Wind turbine generator SCADA data anomaly detection method and related device

Abstract

The invention discloses a wind turbine generator set SCADA data anomaly detection method and a related device, and is applied to the technical field of wind turbine generator sets. Dividing the wind speed range according to a preset interval and dividing the power range according to the preset interval based on a Bins method to obtain a plurality of grids; filling the data points into grids, and determining the row proportion of the data points in each grid in the current row number stronghold and the column proportion of the data points in the current column number stronghold; determining the grid with the maximum row ratio in each row as a row marking grid, and determining the grid with the maximum column ratio in each column as a column marking grid; and determining abnormal data points based on the row mark grids and the column mark grids. According to the method, data anomaly detection is carried out on the SCADA data, so that abnormal data in the SCADA data are removed, the detection result of anomaly detection of the wind turbine generator based on the SCADA data is more accurate, and faults of the wind turbine generator are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine generator systems, and in particular to a method, device, equipment and computer-readable storage medium for detecting abnormality in SCADA data of a wind turbine generator system. Background Art

[0002] The rising failure rate and maintenance difficulties of wind turbines will significantly increase the operation and maintenance costs of wind turbines, which will affect the sustainability of the wind energy industry. Wind turbines are a strongly coupled system with multiple subsystems working together. The abnormal state of the system is likely to be reflected in the coordinated changes of multiple variables. Currently, the SCADA system deployed in the wind farm collects wind turbine SCADA data for abnormal detection of wind turbines, thereby avoiding further damage and catastrophic failures. SCADA data generally contains abnormal data, which affects the accuracy of abnormal detection of wind turbines. Summary of the invention

[0003] The purpose of the present invention is to provide a method, device, equipment and computer-readable storage medium for detecting anomalies in SCADA data of a wind turbine set, which are applied to the technical field of wind turbine sets. The method performs data anomaly detection on SCADA data, thereby removing abnormal data in the SCADA data, making the detection results of anomaly detection of wind turbine sets based on SCADA data more accurate, thereby avoiding the occurrence of wind turbine set failures.

[0004] In order to solve the above technical problems, the present invention provides a method for detecting abnormality of SCADA data of a wind turbine generator set, comprising:

[0005] Acquire SCADA data of a wind turbine; the SCADA data includes a plurality of data points, each of which includes wind speed data and power data;

[0006] Based on the Bins method, the wind speed range is divided into equal intervals according to a first preset interval and the power range is divided into equal intervals according to a second preset interval to obtain a plurality of grids;

[0007] Fill the data points into the grid based on the wind speed data and the power data, and determine the row proportion of the data point in the current row and the column proportion of the data point in the current column in each of the grids;

[0008] Determine the grid with the largest row proportion in each row as a row marking grid, and determine the grid with the largest column proportion in each column as a column marking grid;

[0009] An abnormal grid is determined based on the row marking grid and the column marking grid, and a data point in the abnormal grid is determined as an abnormal data point.

[0010] Optionally, the determining the abnormal grid based on the row marked grid and the column marked grid includes:

[0011] The row proportions of the grids in the left and right directions are accumulated with the column marked grid of each row as the center until the accumulated result is greater than a first preset threshold, and the grids in each row that participate in obtaining the accumulated result are determined as row target grids; wherein the wind speed range thresholds of the grids increase from left to right, and the power range thresholds increase from bottom to top;

[0012] In each row, if the number of target grids in the row is greater than a first preset number and the total number of data points in the row is greater than a second preset number, the grids after the third preset number of grids to the right of the marked grid in the row are marked as abnormal grids;

[0013] In each row, if the number of the row target grids is less than the first preset number, the grids after the fourth preset number of grids to the right of the row mark grid whose row proportion is less than the second preset threshold are marked as abnormal grids;

[0014] In each row, the grids after the fifth preset number of grids to the left of the row marking grid are marked as abnormal grids;

[0015] Determine an inflection point grid from the grids in the top row, determine the column where the inflection point grid is located as the starting column, and determine the column after the sixth preset number of columns to the right of the starting column as the target column;

[0016] In each column from the target column to the rightmost column, the grid below the column marking grid whose column proportion is less than a third preset threshold is marked as the abnormal grid.

[0017] Optionally, determining the inflection point grid from the grid in the top row includes:

[0018] In the top row, the first grid from left to right in which the data point exists is taken as the starting grid, and the first grid from the right of the starting grid whose column ratio is greater than the fourth preset threshold and whose data points are greater than the seventh preset number is determined as the inflection point grid.

[0019] Optionally, before accumulating the row proportions of the grids in the left and right directions with the column mark grid of each row as the center, the method further includes:

[0020] When the row marking grids between adjacent rows differ by an eighth preset number of grids in the horizontal direction, the column marking grids in the lowest row of the adjacent rows are translated upward by a ninth preset number of grids.

[0021] Optionally, the method further includes:

[0022] Selecting the tenth preset number of rows from the top row downward as pre-determined rows, selecting the eleventh preset number of rows above the pre-determined rows as abnormal rows, and the remaining rows as normal rows;

[0023] The grid in the abnormal row is determined as the abnormal grid, and the grid in the normal row is determined as a normal grid; the data points in the normal grid are normal data points.

[0024] Optionally, the method further includes:

[0025] The SCADA data after removing the abnormal data points is input into a neural grid model for secondary abnormal data detection; the neural grid model is an unsupervised model and a semi-supervised model, the unsupervised model includes a reconstruction model and a prediction model, and the encoder of the reconstruction model is a deep autoencoder constrained by MK-MMD distribution.

[0026] Optionally, the method further includes:

[0027] Perform file format check and file quality check on wind turbine CMS data; the file format check includes file name check and file storage type check, and the file quality check includes low amplitude check, zero drift check and impact interference check.

[0028] In order to solve the above technical problems, the present invention provides a wind turbine SCADA data anomaly detection device, comprising:

[0029] The first module is used to obtain SCADA data of the wind turbine; the SCADA data includes multiple data points, each of which includes wind speed data and power data;

[0030] The second module is used to divide the wind speed range into equal intervals according to the first preset interval and divide the power range into equal intervals according to the second preset interval based on the Bins method to obtain multiple grids;

[0031] A third module is used to fill the data points into the grid based on the wind speed data and the power data, and determine the row proportion of the data point in the current row and the column proportion of the data point in the current column in each grid;

[0032] The fourth module is used to determine the grid with the largest row proportion in each row as the row marking grid, and determine the grid with the largest column proportion in each column as the column marking grid;

[0033] The fifth module is used to determine an abnormal grid based on the row marking grid and the column marking grid, and determine the data points in the abnormal grid as abnormal data points.

[0034] In order to solve the above technical problems, the present invention provides a wind turbine SCADA data anomaly detection device, comprising:

[0035] Memory, for storing computer programs;

[0036] A processor is used to implement the above-mentioned wind turbine SCADA data anomaly detection method when executing the computer program.

[0037] In order to solve the above technical problems, the present invention provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the above-mentioned wind turbine SCADA data anomaly detection method is implemented.

[0038] It can be seen that the method of the present invention obtains SCADA data of the wind turbine; the SCADA data contains multiple data points, each data point contains wind speed data and power data; based on the Bins method, the wind speed range is equally spaced according to the first preset spacing and the power range is equally spaced according to the second preset spacing to obtain multiple grids; based on the wind speed data and the power data, the data points are filled into the grids, and the row proportion of the data points in each grid in the total data points of the current row and the column proportion of the data points in the current column are determined; the grid with the largest row proportion in each row is determined as the row marking grid, and the grid with the largest column proportion in each column is determined as the column marking grid; based on the row marking grid and the column marking grid, the abnormal grid is determined, and the data points in the abnormal grid are determined as abnormal data points.

[0039] The method of the present invention performs data anomaly detection on SCADA data, thereby removing abnormal data in the SCADA data, making the detection result of abnormality detection of wind turbines based on SCADA data more accurate, and avoiding the occurrence of wind turbine failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0041] Figure 1 A flow chart of a method for detecting abnormality in SCADA data of a wind turbine generator set provided by an embodiment of the present invention;

[0042] Figure 2 An example diagram of abnormal data points provided by an embodiment of the present invention;

[0043] Figure 3A structural diagram of a wind turbine SCADA data anomaly detection device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] Generally, the design life of a wind turbine is 20 years, and the warranty period is 5 years. This means that behind the rapid development of my country's wind power industry, a large number of wind turbines are approaching or reaching their 20-year design life, and these wind turbines must be equipped with additional maintenance services to achieve subsequent safe production tasks. However, most wind farms are built in remote areas with harsh environments such as mountains, hills, and oceans and few people. The harsh operating environment will inevitably lead to the accumulation of damage to wind turbines, and the failure rate will increase year by year. Therefore, more and more people have noticed that the increased failure rate and maintenance difficulties of wind turbines will significantly increase the operation and maintenance costs of wind turbines, which will affect the sustainability of the wind energy industry.

[0046] Wind turbines are a strongly coupled system with multiple subsystems working together. The abnormal state of the system is likely to be reflected in the coordinated changes of multiple variables. Single variable time series anomaly detection is difficult to correctly reflect the operating state of the system, because the abnormal change of a single variable does not necessarily affect the health of the system. Solve some problems in the data anomaly detection of wind turbine SCADA (Supervisory Control And Data Acquisition) to ensure the reliability and safety of the wind turbine SCADA system operation throughout the project life cycle.

[0047] The SCADA data of wind turbines often contains a large number of bad points that do not conform to the normal output characteristics of wind turbines. For example, under a large measured wind speed, the wind turbine power is much lower than the standard value or even zero. The reasons are many, such as unit failure, maintenance shutdown, unit limited output operation, wind speed sensor failure, etc. These data need to be filtered out.

[0048] In practical applications, most data are screened based on the state recording time. However, wind turbine models are complex, and the state recording rules between units are different. Some units do not even record some abnormal states. In order to meet the subsequent model calculation requirements and improve the applicability and reliability of the model, a Box (grid) method is proposed to add power division on the basis of the traditional Bins method, which can effectively remove bad points in the operating data. The Bins mathematical model is a method of dividing continuous data into discrete intervals.

[0049] The following combination Figure 1 , Figure 1 A flow chart of a method for detecting abnormality in SCADA data of a wind turbine generator system provided by an embodiment of the present invention includes:

[0050] S101: Acquire SCADA data of a wind turbine generator set; the SCADA data includes multiple data points, each of which includes wind speed data and power data.

[0051] This embodiment can obtain SCADA data of wind turbines. Generally, SCADA data can include multiple types of data, such as wind speed, wind direction, rotation speed, current, voltage, vibration and other data. This embodiment can perform anomaly detection on wind speed data and power data in SCADA data, that is, the SCADA data in this embodiment contains multiple data points, and each data point can include wind speed data and power data.

[0052] S102: Based on the Bins method, the wind speed range is divided into equal intervals according to a first preset interval and the power range is divided into equal intervals according to a second preset interval to obtain a plurality of grids.

[0053] In this embodiment, the wind speed data of the data points can be statistically analyzed to determine the wind speed range, and the power data in the data points can be statistically analyzed to determine the power range. Generally, the maximum threshold of the power range can be the sum of the rated power and the redundant power of the wind turbine set. This embodiment does not limit the specific value of the redundant power and can be set based on actual applications.

[0054] This embodiment can divide the wind speed range into equal intervals according to the first preset interval based on the Bins method. This embodiment does not limit the specific method of dividing the wind speed range. Generally, the wind speed range can be divided into equal intervals in the horizontal direction, and the wind speed range threshold of each interval increases from left to right. This embodiment does not limit the specific value of the first preset interval, which can be set based on actual applications, and can generally be 0.25m / s.

[0055] This embodiment can divide the power range into equal intervals according to the second preset interval based on the Bins method. This embodiment does not limit the specific method of dividing the power range. Generally, the power range can be divided into equal intervals in the vertical direction, and the power range threshold of each interval increases from bottom to top. This embodiment does not limit the specific value of the second preset interval, which can be set based on actual applications, and can generally be 25Kw.

[0056] After the wind speed range and the power range are divided into equal intervals, a plurality of grids can be obtained, each grid representing a wind speed value range and a power value range.

[0057] S103: Filling data points into a grid based on wind speed data and power data, and determining the row proportion of the data points in each grid in the total data points in the current row and the column proportion of the data points in the current column.

[0058] S104: Determine the grid with the largest row proportion in each row as the row marking grid, and determine the grid with the largest column proportion in each column as the column marking grid.

[0059] This embodiment can fill the data points into the grid based on the wind speed data and power data of the data points. This embodiment can determine the row ratio of the data point in each grid to all the data points in its row, that is, the ratio of the data point in the grid to all the data points in its row; this embodiment can determine the column ratio of the data point in each grid to all the data points in its column, that is, the ratio of the data point in the grid to all the data points in its column.

[0060] In this embodiment, the grid with the largest row proportion in each row can be determined as the row marking grid, and the grid with the largest column proportion in each column can be determined as the column marking grid.

[0061] According to the distribution pattern of data points, if the wind turbine is in normal operation, the following results will appear: there will always be a grid with the densest distribution of wind speed characteristic points in each row, and there will always be a grid with the densest distribution of power characteristic points in each column, namely the row marking points and column marking points in this embodiment.

[0062] S105: Determine an abnormal grid based on the row-marked grid and the column-marked grid, and determine the data points in the abnormal grid as abnormal data points.

[0063] In this embodiment, an abnormal grid may be determined based on the row-marked grid and the column-marked grid, and a data point in the abnormal grid may be determined as an abnormal data point.

[0064] Since the density of wind speed and power characteristic points in the upper and lower grids with the row mark point and the column mark point as the center will show a decreasing trend, the present embodiment can determine the data inflection point to distinguish the low wind speed section from the high wind speed section. In the low wind speed section, the data points in the grid are horizontally filtered, and the data points in the same power range (row) are counted. The row mark grid is used as the center and extends to the left and right to the set filtering threshold as the row main band. After passing the data inflection point, the high wind speed section is entered, and the data points in the grid are vertically filtered. The column mark grid is used as the center and extends upward and downward to the set filtering threshold as the column main band. The abnormal data points in the non-row main band and / or column main band are eliminated, and the remaining data points in the row main band and column main band are considered to be data under the normal operating state of the wind turbine.

[0065] Specifically, this embodiment can accumulate the row proportions of the left and right grids with the column-marked grid of each row as the center until the accumulated result is greater than the first preset threshold, and determine the grids in each row that participate in obtaining the accumulated result as the row target grid; wherein the wind speed range threshold of the grid increases from left to right, and the power range threshold increases from bottom to top; this embodiment does not limit the size of the first preset threshold, which can generally be 90%, that is, with the column-marked grid of each row as the center, accumulate the row proportions of the left and right grids in turn, and accumulate the row proportions of the left and right grids one at a time. When the accumulated result does not reach the first preset threshold, but the row proportions of the left / right grids have all been accumulated, then only the right / left grids can be accumulated at this time, and the row proportions of one grid can be accumulated each time.

[0066] In this embodiment, before accumulating the row proportions of the grids in the left and right directions with the column mark grid of each row as the center, the following method may also be included:

[0067] Starting from the top row, the tenth preset number of rows are selected downward as pre-determination rows, the eleventh preset number of rows above the pre-determination rows are selected as abnormal rows, and the remaining rows are selected as normal rows; the grids in the abnormal rows are determined as abnormal grids, and the grids in the normal rows are determined as normal grids; the data points in the normal grids are normal data points.

[0068] This embodiment does not limit the specific values ​​of the tenth preset number and the eleventh preset number, and can be set based on actual applications. Specifically, this embodiment can select 6 rows from the top row as pre-determined rows, that is, the top 6 rows are used as pre-determined rows, and the top 2 rows in the pre-determined rows are used as abnormal rows, and the grids in the abnormal rows are determined as abnormal grids, and the grids in the normal rows are determined as normal grids.

[0069] In each row, if the number of row target grids is greater than the first preset number and the total number of row data points is greater than the second preset number, it can be considered that there is a finite power point in the row, and the grids after the third preset number of grids to the right of the row mark grid are marked as abnormal grids; this embodiment does not limit the size of the first preset number, the second preset number and the third preset number; specifically, in each row, if the number of row target grids is greater than 7 and the total number of row data points is greater than 20, the grids after 4 grids to the right of the row mark grid are marked as abnormal grids.

[0070] In each row, if the number of row target grids is less than the first preset number, the grids whose row proportion after the fourth preset number of grids to the right of the row mark grid is less than the second preset threshold are marked as abnormal grids; this embodiment does not limit the size of the fourth preset number and the second preset threshold; specifically, in each row, if the number of row target grids is less than 7, the grids whose row proportion after 4 grids to the right of the row mark grid is less than 3% are marked as abnormal grids.

[0071] In each row, the grids after the fifth preset number of grids to the left of the row marking grid are marked as abnormal grids; this embodiment does not limit the size of the fifth preset number; specifically, in each row, the grids after 6 grids to the left of the row marking grid are marked as abnormal grids.

[0072] Determine the inflection point grid from the grids in the top row, determine the column where the inflection point grid is located as the starting column, and determine the column after the sixth preset number of columns to the right of the starting column as the target column; this embodiment does not limit the specific method of determining the inflection point grid. The inflection point grid of this embodiment is used to divide the high and low wind sections, that is, the rate of change of power after the inflection point grid is close to 0; specifically, this embodiment can use the first grid with data points from left to right in the top row as the starting grid, and determine the grid whose first column from the right of the starting grid has a proportion greater than the fourth preset threshold and whose data points are greater than the seventh preset number as the inflection point grid. This embodiment does not limit the size of the fourth preset threshold and the seventh preset number. Specifically, the grid whose first column from the right of the starting grid has a proportion greater than 1% and whose data points are greater than 3 is determined as the inflection point grid. This embodiment also does not limit the size of the sixth preset number of columns. Generally, the column after 3 columns to the right of the starting column is determined as the target column.

[0073] In each column from the target column to the rightmost column, the grid whose column proportion below the column mark grid is less than the third preset threshold is marked as an abnormal grid; this embodiment does not limit the specific size of the third preset threshold, specifically, the grid whose column proportion below the column mark grid is less than 1% is marked as an abnormal grid.

[0074] In this embodiment, before accumulating the row proportions of the left and right grids with the column mark grids of each row as the center, it also includes: when the row mark grids between adjacent rows differ by an eighth preset number of grids in the horizontal direction, the column mark grids in the lowest row of the adjacent rows are translated upward by a ninth preset number of grids. This embodiment does not limit the size of the eighth preset number and the ninth preset number. Specifically, when the row mark grids between adjacent rows differ by 5 grids in the horizontal direction, the column mark grids in the lowest row of the adjacent rows are translated upward by 2 grids.

[0075] In this embodiment, data points in the abnormal grid can be determined as abnormal data points, and wind speed data and power data of the abnormal data points can be determined as abnormal data. In this embodiment, abnormal data can be removed from the wind turbine SCADA data.

[0076] This embodiment can also perform file format check and file quality check on CMS (Condition Monitoring System) data; in this embodiment, the file format check includes file name check and file storage type check, and the file quality check includes low amplitude check, zero drift check and impact interference check.

[0077] The file format review in this embodiment includes file name check and file storage type check. There are inconsistencies in file naming format and file storage type between each regional end. If the format is not checked in advance and only a fixed file name parsing algorithm and data reading algorithm are adopted, data reading will fail and a large amount of available data will be lost.

[0078] File name check: The vibration files of each region have roughly uniform naming conventions, but the naming details are different, which is reflected in the inconsistent naming format of the channels. In order to make full use of the data with this problem, the channels in the database are expanded, which greatly reduces the data loss caused by the problem of irregular channel naming.

[0079] File format check: Although the vibration data stored in the regional end are all in txt format (text format), the storage format is inconsistent, mainly reflected in the different delimiters. Most wind farms use line feeds as digital delimiters, but a small number of wind farms use spaces as delimiters. If the line feed is used as the basis for reading data alone, it may result in no available data for all units in the wind farm that uses spaces as data delimiters. Therefore, before reading the data, the file storage format needs to be reviewed and the corresponding reading method should be used to ensure smooth reading of the data.

[0080] File quality has a crucial impact on data-driven algorithm models. Based on existing historical data from various regions and previous data analysis experience, files are divided into four quality states to address file quality issues: zero drift, impact interference, low amplitude, and normal.

[0081] Low amplitude: mainly manifested as low signal amplitude level, which may mean that the wind force is too small and the fan blades are rotating slowly. If the amplitude of the vibration data of this channel is always low, it indicates that the sensor of this channel may be detached or damaged. Low amplitude data will make the historical indicator curve of the effective value fluctuate, causing a certain degree of interference and misleading in the diagnosis of fan faults.

[0082] Random shock: There are large non-periodic shocks in the signal, usually from external interference or noise. This type of signal has large kurtosis and spectral kurtosis, but after frequency domain and time domain analysis, no relevant fault information can be seen, which can easily cause false alarms.

[0083] Amplitude mutation: Under normal circumstances, the amplitude of the vibration signal is relatively stable within a certain period of time, and there will be fluctuations in the amplitude within a small range. Therefore, when there is a sudden increase or decrease in the amplitude of the signal, it indicates that the vibration signal is abnormal and the sensor may have problems.

[0084] Zero drift: The vibration signal generally maintains a zero mean, and its time domain waveform is roughly symmetrical about the horizontal axis. However, when the signal has a zero drift, the sensor may have failed and the signal will no longer maintain a zero mean.

[0085] The obvious difference between a signal with normal data quality and an abnormal signal is that the characteristics of each part of the normal signal data are relatively consistent, and there is no obvious mutation. Therefore, this application can divide the data into multiple segments and calculate different statistics for each segment. If the statistical difference is large, it can be determined that the file has abnormal data quality.

[0086] This embodiment does not limit the specific number of segments. Generally, when the periodic impact is relatively sparse, the fault signal may be determined as a random impact anomaly. If the number of segments is too small, the differences between the segments are not obvious. Therefore, after combining a large amount of data analysis, it is found that the number of segments is more appropriate within 5-8.

[0087] Specifically, first calculate the effective value and mean of the entire segment of data, and then divide the wind turbine CMS data into 8 segments, calculate the kurtosis value and mean in each segment, and calculate the variance value and peak-to-peak value of the kurtosis value and the peak-to-peak value of the mean to judge the data quality status.

[0088] Zero drift: The mean of the entire segment of data is greater than a or less than b, or the peak-to-peak value of the mean calculated after dividing into 8 segments is greater than c.

[0089] Impact interference: The kurtosis variance obtained after dividing the data into 8 segments is greater than d or the kurtosis peak-to-peak value obtained after dividing the data into 8 segments is greater than e.

[0090] Low amplitude: The effective value of the entire segment of data is less than f.

[0091] Normal: All situations other than those mentioned above are normal.

[0092] In this embodiment, a, b, c, d, e, and f are constants, and their specific values ​​can be set based on actual applications. Generally, a can be 5, b can be -5, c can be 10, d can be 50, e can be 50, and f can be 0.2.

[0093] In this embodiment, further data anomaly detection can be performed on the SCADA data. Generally, the SCADA data after removing abnormal data points can be input into the neural grid model for secondary abnormal data detection; the neural grid model is an unsupervised model and a semi-supervised model. The unsupervised model includes a reconstruction model and a prediction model. The encoder of the reconstruction model is a deep autoencoder constrained by the MK-MMD distribution.

[0094] Specifically, this embodiment can perform data anomaly detection based on a semi-supervised multivariate time series anomaly detection method combined with an auxiliary discriminator (MK-MMD constrained Deep Autoencoder with Dual-channel Transformer using Auxiliary Discriminator, MDAE-DTAD), and construct an end-to-end semi-supervised anomaly detection framework with an unsupervised model (MK-MMD constrained Deep Autoencoder with Dual-channel Transformer using , MDAE-DT) and a semi-supervised model (Auxiliary Discriminator , AD).

[0095] Based on the above embodiments, the method of the present invention performs data anomaly detection on SCADA data, thereby removing abnormal data in the SCADA data, making the detection result of wind turbine anomaly detection based on SCADA data more accurate, and avoiding the occurrence of wind turbine failure.

[0096] The following is a specific embodiment of a method for detecting abnormality in SCADA data of a wind turbine generator set provided by an embodiment of the present invention:

[0097] The wind speed range is divided into equal intervals of 0.25 m / s in the horizontal direction, and the wind speed range threshold of each interval increases from left to right; the power range is divided into equal intervals of 25 kW in the vertical direction, and the power range threshold of each interval increases from bottom to top; after the division is completed, multiple grids are obtained;

[0098] Fill the data points into the grid based on the wind speed data and the power data, determine the row ratio of the data points in each grid in the total data points of the current row and the column ratio of the data points in the current column, determine the grid with the largest row ratio in each row as the row marking grid, and determine the grid with the largest column ratio in each column as the column marking grid;

[0099] Six rows from the top row are selected as pre-determination rows, that is, the top six rows are used as pre-determination rows; the data in the pre-determination rows do not need to be detected for abnormalities;

[0100] When the row marker grids between adjacent rows differ by 5 grids horizontally, the column marker grid of the lowest row in the adjacent row is shifted upward by 2 grids;

[0101] The row proportions of the left and right grids are accumulated with the column marked grid of each row as the center until the accumulated result is greater than 90%, and the grids in each row that participate in the accumulated result are determined as the row target grids; wherein, the wind speed range threshold of the grid increases from left to right, and the power range threshold increases from bottom to top; each time the row proportions of the left and right grids are accumulated one by one, when the accumulated result does not reach the first preset threshold, but the row proportions of the left / right grids have all been accumulated, then only the right / left grids can be accumulated at this time, and the row proportions of one grid are accumulated each time;

[0102] In each row, if the number of row target grids is greater than 7 and the total number of row data points is greater than 20, the grids 4 grids to the right of the row mark grid are marked as abnormal grids; in each row, if the number of row target grids is less than 7, the grids 4 grids to the right of the row mark grid and the row proportion is less than 3% are marked as abnormal grids; in each row, the grids 6 grids to the left of the row mark grid are marked as abnormal grids;

[0103] The first grid with data points from left to right is taken as the starting grid, and the first column to the right of the starting grid with a percentage greater than 1% and more than 3 data points is determined as the inflection point grid; the column where the inflection point grid is located is determined as the starting column, and the column 3 columns to the right of the starting column is determined as the target column;

[0104] In each column from the target column to the rightmost column, mark the grids whose column proportion below the column marking grid is less than 1% as abnormal grids;

[0105] The two rows at the top of the pre-determined rows are regarded as abnormal rows, and the grids in the abnormal rows are determined as abnormal grids;

[0106] The abnormal data points in the abnormal grid are marked and filtered. The marked abnormal data points can be Figure 2 As shown, the red data points are abnormal data points and the black data points are normal data points.

[0107] The following combination Figure 3 , Figure 3 A wind turbine SCADA data anomaly detection device provided by an embodiment of the present invention may include:

[0108] The first module 100 is used to obtain SCADA data of the wind turbine; the SCADA data includes a plurality of data points, each of which includes wind speed data and power data;

[0109] The second module 200 is used to divide the wind speed range into equal intervals according to the first preset interval and divide the power range into equal intervals according to the second preset interval based on the Bins method to obtain multiple grids;

[0110] The third module 300 is used to fill the data points into the grid based on the wind speed data and the power data, and determine the row proportion of the data point in the current row and the column proportion of the data point in the current column in each grid;

[0111] The fourth module 400 is used to determine the grid with the largest row proportion in each row as a row marking grid, and determine the grid with the largest column proportion in each column as a column marking grid;

[0112] The fifth module 500 is used to determine an abnormal grid based on the row marking grid and the column marking grid, and determine the data points in the abnormal grid as abnormal data points.

[0113] Based on the above embodiments, the method of the present invention performs data anomaly detection on SCADA data, thereby removing abnormal data in the SCADA data, making the detection result of wind turbine anomaly detection based on SCADA data more accurate, and avoiding the occurrence of wind turbine failure.

[0114] Based on the above embodiment, the fifth module 500 may include:

[0115] The first unit is used to accumulate the row proportions of the grids in the left and right directions with the column marked grid of each row as the center until the accumulated result is greater than a first preset threshold, and determine the grids in each row that participate in obtaining the accumulated result as row target grids; wherein the wind speed range thresholds of the grids increase from left to right, and the power range thresholds increase from bottom to top;

[0116] The second unit is used for marking the grids after a third preset number of grids to the right of the row mark grid as the abnormal grids in each row if the number of the row target grids is greater than a first preset number and the total number of row data points is greater than a second preset number;

[0117] A third unit is used for marking the grids whose row proportion is less than the second preset threshold after the fourth preset number of grids to the right of the row mark grid as the abnormal grids if the number of the row target grids is less than the first preset number in each row;

[0118] The fourth unit is used to mark the grids after the fifth preset number of grids to the left of the row marking grid as the abnormal grids in each row;

[0119] A fifth unit is used to determine an inflection point grid from the grids in the top row, determine the column where the inflection point grid is located as the starting column, and determine the column after the sixth preset number of columns to the right of the starting column as the target column;

[0120] The sixth unit is used to mark the grid whose column proportion below the column marking grid is less than a third preset threshold as the abnormal grid in each column from the target column to the rightmost column.

[0121] Based on the above embodiments, the fifth unit may include:

[0122] The first subunit is used to, in the top row, take the first grid from left to right where the data point exists as the starting grid, and determine the first grid from the right of the starting grid whose column ratio is greater than a fourth preset threshold and whose data points are greater than a seventh preset number as the inflection point grid.

[0123] Based on the above embodiments, before the first unit, the device may further include:

[0124] The seventh unit is used to translate the column marking grid in the lowest row of adjacent rows upward by a ninth preset number of grids when the row marking grids between adjacent rows differ by an eighth preset number of grids in the horizontal direction.

[0125] Based on the above embodiments, the device may further include:

[0126] An eighth unit is used to select a tenth preset number of rows from the top row downward as pre-determined rows, select an eleventh preset number of rows above the pre-determined rows as abnormal rows, and the remaining rows as normal rows;

[0127] The ninth unit is used to determine the grid in the abnormal row as the abnormal grid, and determine the grid in the normal row as a normal grid; the data points in the normal grid are normal data points.

[0128] Based on the above embodiments, the device may further include:

[0129] The sixth module is used to input the SCADA data after removing the abnormal data points into the neural grid model for secondary abnormal data detection; the neural grid model is an unsupervised model and a semi-supervised model, the unsupervised model includes a reconstruction model and a prediction model, and the encoder of the reconstruction model is a deep autoencoder constrained by MK-MMD distribution.

[0130] Based on the above embodiments, the device may further include:

[0131] The seventh module is used to perform file format check and file quality check on the CMS data of the wind turbine generator set; the file format check includes file name check and file storage type check, and the file quality check includes low amplitude check, zero drift check and impact interference check.

[0132] Based on the above embodiments, the present invention also provides a wind turbine SCADA data anomaly detection device, which may include a memory and a processor, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the steps provided in the above embodiments can be implemented. Of course, the device may also include various necessary network interfaces, power supplies, and other components.

[0133] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by an execution terminal or a processor, the method provided in the embodiment of the present invention can be implemented; the storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0134] In this article, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

Claims

1. A method for detecting abnormality in SCADA data of a wind turbine generator set, characterized in that: include: Obtain SCADA data from wind turbines; The SCADA data includes a plurality of data points, each of which includes wind speed data and power data; Based on the Bins method, the wind speed range is divided into equal intervals according to the first preset interval and the power range is divided into equal intervals according to the second preset interval to obtain a plurality of grids; Fill the data points into the grid based on the wind speed data and the power data, and determine the row proportion of the data point in the current row and the column proportion of the data point in the current column in each of the grids; Determine the grid with the largest row proportion in each row as a row marking grid, and determine the grid with the largest column proportion in each column as a column marking grid; An abnormal grid is determined based on the row marking grid and the column marking grid, and a data point in the abnormal grid is determined as an abnormal data point.

2. The wind turbine SCADA data anomaly detection method according to claim 1, characterized in that: The determining of abnormal grids based on the row marked grids and the column marked grids includes: The row proportions of the grids in the left and right directions are accumulated with the column marked grid of each row as the center until the accumulated result is greater than a first preset threshold, and the grids in each row that participate in obtaining the accumulated result are determined as row target grids; wherein the wind speed range thresholds of the grids increase from left to right, and the power range thresholds increase from bottom to top; In each row, if the number of target grids in the row is greater than a first preset number and the total number of data points in the row is greater than a second preset number, the grids after the third preset number of grids to the right of the marked grid in the row are marked as abnormal grids; In each row, if the number of the row target grids is less than the first preset number, the grids after the fourth preset number of grids to the right of the row mark grid whose row proportion is less than the second preset threshold are marked as abnormal grids; In each row, the grids after the fifth preset number of grids to the left of the row marking grid are marked as abnormal grids; Determine an inflection point grid from the grids in the top row, determine the column where the inflection point grid is located as the starting column, and determine the column after the sixth preset number of columns to the right of the starting column as the target column; In each column from the target column to the rightmost column, the grid below the column marking grid whose column proportion is less than a third preset threshold is marked as the abnormal grid.

3. The wind turbine SCADA data anomaly detection method according to claim 2 is characterized in that: Determining the inflection point grid from the grid in the uppermost row includes: In the top row, the first grid from left to right in which the data point exists is taken as the starting grid, and the first grid from the right of the starting grid whose column ratio is greater than the fourth preset threshold and whose data points are greater than the seventh preset number is determined as the inflection point grid.

4. The wind turbine SCADA data anomaly detection method according to claim 2, characterized in that: Before accumulating the row proportions of the grids in the left and right directions with the column mark grid of each row as the center, the method further includes: When the row marking grids between adjacent rows differ by an eighth preset number of grids in the horizontal direction, the column marking grids in the lowest row of the adjacent rows are translated upward by a ninth preset number of grids.

5. The wind turbine SCADA data anomaly detection method according to claim 2, characterized in that: Also includes: Selecting the tenth preset number of rows from the top row downward as pre-determined rows, selecting the eleventh preset number of rows above the pre-determined rows as abnormal rows, and the remaining rows as normal rows; Determine the grid in the abnormal row as the abnormal grid, and determine the grid in the normal row as the normal grid; The data points in the normal grid are normal data points.

6. The wind turbine SCADA data anomaly detection method according to claim 1, characterized in that: Also includes: The SCADA data after removing the abnormal data points is input into a neural grid model for secondary abnormal data detection; the neural grid model is an unsupervised model and a semi-supervised model, the unsupervised model includes a reconstruction model and a prediction model, and the encoder of the reconstruction model is a deep autoencoder constrained by MK-MMD distribution.

7. The wind turbine SCADA data anomaly detection method according to claim 1, characterized in that: Also includes: Perform file format check and file quality check on wind turbine CMS data; the file format check includes file name check and file storage type check, and the file quality check includes low amplitude check, zero drift check and impact interference check.

8. A wind turbine SCADA data anomaly detection device, characterized in that: include: The first module is used to obtain SCADA data of wind turbines; The SCADA data includes a plurality of data points, each of which includes wind speed data and power data; The second module is used to divide the wind speed range into equal intervals according to the first preset interval and divide the power range into equal intervals according to the second preset interval based on the Bins method to obtain multiple grids; A third module is used to fill the data points into the grid based on the wind speed data and the power data, and determine the row proportion of the data point in the current row and the column proportion of the data point in the current column in each grid; The fourth module is used to determine the grid with the largest row proportion in each row as a row marking grid, and determine the grid with the largest column proportion in each column as a column marking grid; The fifth module is used to determine an abnormal grid based on the row marking grid and the column marking grid, and determine the data points in the abnormal grid as abnormal data points.

9. A wind turbine SCADA data anomaly detection device, characterized in that: include: Memory, for storing computer programs; A processor is used to implement the wind turbine SCADA data anomaly detection method as claimed in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, the wind turbine SCADA data anomaly detection method as described in any one of claims 1 to 7 is implemented.

Images