Method for positioning grounding fault of flexible power cable of power distribution system

Abstract

The invention discloses a power distribution system flexible power cable grounding fault positioning method, which comprises the following steps of 1, arranging current sensors on grounding wires of each phase branch of a flexible cable system, measuring time domain current instantaneous signals at the head end and the tail end of a mining flexible cable, and performing denoising processing on the acquired signals; 2, deducing a current time domain expression of the mining flexible cable grounding wire, and analyzing time domain characteristics of the mining flexible cable grounding wire according to the expression; 3, Fourier transform is carried out on the time domain signals obtained on the basis of the step 2, amplitudes AS (f) and AR (f) and phases phi S (f) and phi R (f) of harmonics of 0-3 kHz are extracted, and frequency domain characteristic parameters of the harmonics are analyzed. The method is based on joint optimization of artificial intelligence and machine learning, redundant characteristics are reduced, the model generalization ability is improved, the calculation speed and precision are greatly improved, and the method is suitable for large-scale popularization and application. The method is suitable for a complex environment of a coal mine construction site, a more accurate flexible alternating current cable fault line selection method is established, and the method has high engineering practical significance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power systems, in particular to a flexible power cable grounding fault positioning method of a power distribution system. BACKGROUND

[0002] In a flexible cable power supply system, grounding fault is the most common and high-hazard fault type. When single-phase or multi-point grounding occurs in a branch, it will cause system leakage current to increase, voltage imbalance and equipment damage. The traditional line selection method is mainly based on zero sequence current or voltage difference method. Due to the large cable distribution capacitance and complex power supply loop in the flexible power supply system, the line selection method often has the problems of misjudgment or insufficient sensitivity. Therefore, it is of great significance to study a new method that can directly use the grounding line current for high-precision line selection. SUMMARY

[0003] In order to make up for the shortcomings of the prior art, the present application provides a flexible power cable grounding fault positioning method of a power distribution system to solve the problems in the prior art.

[0004] In order to solve the above technical problems, the present application provides the following technical scheme:

[0005] A flexible power cable grounding fault positioning method of a power distribution system, comprising the following steps:

[0006] Step 1, arranging current sensors on the grounding lines of each phase branch of the flexible cable system to measure the time-domain current instantaneous signals of the first and last ends of the mine-used flexible cable 、 . The superscripts S and R represent the cable first-end quantity and the cable last-end quantity, respectively. The collected signals are subjected to denoising processing. An adaptive band-pass filter is used to extract the key frequency band characteristics in the range of 0-3 kHz.

[0007] Step 2, deducing the time-domain expression of the grounding line current of the mine-used flexible cable, as shown in the following formula.

[0008] ,

[0009] Among them, represents the grounding line current at the time of fault, represents the current time-domain signal flowing through the P-phase ( ) conducting core of the independent shielding six-core cable before the fault; and are the current time-domain signals on the P'-phase ( ) metal shielding layer at the first and last ends of the independent shielding six-core cable before the fault, respectively. and represent the mutual resistance and mutual inductance of the unit length conducting core of the cable, respectively.

[0010] Based on the obtained ground line current time domain signal, the sliding window method is used to calculate the instantaneous slope sequence of the current waveform to capture the rapid change trend of the current. A sliding window with a length of N is set, wherein the value of N is associated with the system power frequency period, and the number of sampling points in a quarter to one power frequency period is preferred to ensure that the dynamic process of the fault characteristics can be effectively captured. The window sliding step is set to a single sampling interval, and the sliding window is sequentially slid through the entire current sequence from the starting point of the current signal. For the current data point sequence in the kth window , a first-order linear regression is performed on the time index in the window to calculate the characteristic slope value of the window. The calculation formula is:

[0011] .

[0012] Step 3, based on the time domain signal preprocessed in step 1 and , fast Fourier transform (FFT) is performed to convert them to the frequency domain to obtain the frequency spectrum of the signal and . The harmonic components in the 0-3 kHz frequency band are analyzed. The amplitude AS(f), AR(f) and the phase ΦS(f), ΦR(f) at each characteristic frequency point f are extracted.

[0013] Step 4, based on the time domain and frequency domain features extracted in steps 2 and 3, an initial feature set is constructed. The time domain features include but are not limited to current instantaneous value, sliding window slope maximum value, mean value, variance, etc.; the frequency domain features include amplitude, phase, amplitude ratio, phase difference of key frequency points (such as power frequency, each harmonic). Subsequently, the recursive feature elimination method is used to select the highest feature subset from the initial feature set to distinguish between “fault line” and “non-fault line”, and to construct a reduced feature data set for model training. This step aims to reduce the data dimension, improve the training efficiency and generalization ability of the subsequent model.

[0014] Step 5, based on the above selected ground line current features. The obtained feature set is input into a multi-model fusion learner, including support vector machine (SVM) and random forest (RF) algorithm modules. The support vector machine is used for fault / non-fault binary classification; the random forest is used for line selection.

[0015] Step 6, based on the fault line determined in step 5, a three-layer BP neural network model is constructed with the fault distance as the output and the features in step 4 as the input. A large amount of simulation or historical fault data is used to train the network to learn the complex nonlinear mapping relationship between the fault point distance and the time-frequency features of the grounding line current. The features of the to-be-tested sample are input into the trained BP network, and the output is the preliminary fault distance estimate .

[0016] Step 7, in order to improve the positioning accuracy, the confidence weighted fusion method is used for the above n preliminary estimates to obtain the final fault distance. First, the confidence of the output fault distance is calculated, and the confidence is determined by the uncertainty, error and consistency of different prediction results, etc. Comprehensive factors are normalized to [0, 1]. Then the comprehensive confidence of each prediction is normalized as a weight and all fault distances are weighted and averaged to obtain the final fault point position estimate. The calculation formula is as follows:

[0017] ,

[0018] Wherein is the confidence, is the weight.

[0019] One or more technical solutions provided in the embodiments of the application have at least the following technical effects or advantages:

[0020] The application is based on artificial intelligence and machine learning joint optimization, reduces redundant features and improves model generalization ability, greatly improves the calculation speed and accuracy.

[0021] The application is based on the research on the complex coupling characteristics of the multi-layer conductor of the flexible alternating current cable, and an equivalent analysis model more suitable for single-phase grounding fault of the flexible alternating current cable is established, which lays a theoretical foundation for subsequent fault line selection.

[0022] The application is suitable for the complex environment of coal mine construction site, and establishes a more accurate flexible alternating current cable fault line selection method, which has high engineering practical significance. BRIEF DESCRIPTION OF DRAWINGS

[0023] Fig. 1 is the flowchart of the application;

[0024] Fig. 2 is a schematic diagram of a fault circuit distributed parameter model;

[0025] Fig. 3 is a schematic diagram of a single-phase grounding fault structure of a flexible cable. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0027] As shown in the figure, a power distribution system flexible power cable grounding fault positioning method comprises the following steps: Figs. 1-3

[0028] Step 1, arranging current sensors on the grounding lines of each phase branch of the flexible cable system, measuring the time-domain current transient signals of the first and last ends of the mine-used flexible cable The superscripts S and R represent the first end quantity and the last end quantity of the cable respectively. The collected signals are subjected to denoising processing. An adaptive band-pass filter is used to extract the key frequency band features in the range of 0-3 kHz.

[0029] Step 2, deducing the time-domain expression of the grounding line current of the mine-used flexible cable, as shown in the following formula.

[0030]

[0031] Among them represents the grounding line current when a fault occurs, represents the time-domain current signal flowing through the P-phase ( ) conducting core of the independent shielding six-core cable before a fault occurs; and are the time-domain current signals on the first end and the last end P'-phase ( ) metal shielding layer of the independent shielding six-core cable before a fault occurs. and represent the mutual resistance and mutual inductance of the unit length conducting core of the cable.

[0032] Based on the obtained grounding line current time-domain signal, the present application uses a sliding window method to calculate the instantaneous slope sequence of the current waveform to capture the rapid change trend of the current. A sliding window with a length of N is set, wherein the value of N is associated with the system power frequency period, and the number of sampling points in one quarter to one power frequency period is preferably selected to ensure that the dynamic process of the fault characteristics can be effectively captured. The window sliding step is set as a single sampling interval, and the sliding window is sequentially slid over the entire current sequence from the starting point of the current signal. For the current data point sequence in the kth window, a first-order linear regression is performed on the time index in the window to calculate the characteristic slope value of the window. The calculation formula is as follows:

[0033] ​​ ,

[0034] Step 3, based on the pre-processed time-domain signal in step 1 and Fast Fourier Transform (FFT) is performed to convert it to the frequency domain to obtain the frequency spectrum of the signal and . The harmonic components in the 0-3 kHz frequency band are analyzed. The amplitude AS(f), AR(f) and phase ΦS(f), ΦR(f) at each characteristic frequency point f are extracted.

[0035] Step 4, based on the time-domain and frequency-domain features extracted in steps 2 and 3, an initial feature set is constructed. The time-domain features include but are not limited to current instantaneous value, sliding window slope maximum value, mean value, variance, etc.; the frequency-domain features include amplitude, phase, amplitude ratio, phase difference of key frequency points (such as power frequency, each harmonic). Then, the recursive feature elimination method is used to select the feature subset with the highest discrimination between "fault line" and "non-fault line" from the initial feature set, to construct a reduced feature dataset for model training. This step aims to reduce the data dimension and improve the training efficiency and generalization ability of the subsequent model.

[0036] Step 5, based on the grounding line current features selected above. The obtained feature set is input into a multi-model fusion learner, including support vector machine (SVM) and random forest (RF) algorithm modules. Support vector machine is used for fault / non-fault binary classification; random forest is used for line selection.

[0037] Step 6, based on the fault line determined in step 5, a three-layer BP neural network model is constructed with fault distance as output and step 4 features as input. A large amount of simulation or historical fault data is used to train the network to learn the complex nonlinear mapping relationship between fault point distance and grounding line current time-frequency features. The features of the sample to be tested are input into the trained BP network, and the output is the preliminary fault distance estimate .

[0038] Step 7, to improve the positioning accuracy, the confidence weighted fusion method is used for the above n preliminary estimates to obtain the final fault distance. First, the confidence of the output fault distance is calculated, which is determined by the uncertainty, error and consistency of different prediction results, etc., and is normalized to [0, 1] through normalization. Then the comprehensive confidence of each prediction is normalized as a weight and all fault distances are weighted and averaged to obtain the final fault point position estimate. The calculation formula is as follows:

[0039] ,

[0040] where is a confidence, is a weight.

[0041] It is apparent to a person skilled in the art that the present application is not limited to the details of the above-described exemplary embodiments, but can be implemented in other concrete forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered in all respects as illustrative and not restrictive, the scope of the present application being defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and range of equivalency of the claims are embraced therein.

[0042] Furthermore, it should be understood that although the present specification describes exemplary embodiments, not every embodiment contains only one independent technical solution, and the present specification is described in this way only for the sake of clarity, and a person skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment have been appropriately combined to form other embodiments which are easily understood by a person skilled in the art.

Claims

1. A method for locating grounding faults in flexible power cables in a power distribution system, characterized in that, Includes the following steps: Step 1: Install current sensors on the grounding wires of each phase branch of the flexible cable system to measure the instantaneous time-domain current signal at the beginning and end of the mining flexible cable. , The acquired signals are then denoised. Step 2: Derive the time-domain expression for the grounding current of the flexible mining cable, and analyze its time-domain characteristics based on the expression; Step 3: Apply Fourier transform to the time-domain signal obtained in Step 2, extract the amplitude AS(f), AR(f) and phase ΦS(f), ΦR(f) of the harmonics in the 0-3kHz range, and analyze their frequency domain characteristic parameters. Step 4: Based on the time-domain and frequency-domain features from Step 2 and Step 3, construct an initial feature dataset and filter out the grounding current characteristics that can reflect the faulty and non-faulty lines. Step 5: Based on the grounding wire current characteristics obtained in Step 4, input the obtained feature set into the multi-model fusion learner to learn the characteristics of faulty lines and non-faulty lines. Step 6: Based on the route selection results in Step 5, use a BP neural network to train the relationship between the distance to the fault point and the above features.

2. The method for locating grounding faults in flexible power cables of a power distribution system according to claim 1, characterized in that: In step 1, current sensors are placed on the grounding wires of each phase branch of the flexible cable system to measure the instantaneous time-domain current signals at the beginning and end of the mining flexible cable. , The superscripts S and R represent the quantities at the beginning and end of the cable, respectively; the acquired signals are denoised; and an adaptive bandpass filter is used to extract key frequency band features in the range of 0–3 kHz.

3. The method for locating grounding faults in flexible power cables of a power distribution system according to claim 1, characterized in that: In step 2, the time-domain expression for the grounding current of the mining flexible cable is derived as shown in the following formula; ， in Indicates the grounding current during a fault. This indicates the P phase before a fault occurs in the independently shielded six-core cable. The time-domain signal of the current flowing through the conductor core; and These are the first and last P' phases of the independently shielded six-core cable before a fault occurred. The current time-domain signal on the metal shielding layer; and These represent the mutual resistance and mutual inductance per unit length of the conductor core in the cable, respectively. Based on the grounding wire current time-domain signal obtained above, this invention uses the sliding window method to calculate the instantaneous slope sequence of the current waveform to capture the rapid change trend of the current; a sliding window of length N is set, where the value of N is related to the system power frequency period; the window sliding step size is set to a single sampling interval, and starting from the starting point of the current signal, the sliding window is slid across the entire current sequence sequentially; for the current data point sequence within the k-th window... Compare it with the time index within the window. Perform a first-order linear regression and calculate the characteristic slope value within the window. The calculation formula is as follows: 。 4. The method for locating grounding faults in flexible power cables of a power distribution system according to claim 1, characterized in that: Based on the time-domain signal preprocessed in step 1 and Perform a Fast Fourier Transform to convert it to the frequency domain, and obtain the signal's spectrum. and The analysis focuses on harmonic components within the 0-3 kHz frequency band; the amplitudes AS(f), AR(f), and phases ΦS(f), ΦR(f) at each characteristic frequency point f are extracted.

5. The method for locating grounding faults in flexible power cables of a power distribution system according to claim 1, characterized in that: Based on the time-domain and frequency-domain features extracted in steps 2 and 3, an initial feature set is constructed. The time-domain features include the instantaneous current value, the maximum slope of the sliding window, the mean, and the variance. The frequency-domain features include the amplitude, phase, amplitude ratio, and phase difference of key frequency points. Subsequently, a recursive feature elimination method is used to select the feature subset with the highest distinguishability between "faulty lines" and "non-faulty lines" from the initial feature set, thus constructing a simplified feature dataset for model training.

6. The method for locating grounding faults in flexible power cables of a power distribution system according to claim 1, characterized in that: Based on the grounding wire current characteristics obtained from the above screening, the obtained feature set is input into a multi-model fusion learner, which includes two algorithm modules: support vector machine and random forest. Support vector machine is used for fault / non-fault binary classification; random forest is used for line selection judgment.

7. The method for locating grounding faults in flexible power cables of a power distribution system according to claim 1, characterized in that: In step 6, based on the faulty line determined in step 5, a three-layer BP neural network model is constructed, with the fault distance as the output and the features proposed in step 4 as the input. This network is trained using a large amount of simulation or historical fault data to learn the complex nonlinear mapping relationship between the fault distance and the time-frequency characteristics of the grounding wire current. The features of the sample to be tested are input into the trained BP network, and its output is the preliminary estimate of the fault distance. .

8. The method for locating grounding faults in flexible power cables of a power distribution system according to claim 1, characterized in that: In step 7, to improve positioning accuracy, a confidence-weighted fusion method is used to obtain the final fault distance from the above n preliminary estimates. First, the confidence level of the output fault distance is calculated. The confidence level is determined by the uncertainty, error, and consistency factors between different prediction results of the model prediction, and is normalized to [0,1]. Then, the comprehensive confidence level of each prediction is normalized into weights, and all fault distances are weighted and averaged to obtain the final fault location estimate. The calculation formula is as follows: ， in It's the confidence level. It's the weight.

Images