Converter station fault type identification method and computer equipment

Abstract

The invention belongs to the technical field of fault identification, and particularly relates to a converter station fault type identification method and computer equipment. The method comprises the following steps: acquiring operation data of a converter station, extracting feature data required for identifying a fault type of a certain region in the converter station from the operation data, and inputting the extracted feature data into a fault type identification model corresponding to the region to obtain the fault type of the region; wherein the fault type identification model corresponding to a certain region is obtained by training a machine learning model by using a data set of feature data of the region containing a fault type label; areas in the converter station comprise a valve area, a converter transformer area, a pole area, a bipolar area, a direct current filter area and a direct current line area. According to the method, one machine learning model aims at one area, the accuracy of fault type identification of the area can be achieved only through a small amount of targeted feature data, and the requirement of a modern electric power system for fault processing is met.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault identification, and in particular relates to a converter station fault type identification method and computer equipment. Background Art

[0002] Flexible DC transmission is a new generation of DC transmission technology, which mainly includes converter stations and is a DC transmission technology in high-voltage DC transmission. The converter station can change the voltage, adjust the voltage, and maintain voltage stability between the bus side and the distribution line side to ensure the stable and reliable operation of the distribution system. If a converter station fails, it will cause a large power loss, seriously threatening the safety of AC and DC power grids and equipment, and may even face the risk of power outages. Therefore, it is extremely important to quickly locate and accurately identify the faults of the converter station.

[0003] With the development of artificial intelligence technology, machine learning will play an increasingly important role in fault detection and diagnosis, and machine learning technology will become a more intelligent and efficient solution for industrial production and equipment maintenance. In practical applications, a machine learning model is often set up to detect which specific area of ​​the entire converter station has a fault and the corresponding fault type. In other words, this solution has high requirements for this machine learning model, which must be able to learn both which specific area has a fault and the type of fault that has occurred. However, the entire converter station is large and contains many areas, each of which exhibits different fault characteristics. Setting a single model cannot take into account the characteristics of all areas, which will lead to low model recognition accuracy, that is, low fault type recognition accuracy; and if all fault feature data of all areas are used as input to this machine learning model, the data volume will be large and the model training time will be long. Summary of the invention

[0004] The object of the present invention is to provide a converter station fault type identification method and computer equipment to solve the problem of low fault type identification accuracy caused by the use of a single machine learning model for fault type identification in the entire converter station.

[0005] In order to solve the above technical problems, the present invention provides a method for identifying the fault type of a converter station, comprising the following steps: obtaining the operating data of the converter station, extracting the characteristic data required for identifying the fault type of a certain area in the converter station, and inputting the extracted characteristic data into a fault type identification model corresponding to the area to obtain the fault type of the area; wherein the fault type identification model corresponding to a certain area is obtained by training a machine learning model using a data set of characteristic data of the area containing fault type labels.

[0006] Furthermore, the areas in the converter station include a valve area, a converter transformer area, a pole area, a bipolar area, a DC filter area and a DC line area.

[0007] Furthermore, the characteristic data required for a certain regional fault type is characteristic data selected from pre-selected characteristic data and having a high correlation with the identification of the regional fault type; the pre-selected characteristic data include: at least one of the statistical characteristics of each sub-signal, frequency domain characteristics, discrete characteristics reflecting the discrete degree of the signal, and random characteristics reflecting the uncertainty of the signal; wherein, wavelet transform is performed on the converter station operation data to obtain sub-signals of different frequency bands.

[0008] Furthermore, the statistical features include at least one of the mean, standard deviation, maximum value, minimum value, skewness, kurtosis, median, sum of absolute values, and sum of squares of the signal.

[0009] Furthermore, the frequency domain characteristics include at least one of the power spectrum, power spectrum energy, maximum power frequency, and maximum power of the signal.

[0010] Furthermore, the discrete features include entropy of the signal.

[0011] Further, the random characteristics include the coefficient of variation of the signal.

[0012] Furthermore, the converter station operation data is data obtained by preprocessing the original collected converter station operation data; the preprocessing includes filtering to filter out irrelevant frequency noise.

[0013] Furthermore, a Butterworth filter is used to implement the filtering.

[0014] Furthermore, the machine learning model is a convolutional neural network model.

[0015] In order to solve the above technical problem, the present invention further provides a computer device, including a processor, wherein the processor is used to execute a computer program to implement the steps of the converter station fault type identification method introduced above.

[0016] The present invention is a pioneering invention with the following beneficial effects: the present invention changes the idea of ​​the application object of the machine learning model in the prior art, and adjusts the idea of ​​applying a machine learning model in the prior art to the entire converter station, thereby using the machine learning model to identify which specific area in the entire converter station has a fault and the corresponding fault type, to a machine learning model applied to a specific area in the converter station, thereby using the machine learning to identify the fault type of the specific area, and the input of the corresponding machine learning model is only some feature data closely related to the fault type identification of the area, and does not require all the feature data in the entire converter station. Overall, the present invention uses a machine learning model for one area, and only a small amount of targeted feature data is needed to complete the accuracy of fault type identification in the area, meeting the needs of modern power systems for fault handling. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a method flow chart in an embodiment of the method of the present invention;

[0018] Figure 2 is an original data graph in an embodiment of the method of the present invention;

[0019] Figure 3 It is a matrix diagram of the original data after dimension reduction and normalization in the embodiment of the method of the present invention;

[0020] Figure 4 is a feature extraction flow chart in an embodiment of the method of the present invention;

[0021] Figure 5 It is a waveform diagram of a fault recording signal in an embodiment of the method of the present invention. DETAILED DESCRIPTION

[0022] The core idea of ​​the present invention is that a machine learning model corresponds to an area in the converter station, and the entire converter station is divided into valve area, converter transformer area, pole area, bipolar area, DC filter area and DC line area. Only the feature data required to identify the type of fault in a certain area is used, and input into the trained machine learning model corresponding to the area, so that the type of fault in the area can be identified. The present invention only needs a small amount of targeted feature data to complete the accuracy of fault type identification in the area, meeting the needs of modern power systems for fault handling. In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0023] Converter station fault type identification method embodiment:

[0024] An embodiment of a converter station fault type identification method of the present invention, the entire implementation process is as follows Figure 1 As shown, the details are as follows:

[0025] Step 1: Define the area and fault type of the converter station. Specifically, divide the area of ​​the entire converter station into six areas: valve area, converter transformer area, pole area, bipolar area, DC filter area and DC line area. Determine the possible fault type in each area, such as switch fault, short circuit fault, etc., and obtain the fault recording data of each fault type in each area (including normal data, fault type including normal). Specifically:

[0026] The required fault recording data is obtained by preprocessing the data of the converter station, specifically, it is extracted from the data of the converter station. The data of the converter station is stored in the CSV file format. The CSV file is a plain text file. CSV stands for "Comma-Separated Values", with commas as delimiters between fields and line breaks as delimiters between records.

[0027] In the original data of the fault recording, each data is an N×M matrix, such as Figure 2 As shown in the figure, N is generally greater than or equal to 10,000, and M is generally greater than or equal to 30. With such a data scale, if each type has 1,000 samples, the entire model training time will be very long and consume a lot of computing resources. Therefore, a more optimized processing method in this application is to first reduce the dimensionality of the sample data and convert it into a 60×M (for example only) matrix. The normalized result is as follows Figure 3 As shown in the figure, the color in the figure represents the value, which greatly reduces the data size, not only reducing the training time of the model, but also reducing the response time of the model. The final fault recording data is as follows Figure 5 As shown. Moreover, these data are all time series data, usually voltage, current and other signals. These data can refer to the data in Table 1 below (the current criterion in Table 1 is actually the current value, and the voltage criterion is actually the voltage value). These data are preprocessed, including filtering and normalization to eliminate noise, in order to improve the quality and efficiency of subsequent processing. Among them, a bandpass Butterworth filter can be used to eliminate irrelevant frequency noise. The filter provides maximum attenuation at the cutoff frequency, allowing the signal components within the specified frequency range to remain unchanged and remove other components. The expression is:

[0028]

[0029] Where H(ω) represents the filter output, ω represents the frequency, and ω c represents the cutoff frequency, and n represents the filter order.

[0030] Table 1

[0031] IDL_I(S2+PPR1A): Current criterion (pole 1 protection system A) UDL_SW_IV(S2+PPR1A): Voltage criterion (pole 1 protection system A) IDL_I(S2+PPR1B): Current criterion (pole 1 protection system B) UDL_SW_IV(S2+PPR1B: Voltage criterion (pole 1 pole protection system B) IDL_I(S2+PPR1C): Current criterion (pole 1 protection system C) UDL_SW_IV(S2+PPR1C): Voltage criterion (pole 1 protection system C) IDL_I(S2+PPR2A): Current criterion (pole 2 pole protection system A) UDL_SW_IV(S2+PPR2A): Voltage criterion (pole 2 pole protection system A) IDL_I(S2+PPR2B): Current criterion (pole-to-pole protection system B) UDL_SW_IV(S2+PPR2B): Voltage criterion (pole-to-pole protection system B) IDL_I(S2+PPR2C): Current criterion (pole-to-pole protection system C) UDL_SW_IV(S2+PPR2C): Voltage criterion (pole 2 pole protection system C)

[0032] Step 2: Extract features from the acquired fault recording data and select feature data closely related to the fault type identification of each area.

[0033] The benefits of feature extraction are that, first, it can extract representative features from the original signal, convert the original high-dimensional signal data into low-dimensional model data, reduce the data dimension, greatly reduce the complexity of data processing, and reduce the complexity of calculation; secondly, the use of low-dimensional feature data significantly improves the computational efficiency of model training and prediction, meeting the requirements of the power system for rapid fault identification.

[0034] When dealing with specific Figure 4 As shown, firstly, each data is subjected to wavelet transform and decomposed into sub-signals of different frequency bands. Then, for each sub-signal, its statistical characteristics, frequency domain characteristics, discrete characteristics reflecting the discrete degree of the signal, and random characteristics reflecting the uncertainty of the signal are calculated (one, two, or more than three of these characteristics can be selected according to the actual situation). Then, feature selection is performed to screen out the features closely related to each area.

[0035] Among them, multi-scale wavelet transform is used to capture the time-frequency characteristics of the signal. Wavelet transform provides better time and frequency localization information than traditional Fourier transform, and its expression is:

[0036]

[0037] Where W f represents the wavelet coefficient, a represents the scale parameter, b represents the translation parameter, ψ() represents the mother wavelet function, f(t) represents the original signal, and t represents time. This transform analyzes signal components of different scales (frequency) and positions (time), and is particularly suitable for non-stationary signals.

[0038] Statistical features include: mean, standard deviation, maximum, minimum, skewness, kurtosis, median, sum of absolute values, sum of squares, etc. Frequency domain features include: power spectrum, power spectrum energy, maximum power frequency, maximum power, etc. These statistical features and frequency domain features can be selected from one, two or more than three according to actual conditions. Discrete features include: coefficient of variation. Random features include: entropy of the signal.

[0039] The features that are most relevant to fault type identification in each region may be the same or different.

[0040] Step three: Build a dedicated machine learning model for each region.

[0041] The machine learning model can be a convolutional neural network (CNN) model, and CNN has residual blocks to improve depth and accuracy.

[0042] In step 4, the feature data including the fault type labels corresponding to each area obtained in step 2 are taken as a data set, and divided into a training set and a test set. The machine learning model constructed in step 3 is trained and tested accordingly to obtain a fault type recognition model corresponding to each area.

[0043] In this step, the accuracy and generalization ability of the model are verified on an independent test set, and cross-validation is performed to ensure that the model can maintain stable recognition performance under different conditions.

[0044] Step 5: deploy the model obtained in step 4 to the actual monitoring system to monitor the recorded data in real time. Specifically, the operation data of the converter station is obtained, and the characteristic data required to predict the fault type of a certain area in the converter station is extracted from it, and the extracted characteristic data is input into the fault type identification model corresponding to the area to predict the fault type of the area, so that when a new fault occurs, the deployed model can be used to quickly identify the fault type and location.

[0045] Step six: Collect the model recognition results of step five and actual fault handling feedback, analyze the model performance, identify possible misjudgments and missed judgments, adjust and optimize the model parameters based on the feedback, and improve the model recognition accuracy and response speed.

[0046] Step 7: Using the developed user interface, operators can more easily view fault alarms, identification results and historical records, and automatically generate fault reports, providing technicians with detailed fault data and analysis to assist in quickly and accurately handling faults.

[0047] Computer equipment example:

[0048] A computer device embodiment of the present invention includes a memory, a processor, an internal bus and a computer program stored in the memory, and the processor and the memory communicate and exchange data with each other through the internal bus. The processor executes the computer program to implement the steps of the method described in the embodiment of the converter station fault type identification method of the present invention. The processor can be a processing device such as a microprocessor MCU, a programmable logic device FPGA, etc.; the memory can be various memories that use electrical energy to store information, such as RAM, ROM, etc., or it can be a memory using other methods.

[0049] In summary, in view of the problems such as low accuracy of traditional fault identification methods, and the fact that different fault types are distributed in different areas (valve area, converter area, pole area, bipolar area, DC filter area and DC line area), which makes the fault characteristics of each area different, and the problem that the existing single model cannot take into account the characteristics of all areas, the present invention proposes a fault type identification method for fault characteristics of different areas. Each area corresponds to its own proprietary fault type identification model, and each fault type identification model has its own corresponding feature data according to the characteristics of the area, so as to improve the accuracy and efficiency of fault identification and meet the needs of modern power systems for fault handling.

[0050] Specific implementation methods are given above, but the present invention is not limited to the described implementation methods. The basic idea of ​​the present invention lies in the above basic scheme. For ordinary technicians in this field, it does not take creative work to design various deformed models, formulas, and parameters according to the teachings of the present invention. Changes, modifications, substitutions, and variations of the implementation methods without departing from the principles and spirit of the present invention still fall within the scope of protection of the present invention.

Claims

1. A method for identifying fault types of converter stations, characterized in that: The steps include: The converter station operation data is obtained, characteristic data required for identifying the fault type of a certain area in the converter station is extracted, and the extracted characteristic data is input into a fault type identification model corresponding to the area to obtain the fault type of the area; Among them, the fault type recognition model corresponding to a certain area is obtained by training the machine learning model using a data set of feature data of the area containing fault type labels.

2. The converter station fault type identification method according to claim 1, characterized in that: The areas in the converter station include valve area, converter transformer area, pole area, bipolar area, DC filter area and DC line area.

3. The method for identifying converter station fault types according to claim 1, characterized in that: The characteristic data required for a certain regional fault type is characteristic data selected from pre-selected characteristic data and having a high correlation with the identification of the regional fault type; the pre-selected characteristic data includes: at least one characteristic data of the statistical characteristics of each sub-signal, frequency domain characteristics, discrete characteristics reflecting the discrete degree of the signal, and random characteristics reflecting the uncertainty of the signal; wherein, wavelet transform is performed on the converter station operation data to obtain sub-signals of different frequency bands.

4. The converter station fault type identification method according to claim 3, characterized in that: The statistical features include at least one of the signal's mean, standard deviation, maximum value, minimum value, skewness, kurtosis, median, sum of absolute values, and sum of squares.

5. The converter station fault type identification method according to claim 3, characterized in that: The frequency domain features include at least one of the power spectrum, power spectrum energy, maximum power frequency, and maximum power of the signal.

6. The converter station fault type identification method according to claim 3, characterized in that: Discrete features include the entropy of the signal.

7. The converter station fault type identification method according to claim 3, characterized in that: Random characteristics include the coefficient of variation of the signal.

8. The converter station fault type identification method according to claim 1, characterized in that: The converter station operation data is data obtained by preprocessing the original collected converter station operation data; The pre-processing includes filtering to remove irrelevant frequency noise.

9. The converter station fault type identification method according to claim 8, characterized in that: The filtering is implemented using a Butterworth filter.

10. A computer device comprising a processor, characterized in that: The processor is used to execute a computer program to implement the steps of the method according to any one of claims 1 to 9.

Images