; the summation of the squares of the voltage-current at each moment in the zero-sequence voltage sequence corresponding to the sample is obtained to obtain the third coefficient of the sample; the first coefficient is divided by the product of the second and third coefficients to obtain the ratio result; the square root result of the sample is obtained by subtracting the ratio result from the first preset value; the phase space trajectory singularity factor between the two samples is obtained by mapping the product of the preset exponential coefficient and the absolute value of the difference between the square root result of the sample and the square root result of another sample using an exponential function with the natural constant as the base.

[0049] The specific calculation model for the singularity factor of the phase space trajectory between two samples is as follows:

[0050] ,

[0051] in, λ represents the phase space trajectory singularity factor between the i-th sample and the j-th sample; λ is a preset exponential coefficient used to adjust the penalty strength for correlation differences, and its preferred value in this embodiment is 10.0; since the phase space trajectory singularity factor aims to strongly eliminate unrelated random noise, a steep penalty boundary needs to be constructed. When λ=10.0, even if the correlation feature value has a difference of only 0.3, the exponential function exp(10×0.3) can produce a distance amplification effect of about 20 times, ensuring that the noise sample is completely pushed away from the fault cluster and achieving physical isolation effect; exp represents the exponential function with the natural constant as the base; T represents the number of moments in the zero-sequence current sequence or zero-sequence voltage sequence; Let be the sampled value of the zero-sequence current of the i-th sample at time t. The sampled value of the zero-sequence voltage of the i-th sample at time t. Let be the sampled value of the zero-sequence current of the j-th sample at time t. The sampled value of the zero-sequence voltage of the j-th sample at time t; This represents the product of the voltage and current of i samples at time t; The result of multiplying the voltage and current of the i-th sample at time t is... The first coefficient of the i-th sample. The second coefficient of the i-th sample. The third coefficient is for the i-th sample; the first preset value is 1.

[0052] The phase space trajectory singularity factor uses zero-sequence voltage as a reference and measures the degree of physical coupling between zero-sequence current and voltage to address the difficulty in distinguishing between random noise with similar frequency domain characteristics and real faults. The phase space trajectory singularity factor calculates the sine of the angle between the current vector and the voltage vector in multidimensional space. For voltage-driven real fault signals, the two are highly correlated, with a very small angle, and the sine value approaches 0; while for voltage-independent random noise, the two are orthogonal or uncorrelated, with an angle close to 90 degrees, and the sine value approaches 1.

[0053] To amplify this difference and transform it into a distance metric, the formula calculates the absolute value of the difference between the i-th and j-th samples on this geometric feature. When the i-th sample is random noise (eigenvalue close to 1) and the j-th sample is a real fault (eigenvalue close to 0), the difference between the two will be close to 1. This significant numerical difference directly reflects the fundamental difference in their physical generation mechanisms.

[0054] Finally, the formula introduces an exponential function and a pre-defined exponential coefficient λ to perform a non-linear mapping of this difference. This transforms a correlation difference close to 1 into a large distance penalty value (i.e., (Much greater than 1). This means that even if random noise is disguised as a fault signal in the spectrum, as long as it lacks a physical correlation with the voltage, The factor will forcibly increase the distance between itself and the faulty sample, making it an isolated point far from the fault cluster in the clustering space. Conversely, if both samples are real fault signals, since they are strongly correlated, the difference will be close to 0. The value will approach 1, thus not affecting the normal clustering of similar faults. This design completely solves the problem of accurate fault identification under random broadband noise interference, ensuring the physical reliability of the identification results.

[0055] Therefore, the spectral fourth-order moment deviation factor and phase space trajectory singularity factor between each pair of samples can be obtained through the above analysis.

[0056] Step S3: Correct the original Euclidean distance between two samples by adjusting the spectral fourth moment deviation factor and the phase space trajectory singularity factor between the two samples to obtain the corrected Euclidean distance between the two samples; perform clustering based on the corrected Euclidean distance between each pair of samples to obtain different clusters.

[0057] In this embodiment, two optimization factors (spectral fourth-moment deviation factor and phase space trajectory singularity factor) are introduced into the Euclidean distance calculation formula using a multiplicative weighted approach. Specifically, the original Euclidean distance is multiplied by the two optimization factors to obtain the final corrected distance. This ensures that any feature difference in any dimension (whether spectral morphology or physical correlation) can have a decisive impact on the final distance. That is, as long as there are significant spectral differences (large spectral fourth-moment deviation factor) or significant correlation differences (large phase space trajectory singularity factor) between samples, the final distance will be... This will be amplified dramatically, causing the two samples to be determined as non-adjacent in the clustering space and unable to be assigned to the same cluster.

[0058] Therefore, it is necessary to calculate the original Euclidean distance between the two samples. Specifically, the original Euclidean distance between the two samples is obtained by using the Euclidean distance calculation formula based on the zero-sequence current sequences corresponding to the two samples.

[0059] Furthermore, the original Euclidean distance between the two samples is corrected by the spectral fourth moment deviation factor and the phase space trajectory singularity factor, resulting in the corrected Euclidean distance between the two samples.

[0060] Specifically, the corrected Euclidean distance between two samples is obtained by multiplying the original Euclidean distance between the two samples, the spectral fourth moment deviation factor, and the phase space trajectory singularity factor.

[0061] Therefore, clustering can be performed subsequently based on the corrected Euclidean distance between every two samples.

[0062] Before performing clustering, to adapt to the corrected distance space distribution, the existing K-distance graph method is used to adaptively determine two key parameters of the DBSCAN algorithm: the neighborhood radius ϵ and the minimum number of contained points MinPts. First, MinPts is set to a fixed integer (MinPts ≥ 4) based on the dimension of the feature space. Then, the distance from each sample point in the dataset to its MinPts nearest neighbor is calculated; note that the distance calculation here must use the corrected Euclidean distance. All calculated k-distance values ​​are sorted in descending order, and a distance curve is plotted. The inflection point with the most abrupt change in slope is found in the curve; the distance value corresponding to this inflection point is selected as the optimal neighborhood radius ϵ. This method can objectively delineate the boundary between the core cluster and noise based on the actual density of the feature space after stretching by the optimization factor.

[0063] Subsequently, using the determined parameters ϵ, MinPts, and the corrected Euclidean distance instead of the original Euclidean distance, the standard procedure of the DBSCAN algorithm is run: all samples are traversed, and the number of their neighbors in the ϵ-neighborhood under the corrected Euclidean distance is calculated. If the number of neighbors is greater than MinPts, it is marked as a core point and the cluster is expanded; if a sample cannot be contained in any cluster, it is marked as a noise point. In this way, the algorithm can automatically cluster samples with similar amplitude, similar spectral shape, and similar voltage correlation into one class, while eliminating inverter interference and random noise.

[0064] Specifically, after the DBSCAN algorithm completes clustering, it will obtain several clusters composed of samples and a set of noise points marked as outliers.

[0065] Step S4: Obtain fault clusters based on the representative spectral fourth-order moment ratio and the square of the representative phase space correlation coefficient of each type of cluster; identify whether the coal mine power supply system is abnormal based on the fault clusters.

[0066] The above method uses the modified Euclidean distance between every two samples to obtain different clusters. Further, to determine the physical meaning of each cluster, statistical analysis of the central features of each cluster is required. First, the feature mean of all samples in each cluster is calculated, and the representative spectral fourth-order moment ratio of each cluster is extracted. and the squared representative phase space correlation coefficient .

[0067] The central features of each cluster are then extracted and compared with preset physical criteria to obtain the faulty clusters.

[0068] Specifically, if the ratio of the fourth moment of the representative spectrum of a cluster is less than the spectrum discrimination threshold, and the square of the representative phase space correlation coefficient of the cluster is greater than the correlation discrimination threshold, then the cluster is a fault cluster.

[0069] Among them, the reference value of the spectrum discrimination threshold is 0.2. The representative spectrum fourth-order moment ratio is used to determine whether the spectrum is flat and whether it conforms to broadband characteristics. Fault clusters usually show a broadband flat spectrum, and their representative spectrum fourth-order moment ratio is small. If the cluster is a cluster affected by frequency converter interference, it shows a discrete peak spectrum, and its representative spectrum fourth-order moment ratio is large.

[0070] The reference value for the relevant discrimination threshold is 0.6, which is used to determine whether it is driven by voltage. Fault clusters usually show a strong correlation with voltage, which conforms to causal characteristics. Their representative phase space correlation coefficient square is large. If the representative phase space correlation coefficient square is small, it may be random noise interference from the frequency converter, which shows no correlation or weak correlation.

[0071] If a certain type of cluster Smaller (indicating a flat spectrum, consistent with broadband characteristics) and If the value is close to 1 (indicating a strong correlation with voltage and conforming to causal characteristics), then this cluster is determined to be a high-resistance grounding fault cluster.

[0072] If a certain type of cluster If the value is large (indicating a significant spike), the cluster is identified as an inverter harmonic interference cluster regardless of its correlation. If the real-time collected samples are classified into this cluster, the system only records the inverter operation event and does not trip the circuit breaker.

[0073] If a certain type of cluster Smaller but If the value is close to 0 (indicating no correlation with voltage), or if the sample is directly marked as a noise point, it is determined to be random electromagnetic interference and filtered out by the system.

[0074] Furthermore, the status of the coal mine power supply system is identified as abnormal based on the fault clusters. Specifically, the cluster to which the latest sample belongs is determined based on the corrected Euclidean distance between the latest sample collected in real time and the cluster centers of each cluster. If the latest sample belongs to a fault cluster, then the coal mine power supply system is abnormal. In this case, the abnormality of the power supply system is a single-phase ground fault, and a fault selection command and alarm signal are output simultaneously.

[0075] Through the aforementioned feature-based inversion discrimination logic, this application achieves automatic mapping from unsupervised clustering results to specific fault types, thus completing the accurate identification of abnormal states in coal mine power supply systems.

[0076] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0077] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0078] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent identification of abnormal states in a coal mine power supply system, characterized in that, The method includes: Zero-sequence current and zero-sequence voltage of each feeder branch of the coal mine power supply system are collected for a preset time length and formed into zero-sequence current sequence and zero-sequence voltage sequence, respectively; a zero-sequence current sequence is taken as a sample and the frequency domain amplitude sequence of the sample is obtained. The spectral fourth-order moment deviation factor between the two samples is obtained from the frequency domain amplitudes of each frequency domain amplitude sequence; the phase space trajectory singularity factor between the two samples is obtained using the zero-sequence current sequence and zero-sequence voltage sequence corresponding to the two samples. The original Euclidean distance between two samples is corrected by the spectral fourth-order moment deviation factor and the phase space trajectory singularity factor to obtain the corrected Euclidean distance between the two samples; clustering is then performed based on the corrected Euclidean distance between each pair of samples to obtain different clusters; Fault clusters are obtained based on the representative spectral fourth-order moment ratio and the square of the representative phase space correlation coefficient of each cluster; the state of the coal mine power supply system is identified as abnormal based on the fault clusters.

2. The intelligent identification method for abnormal states of a coal mine power supply system according to claim 1, characterized in that, The specific calculation model for the fourth-order moment deviation factor of the spectrum is as follows: ， in, denoted as the fourth-order moment deviation factor of the spectrum between the i-th sample and the j-th sample; μ represents the weighting coefficient; tanh() represents the hyperbolic tangent function; ln() represents the logarithm with the natural constant e as the base; N represents the number of frequency domain amplitudes in the frequency domain amplitude sequence; This represents the k-th frequency domain amplitude in the frequency domain amplitude sequence of the i-th sample; This represents the k-th frequency domain amplitude in the frequency domain amplitude sequence of the j-th sample.

3. The intelligent identification method for abnormal states of a coal mine power supply system according to claim 1, characterized in that, The method of obtaining the phase space trajectory singularity factor between two samples using the zero-sequence current sequence and zero-sequence voltage sequence corresponding to the two samples includes: Multiply the zero-sequence current at a certain moment in the zero-sequence current sequence corresponding to a sample by the zero-sequence voltage at the same moment in the zero-sequence voltage sequence corresponding to the sample to obtain the voltage-current multiplication result at that moment; sum the voltage-current multiplication results at each moment in the zero-sequence current sequence corresponding to the sample and then square them to obtain the first coefficient of the sample; sum the squares of the zero-sequence current at each moment in the zero-sequence current sequence corresponding to the sample to obtain the second coefficient of the sample; sum the squares of the voltage-current at each moment in the zero-sequence voltage sequence corresponding to the sample to obtain the third coefficient of the sample; divide the first coefficient by the product of the second coefficient and the third coefficient to obtain the ratio result; subtract the ratio result from the first preset value and take the square root to obtain the square root result of the sample; map the product of the preset exponential coefficient and the absolute value of the difference between the square root result of the sample and the square root result of another sample using an exponential function with the natural constant as the base to obtain the phase space trajectory singularity factor between the two samples.

4. The intelligent identification method for abnormal states of a coal mine power supply system according to claim 1, characterized in that, Before the step of correcting the original Euclidean distance between two samples by adjusting the spectral fourth-order moment deviation factor and the phase space trajectory singularity factor between the two samples to obtain the corrected Euclidean distance between the two samples, the following steps are included: The original Euclidean distance between two samples is obtained using the Euclidean distance calculation formula based on the zero-sequence current sequences corresponding to the two samples.

5. The intelligent identification method for abnormal states of a coal mine power supply system according to claim 1, characterized in that, The process of correcting the original Euclidean distance between two samples based on the spectral fourth-order moment deviation factor and the phase space trajectory singularity factor to obtain the corrected Euclidean distance between the two samples includes: The corrected Euclidean distance between two samples is obtained by multiplying the original Euclidean distance between the two samples, the spectral fourth moment deviation factor, and the phase space trajectory singularity factor.

6. The intelligent identification method for abnormal states of a coal mine power supply system according to claim 1, characterized in that, The method of obtaining fault clusters based on the representative spectral fourth-order moment ratio and the squared representative phase space correlation coefficient of each cluster includes: If the ratio of the fourth moment of the representative spectrum of a cluster is less than the spectrum discrimination threshold, and the square of the representative phase space correlation coefficient of the cluster is greater than the correlation discrimination threshold, then the cluster is a fault cluster.

7. The intelligent identification method for abnormal states of a coal mine power supply system according to claim 1, characterized in that, The step of identifying whether the coal mine power supply system is abnormal based on the fault cluster includes: The cluster to which the latest sample belongs is determined by the corrected Euclidean distance between the latest sample collected in real time and the cluster centers of each type of cluster. If the latest sample belongs to the fault cluster, then the coal mine power supply system is in an abnormal state.

Images