Fault line selection algorithm and system for multi-power electronic source power supply system, and storage medium

Abstract

The invention provides a fault line selection algorithm and system for a multi-power electronic source power supply system and a storage medium. The fault line selection algorithm for the multi-power electronic source power supply system comprises the steps of 1, establishing a fault historical data set; 2, carrying out data preprocessing and feature extraction; 3, forming a fault data training sample, and carrying out model training through a random forest algorithm; 4, acquiring data when an actual fault occurs, and performing feature extraction; and step 5, comparing measured data with the training model by using a fault line selection model established by using a random forest algorithm so as to determine a fault position. According to the fault line selection algorithm and system for the multi-power electronic source power supply system and the storage medium, model training is directly carried out by utilizing a simulation model or fault recording data, characteristic value extraction and analysis are carried out by utilizing actually measured data, rapid positioning can be realized, and the accuracy of fault line selection of the multi-power electronic source power supply system can be ensured.

Description

Technical Field

[0001] This invention relates to the field of electrical engineering and new energy power generation technology, and in particular to a fault selection algorithm, system and storage medium for a multi-power electronic source power supply system. Background Technology

[0002] Photovoltaic power plants, as a common renewable energy generation system, are clean and sustainable, playing a vital role in energy transition and carbon emission reduction. As the scale of photovoltaic power plant systems continues to expand, they form multi-power electronic source power supply systems composed of photovoltaic inverters, transformers, busbars, and other equipment. Failures in these devices can lead to abnormal current, voltage fluctuations, and power losses, reducing the power generation capacity of the photovoltaic power plant and the power quality of the grid. Timely and accurate fault selection is crucial to ensuring the normal operation of the power plant.

[0003] Chinese patent application CN201510933364.X discloses a method for selecting single-phase fault lines in a low-current grounding system. This method uses wavelet decomposition to obtain the wavelet energy of the zero-sequence current of the feeder, and establishes a sample set based on the obtained wavelet energy. After normalization, a normalized sample set X* is obtained. A Spiking neural network is initially constructed using an SRM neuron model. The SpikeProp method is used to calculate the weights of each synapse between neurons in the input layer H, neurons in the hidden layer I, and neurons in the output layer J. After correcting the connection weights, the Spiking neural network for fault selection is obtained. Fault data is input into the Spiking neural network, and the line marked with Φ0 in the output layer is identified as the faulty line. This method is convenient to operate, can quickly and effectively determine whether a grounding fault has occurred in the power grid, and, after determining the grounding fault, identifies the faulty line, thus providing a basis for timely fault elimination.

[0004] Chinese patent application CN201811375953.0 discloses a method for selecting a fault line in a distribution network with a single-phase open circuit. The method includes analyzing the topology of the distribution network system, collecting the voltage and current at the outgoing line, calculating the positive sequence current change and comparing it with the threshold of the positive sequence current diagram, calculating the positive sequence voltage change and comparing it with the threshold of the positive sequence voltage change, calculating the line impedance and comparing it with the line impedance threshold, obtaining the fault path line, and finally selecting the line with the lowest level among the fault path lines as the line where the fault occurred.

[0005] Chinese patent application CN202211635719.3 discloses a fault location method considering a large number of distributed photovoltaic power generation systems connected to a distribution network. The method includes: determining the composition of mixed measurement data based on the system measurement vectors of the active distribution network; analyzing the correlation coefficient matrix using signal correlation theory to determine the reference measurement time of the mixed measurement system and constructing a delay error function; constructing a linear state estimation model of the active distribution network based on the real and imaginary parts of the three-phase node voltages; solving for the state variables of the active distribution network using an iterative weighted least squares method; and constructing a fault location model and determining the location of the fault point by introducing virtual nodes.

[0006] Existing fault location algorithms are mainly designed for low-voltage distribution networks in traditional AC power transmission and distribution systems. They are not well-suited for multi-power electronic power supply systems connected to distributed photovoltaic power stations and cannot quickly and accurately locate faults.

[0007] The existing technologies described above differ significantly from this invention. Existing related patents primarily concern low-voltage equipment insulation monitoring and current signal measurement based on insulation fault circuit current sensors, but do not address fault location in multi-power electronic source power supply systems, nor do they cover intelligent processing algorithms suitable for fault location. Existing fault location algorithms are mainly designed for low-voltage distribution networks in traditional AC transmission and distribution systems, and are not well-suited for multi-power electronic source power supply systems connected to distributed photovoltaic power plants, failing to achieve fast and accurate fault location. Therefore, we have invented a new fault location algorithm, system, and storage medium for multi-power electronic source power supply systems. Summary of the Invention

[0008] The purpose of this invention is to provide a fault selection algorithm, system, and storage medium for multi-power electronic power supply systems that is highly applicable and can quickly and accurately perform fault selection.

[0009] The objective of this invention can be achieved through the following technical measures: a fault selection algorithm for a multi-power electronic source power supply system, which includes:

[0010] Step 1: Establish a fault history dataset;

[0011] Step 2: Perform data preprocessing and feature extraction;

[0012] Step 3: Generate fault data training samples and train the model using the random forest algorithm;

[0013] Step 4: Obtain data at the actual time of the fault and extract its features;

[0014] Step 5: The fault location is determined by comparing the fault location model established using the random forest algorithm with the training model using the measured data.

[0015] The objective of this invention can also be achieved through the following technical measures:

[0016] In step 1, acquire historical fault datasets for single-phase ground faults, two-phase ground faults, two-phase phase-to-phase faults, and three-phase faults. These historical fault datasets are acquired from the simulation platform or collected from the fault recording device of the actual power plant.

[0017] In step 2, after obtaining the historical fault dataset, the data is preprocessed to filter out invalid data caused by random noise and interference.

[0018] In step 2, feature extraction consists of three parts: empirical mode decomposition, Hilbert spectral analysis, and eigenvalue calculation.

[0019] In step 2, feature extraction specifically includes:

[0020] (a) Empirical mode decomposition is performed on the zero-sequence current data to obtain the modes IMF1-IMF. n ;

[0021] (b) Select m modes, m≤n, and perform Hilbert spectral analysis on them to obtain the modal components;

[0022] (c) Calculate the mean, standard deviation, and energy of the data obtained in (b);

[0023] (d) Obtain features in 3×m×3 dimensions.

[0024] In step c, the formulas for calculating the mean (Mean), standard deviation (Std), and energy (Energy) are as follows:

[0025]

[0026] x i- Let N represent the i-th discrete data point, and N represent the number of discrete data points. denoted by , x(t) represents the average value, and x(t) represents the real-time data.

[0027] In step 3, the features extracted from the fault dataset are used as the sample set X, and the corresponding fault locations are used as the label set y. The training sets X_train and y_train, and the test sets X_test and y_test are selected using a stratified sampling method. X_train and y_train are input into the Random Forest (RF) algorithm model for training to obtain the trained fault line selection model. X_test and y_test are input into the trained fault line selection model to evaluate the accuracy of the model.

[0028] In step 4, real-time fault data of the power supply system is collected from the fault recording device of the actual power station and used as the data source for fault line selection.

[0029] In step 4, feature extraction specifically includes:

[0030] (4a) Empirical mode decomposition is performed on the zero-sequence current data in the real-time fault data of the power supply system collected from the fault recording device of the actual power station to obtain the mode IMF1-IMF. n ;

[0031] (4b) Select m modes, m≤n, and perform Hilbert spectral analysis on them to obtain the modal components.

[0032] (4c) Calculate the mean, standard deviation, and energy of the data obtained in (4b);

[0033] (4d) Obtain features in 3×m×3 dimensions.

[0034] In step 5, the feature extraction results of the actual fault occurrence data are compared with the training model structure to determine the fault location. That is, the real-time fault data after feature extraction is added as input to the training random forest algorithm (RF) model, and the fault location in step 4 is obtained based on the RF training model results.

[0035] The objective of this invention can also be achieved through the following technical measures: a fault location system for a multi-power electronic power supply system, wherein the fault location system for a multi-power electronic power supply system uses a fault location algorithm for a multi-power electronic power supply system to determine the location of the line fault.

[0036] The objective of this invention can also be achieved through the following technical measures: a computer-readable storage medium, the computer-readable storage medium including a stored computer program; wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute a fault selection algorithm for a multi-power electronic source power supply system.

[0037] The fault location algorithm, system, and storage medium for multi-power electronic source power supply systems proposed in this invention are based on fault data training samples, feature extraction, and random forest algorithm model training. The algorithm involves establishing a historical fault dataset, data preprocessing, and feature extraction to form fault data training samples, followed by model training using the random forest algorithm. After feature extraction using measured data from distributed photovoltaic power stations, the fault location model built using the random forest algorithm is compared with the trained model using the measured data to determine the fault location. This fault location algorithm for multi-power source power supply systems is not limited by the power supply system topology; it directly uses simulation models or fault waveform data for model training and utilizes measured data for feature value extraction and analysis, enabling rapid location. Simultaneously, the current measured data can be used as training data to improve the accuracy of the training model, thereby ensuring the accuracy of fault location in multi-power source power supply systems. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of a single-phase ground fault in a specific embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of a two-phase ground fault in a specific embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram of a two-phase short-circuit fault in a specific embodiment of the present invention;

[0041] Figure 4 This is a schematic diagram of a three-phase short-circuit fault in a specific embodiment of the present invention;

[0042] Figure 5 This is a schematic diagram of AC side fault types in a specific embodiment of the present invention;

[0043] Figure 6 This is a schematic diagram of AC side fault simulation in a specific embodiment of the present invention;

[0044] Figure 7 A flowchart of a specific embodiment of the fault selection algorithm for a multi-power electronic source power supply system of the present invention;

[0045] Figure 8 This is a graph showing the relationship between the number of decision trees and accuracy in a specific embodiment of the present invention. Detailed Implementation

[0046] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, 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.

[0047] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, and / or combinations thereof.

[0048] The output of the distributed photovoltaic output unit is three-phase AC power. The faults on the low-voltage AC side mainly include single-phase grounding short circuit faults, two-phase short circuit faults, three-phase grounding short circuit faults, and two-phase grounding faults.

[0049] (a) Single-phase ground fault

[0050] Single-phase faults are the most frequent type of fault, thus attracting much research and analysis. When an asymmetrical short-circuit fault occurs, the asymmetrical short-circuit fault analysis method is introduced. Using phase A as the reference phase, the fault diagram is as follows: Figure 1 As shown.

[0051] At this time, the sequence voltages at each point of the fault are:

[0052]

[0053] in, The voltage at fault point a before the short circuit occurred. and These represent the positive, negative, and zero-sequence components of the current and voltage at the fault location, respectively. ff(1) X ff(2) and X ff(0) These represent the input impedances of the sequence networks at this point. There are six unknown variables, and a complete solution requires considering the boundary conditions for asymmetric short circuits.

[0054] Boundary conditions at the location of phase A fault:

[0055]

[0056] in, This represents the voltage at the grounding point of phase A. and This indicates the fault current of the normal two phases B and C at the grounding point.

[0057] When expressed in order components, its boundary conditions can be transformed into the following equation:

[0058]

[0059] The sequence current at fault point a can be calculated:

[0060]

[0061] That is, the fault current at fault point a:

[0062]

[0063] It can be seen that when a single-phase short-circuit fault occurs, the short-circuit current at the fault point is related to the magnitude of the input impedance in the positive, negative and zero sequence networks, and its value is 3 times the value of each sequence current.

[0064] (b) Two-phase-to-ground short circuit fault

[0065] like Figure 2 As shown, the boundary conditions for the failure are:

[0066]

[0067] When expressed in order components, its boundary conditions can be transformed as shown below:

[0068]

[0069] By combining the above equations, we can obtain the sequence currents as shown below:

[0070]

[0071] The absolute values ​​of the short-circuit currents at fault points b and c are shown below:

[0072]

[0073] It can be seen that when a two-phase ground fault occurs in a line, the absolute values ​​of the currents in the two faulty phases are equal. The value of m is related to X. ff(0) / X ff(2) The ratio is related; when the ratio approaches 0, When the ratio approaches infinity, m = 1.5. Therefore, the range of values ​​for m is... The range of variation is not large.

[0074] (c) Two-phase short circuit fault

[0075] like Figure 3 As shown, the boundary conditions at the location of the fault are:

[0076]

[0077] When expressed in order components, its boundary conditions can be transformed into:

[0078]

[0079] Combining the above equation, the sequence currents can be obtained as shown below:

[0080]

[0081] The short-circuit current values ​​at fault points b and c are shown below:

[0082]

[0083] It can be seen that when a phase-to-phase short circuit occurs on a line, the currents in the two faulty phases are equal in magnitude and opposite in direction, and their absolute values ​​are... The normal phase positive sequence current value is times that of the phase positive sequence current.

[0084] (d) Three-phase short circuit fault

[0085] Three-phase faults have the lowest probability of occurrence among short-circuit faults, but when a three-phase short circuit occurs, it often causes a relatively serious fault, and therefore requires sufficient attention.

[0086] like Figure 4 As shown, before the short circuit occurs, the circuit is in a steady state, where the expressions for the current and voltage of phase A are:

[0087]

[0088] U a =U m sin(ωt+α)

[0089] In the formula:

[0090]

[0091] Where Um is the voltage amplitude, R is the line resistance of each phase, R′ is the short-circuit point resistance, L is the line reactance of each phase, L′ is the short-circuit point reactance, ω is the grid angular frequency, and α is the initial phase angle of voltage and current.

[0092] A three-phase short circuit occurs at point f, effectively dividing the circuit into two independent circuits. The right half, which was operating in steady state, abruptly becomes a short-circuited circuit without power supply, and the current in this loop decays exponentially from its steady-state value to zero. Meanwhile, the left half, connected to the power supply, experiences a decrease in current from (R+R′)+jw(L+L′) per phase impedance during steady-state operation to R+jwL, and its current value also reaches a new steady-state value as impedance changes. Since the left-side circuit remains symmetrical after the short circuit, we will analyze one phase, A, as an example.

[0093] The differential equation for phase A circuit can be written as follows:

[0094]

[0095] The above equation is a first-order linear non-homogeneous differential equation with constant coefficients. Its solution is the total current during a short circuit in phase A, consisting of a particular solution and a general solution. The particular solution of the equation corresponds to the steady-state short-circuit current, expressed as:

[0096]

[0097] In the formula, Let R+jwL be the magnitude of the impedance of each term in the short-circuit loop. This represents the amplitude of the steady-state short-circuit current. α is the impedance angle of each phase of the short-circuit loop; α is the initial phase angle of the power supply voltage, i.e., the phase angle at t=0, also known as the closing angle. Therefore, it can be seen that the steady-state short-circuit current remains constant during transient changes and is a periodic component.

[0098] The general solution expression of the equation is:

[0099]

[0100] In the formula: p is the characteristic root of the homogeneous equation; is the time constant of the short-circuit loop; C is the integration constant, the value of which is determined by the initial conditions and serves as the initial value of the current.

[0101] The total current expression for the short-circuit current in phase A is:

[0102]

[0103] Since the three-phase short-circuit current is symmetrical, we can obtain the short-circuit calculation expressions for the other two phases B and C of the three-phase line by simply replacing α with (α-120°) and (α+120°).

[0104] like Figure 5 The diagram shows the fault types on the AC side. Fault simulations were performed on the low-voltage AC side for single-phase-to-ground short-circuit faults, two-phase-to-ground short-circuit faults, two-phase-to-phase short-circuit faults, and three-phase-to-ground short-circuit faults. Their probabilities of occurrence in the power system are 70%, 10%, 15%, and 5%, respectively. Information on each fault was collected to provide data support for training the line selection algorithm model.

[0105] A voltage detection point is set at the inverter outlet, and the simulation duration is 0.5s. When 0≤t<0.2s, the system is in normal state; when t=0.2s, single-phase ground fault, two-phase ground fault, two-phase short circuit fault and three-phase ground fault are set with a duration of 0.3s respectively.

[0106] Figure 6 As a result of the operation, when a ground fault occurs in a certain phase, the voltage of that phase drops rapidly; when a short circuit occurs between two phases, the voltages of both phases drop simultaneously, but the drop is smaller than that of a ground fault.

[0107] For the aforementioned faults such as single-phase-to-ground short circuit, two-phase-to-ground short circuit, two-phase-to-phase short circuit, and three-phase short circuit, this invention establishes a historical fault dataset, performs data preprocessing and feature extraction to form fault data training samples, and trains the model using a random forest algorithm. After extracting features from measured data from distributed photovoltaic power stations, the fault location model established using the random forest algorithm is compared with the measured data to determine the fault location. The fault location flowchart of this patent is shown below. Figure 7 As shown.

[0108] The process for training the model is as follows:

[0109] (1) Obtain historical fault datasets such as single-phase ground fault, two-phase ground fault, two-phase phase-to-phase fault, and three-phase fault. The historical fault datasets (zero-sequence current data) can be obtained from the simulation platform or collected from the fault recording device of the actual power plant. For details of the fault characteristics, please refer to the fault current characteristic formulas for each type of fault.

[0110] (2) Data preprocessing. After obtaining the fault dataset, the data must first be preprocessed. Whether it is a simulation platform or a waveform recording device, data corruption may occur during the transmission of large amounts of data. Therefore, before feature extraction, random noise and invalid data caused by interference are filtered out.

[0111] (3) Feature extraction. Feature extraction mainly includes three parts: empirical mode decomposition, Hilbert spectral analysis, and eigenvalue calculation.

[0112] (A) Empirical Mode Decomposition:

[0113] Input: Original signal x(t)

[0114] Output: IMF components, residual components

[0115] a) Initialization of the empirical mode decomposition process for fault data characteristics: y(t) = x(t), i = 1

[0116] b) Analyze and process the decomposed signal y(t), find all local extreme points in the time domain data, and use cubic spline interpolation to delineate all minimum and maximum points into lines, constructing upper and lower envelopes with the same form as y(t).

[0117] c) Calculate the average of the upper and lower envelopes, and the mean sequence is represented as m. i (t), subtract m from the decomposed signal i (t) yields h i (t), that is:

[0118] h i (t)=y(t)-m i(t)

[0119] d) The IMF must meet two conditions: Condition 1 - In the entire waveform of the signal, the difference between the number of extreme points and the number of zero-crossing points is less than 1; Condition 2 - In any time period, the mean of the upper and lower envelopes constructed from local maxima and local minima respectively is 0. (The question then returns to the previous statement, which is unclear without further context.) i (t) Whether these two conditions are met, if they are met, then h i (t) is the i-th IMF component decomposed from the original signal, i.e., imf i (t)=h i (t); if it does not meet the requirement, then let y(t) = h i (t), repeat steps (2)-(3) k times until h is decomposed. ik (t) satisfies two conditions, at which point imf i (t)=h ik (t).

[0120] e) Separate the decomposed IMF components from the decomposed signal; the remaining part is the difference signal r. i (t):

[0121] r i (t)=y(t)-imf i (t)

[0122] f) Let y(t) = r i (t), repeat steps (2)-(5) until r(t) becomes a monotonic function and the iteration stops. The final signal decomposition result is obtained:

[0123]

[0124] (B) Hilbert spectral analysis:

[0125] The initial signal can be represented as:

[0126]

[0127] Applying the Hilbert transform to each IMF in the above equation, we obtain:

[0128]

[0129] Constructing an analytic signal:

[0130]

[0131] Phase function:

[0132]

[0133] The instantaneous frequency is obtained from the phase function:

[0134]

[0135] (C) Feature value extraction

[0136] Feature extraction is a crucial step in machine learning algorithms. As input to the model, features significantly impact its quality; good features improve accuracy and reduce generalization error, while poor features decrease accuracy and increase generalization error. Zero-sequence current is time-series data. Generally, time-series data cannot be directly input into a classifier model; it requires methods to extract features from the raw data, and then using the feature set as the dataset to train the model.

[0137] The overall process of feature extraction is as follows:

[0138] (a) Empirical mode decomposition is performed on the zero-sequence current data to obtain the modes IMF1-IMF. n ;

[0139] (b) Select m modes (m≤n) and perform Hilbert spectral analysis on them to obtain the modal components.

[0140] Amplitude, frequency, and phase data in the time domain (3*m dimensions in total);

[0141] (c) Calculate the mean, standard deviation, and energy of the data obtained in (b) using the following formulas:

[0142]

[0143] x i- Let N represent the i-th discrete data point, and N represent the number of discrete data points. denoted by , x(t) represents the average value, and x(t) represents the real-time data.

[0144] (d) Obtain features in 3×m×3 dimensions.

[0145] Since the number of modal components obtained by EMD decomposition is determined by the specific signal rather than preset, it is necessary to determine the value of m when extracting features based on the specific dataset. This section mainly focuses on how to select an appropriate number of IMFs.

[0146] (4) Training the RF model. The features extracted from the fault dataset are used as the sample set X, and the corresponding fault locations are used as the label set y. A stratified sampling method is used to select 70%-80% as the training set X_train and y_train, and 30%-20% as the test set X_test and y_test. X_train and y_train are input into the RF model for training to obtain the trained fault selection model. X_test and y_test are then input into the trained fault selection model to evaluate its accuracy and other aspects.

[0147] The application process is as follows:

[0148] (1) Obtain data when the actual fault occurs. Collect real-time fault data of the power supply system from the fault recording device of the actual power station, and use it as the data source for fault line selection.

[0149] (2) Feature extraction. In practical applications, the feature extraction steps are the same as those in the training model. The difference is that the data source currently used comes from the fault recording device of the actual power plant to collect real-time fault data of the power supply system.

[0150] (3) RF fault line selection model. The real-time fault data after feature extraction is added as input to the RF model for training, and the fault line selection location is obtained based on the RF training model results.

[0151] The following are several specific embodiments of the application of the present invention.

[0152] Example 1

[0153] In a specific embodiment 1 of the present invention, a diagnostic model training sample set is constructed based on historical grounding fault data of a 0.8kV photovoltaic power generation connection to a distribution line. This sample set contains 1000 samples and 89 grounding fault types. During the sample training process, 80% of the 1000 sets of historical grounding fault data are allocated as the training set for training parameters, and 20% as the test set for testing result accuracy. To determine the optimal hyperparameters of the diagnostic model, the maximum depth of the random forest is set to 50 layers in this example test. Based on this, random forests composed of 50, 100, 150, 200, ... 500 decision trees are used sequentially for diagnostic testing. The relationship between the number of decision trees and the diagnostic accuracy is as follows: Figure 8 As shown.

[0154] When the decision tree is set to 230, the system achieves a precision rate of 41%, a recall rate of 50%, and an online fault location accuracy exceeding 90%.

[0155] Example 2

[0156] In a specific embodiment 2 of the present invention, a diagnostic model training sample set containing 500 samples is constructed based on historical phase-to-phase short-circuit fault data of a 0.8kV photovoltaic power generation line connected to the distribution network. During sample training, 80% of the 500 sets of historical phase-to-phase short-circuit fault data are allocated as a training set for training parameters, and 20% are allocated as a test set for testing result accuracy. To determine the optimal hyperparameters of the diagnostic model, the maximum depth of the random forest is set to 25 layers in this example test. Based on this, random forests composed of 10, 20, 30, 40, ... 100 decision trees are used sequentially for diagnostic testing.

[0157] When the decision tree is set to 35 trees, the system achieves a precision rate of 39%, a recall rate of 46%, and an online fault location accuracy exceeding 90%.

[0158] This invention presents a fault location algorithm for multi-power electronic source power supply systems, based on fault data feature value extraction and random forest algorithm training. It is not limited by the power supply system topology, directly utilizes simulation models or fault waveform data for model training, and uses measured data for feature value extraction and analysis, enabling rapid fault location. Simultaneously, the measured data can be used as training data to improve the accuracy of the trained model, thereby ensuring the accuracy of fault location in multi-power source power supply systems.

[0159] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0160] Except for the technical features described in the specification, all other technologies are known to those skilled in the art.

Claims

1. A fault selection algorithm for a multi-power electronic source power supply system, characterized in that, The fault location algorithm for this multi-power electronic source power supply system includes: Step 1: Establish a fault history dataset; Step 2: Perform data preprocessing and feature extraction; Step 3: Generate fault data training samples and train the model using the random forest algorithm; Step 4: Obtain data at the actual time of the fault and extract its features; Step 5: The fault location is determined by comparing the fault location model established using the random forest algorithm with the training model using the measured data.

2. The fault selection algorithm for a multi-power electronic source power supply system according to claim 1, characterized in that, In step 1, acquire historical fault datasets for single-phase ground faults, two-phase ground faults, two-phase phase-to-phase faults, and three-phase faults. These historical fault datasets are acquired from the simulation platform or collected from the fault recording device of the actual power plant.

3. The fault selection algorithm for a multi-power electronic source power supply system according to claim 1, characterized in that, In step 2, after obtaining the historical fault dataset, the data is preprocessed to filter out invalid data caused by random noise and interference.

4. The fault selection algorithm for a multi-power electronic source power supply system according to claim 1, characterized in that, In step 2, feature extraction consists of three parts: empirical mode decomposition, Hilbert spectral analysis, and eigenvalue calculation.

5. The fault selection algorithm for a multi-power electronic source power supply system according to claim 4, characterized in that, In step 2, feature extraction specifically includes: (a) Empirical mode decomposition is performed on the zero-sequence current data to obtain the modes IMF1-IMF. n ; (b) Select m modes from the set, where m ≤ n, and perform Hilbert spectral analysis on them to obtain the modal components. (c) Calculate the mean, standard deviation, and energy of the data obtained in (b); (d) Obtain features in 3×m×3 dimensions.

6. The fault selection algorithm for a multi-power electronic source power supply system according to claim 5, characterized in that, In step c, the formulas for calculating the mean (Mean), standard deviation (Std), and energy (Energy) are as follows: x i- Let represent the i-th discrete data point, N represent the number of discrete data points, x represent the average value, and x(t) represent the real-time data.

7. The fault selection algorithm for a multi-power electronic source power supply system according to claim 1, characterized in that, In step 3, the features extracted from the fault dataset are used as the sample set X, and the corresponding fault locations are used as the label set y. The training sets X_train and y_train, and the test sets X_test and y_test are selected using a stratified sampling method. X_train and y_train are input into the Random Forest (RF) algorithm model for training to obtain the trained fault line selection model. X_test and y_test are input into the trained fault line selection model to evaluate the accuracy of the model.

8. The fault selection algorithm for a multi-power electronic source power supply system according to claim 1, characterized in that, In step 4, real-time fault data of the power supply system is collected from the fault recording device of the actual power station and used as the data source for fault line selection.

9. The fault selection algorithm for a multi-power electronic source power supply system according to claim 8, characterized in that, In step 4, feature extraction specifically includes: (a) Empirical mode decomposition (EMD) is performed on the zero-sequence current data in the real-time fault data of the power supply system collected from the fault recording device of the actual power plant to obtain the modes IMF1-IMF. n ; (b) Select m modes from the set, where m ≤ n, and perform Hilbert spectral analysis on them to obtain the modal components. (c) Calculate the mean, standard deviation, and energy of the data obtained in (b); (d) Obtain features in 3×m×3 dimensions.

10. The fault location algorithm for a multi-power electronic source power supply system according to claim 1, characterized in that, In step 5, the feature extraction results of the actual fault occurrence data are compared with the training model structure to determine the fault location. That is, the real-time fault data after feature extraction is added as input to the training random forest algorithm (RF) model, and the fault location in step 4 is obtained based on the RF training model results.

11. A fault location system for a multi-power electronic source power supply system, characterized in that, The fault location system for the multi-power electronic power supply system uses the fault location algorithm for the multi-power electronic power supply system described in any one of claims 1-10 to determine the location of the line fault.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program; wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform a fault selection algorithm for a multi-power electronic source power supply system as described in any one of claims 1-10.

Images