A method and device for automatically marking fault data

Abstract

The present disclosure provides a method and device for automatic fault data labeling, which belongs to the field of data processing technology. The method includes: cleaning a first data set once to obtain a second data set; the first data set is a historical operation data set of a target device; the first data set includes a fault data set to be labeled; determining key feature parameters according to the fault data set to be labeled, and performing a second cleaning on the second data set based on the key feature parameters to obtain a third data set; determining an over-limit threshold corresponding to the fault type based on the third data set according to the fault type; labeling the fault data set to be labeled based on the over-limit threshold to obtain a fault labeling data set. The present disclosure can solve the problems of low fault data set labeling accuracy and large workload in the process of wind turbine fault diagnosis, and improve the work efficiency of the wind turbine fault diagnosis sample data extraction process.

Description

Technical Field

[0001] The present disclosure belongs to the field of data processing technology, and more specifically, to a method and device for automatically marking fault data. Background Art

[0002] With the rapid development of wind power generation industry, the installed capacity of wind turbines has been increasing year by year. As an important medium for obtaining wind energy resources, wind turbines are mostly installed in areas with harsh conditions, which brings severe challenges to the safe and stable operation of the units. The wind turbine data acquisition and supervisory control (SCADA) system contains hundreds of parameters related to the operating performance of wind turbines. The high-dimensional massive data accumulated over a long period of time can provide a wealth of information.

[0003] Because SCADA data is extremely large, the existing fault data annotation is generally done manually, which is labor-intensive and has low accuracy. Therefore, there is an urgent need for a method and device for automatic fault data annotation. Summary of the invention

[0004] The purpose of the present invention is to provide a method and device for automatic fault data labeling to solve the problems of low accuracy and large workload of manual labeling.

[0005] A first aspect of an embodiment of the present disclosure provides a method for automatically labeling fault data, comprising:

[0006] The first data set is cleaned once to obtain a second data set; the first data set is a historical operation data set of the target device; the first data set includes a fault data set to be annotated;

[0007] Determine key characteristic parameters according to the fault data set to be annotated, and perform secondary cleaning on the second data set based on the key characteristic parameters to obtain a third data set;

[0008] According to the fault type, determining an over-limit threshold corresponding to the fault type based on the third data set;

[0009] The to-be-annotated fault data set is annotated based on the over-limit threshold to obtain a fault annotated data set.

[0010] A second aspect of the embodiments of the present disclosure provides a device for automatically labeling fault data, comprising:

[0011] A first data cleaning unit is used to clean the first data set once to obtain a second data set; the first data set is a historical operation data set of the target device; the first data set includes a fault data set to be marked;

[0012] A second data cleaning unit, configured to determine key characteristic parameters according to the fault data set to be labeled, and perform secondary cleaning on the second data set based on the key characteristic parameters to obtain a third data set;

[0013] a fault threshold determination unit, configured to determine, according to the fault type, an over-limit threshold corresponding to the fault type based on the third data set;

[0014] The data labeling unit is used to label the to-be-labeled fault data set based on the over-limit threshold to obtain a fault labeling data set.

[0015] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of the above-mentioned method for automatic fault data labeling when executing the computer program.

[0016] According to a fourth aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method for automatic fault data labeling are implemented.

[0017] The beneficial effects of the method and device for automatically labeling fault data provided by the embodiments of the present disclosure are:

[0018] On the one hand, the automatic fault data labeling method proposed in the present disclosure is different from the existing idea of ​​data cleaning based only on the wind speed-power curve in the data cleaning stage. The present disclosure performs a secondary deep cleaning of the data based on the key feature parameters corresponding to the fault data set to be labeled based on the machine learning algorithm (isolation forest algorithm), further improving the data quality and obtaining a more accurate threshold.

[0019] On the other hand, the present disclosure can realize automatic labeling of fault data, greatly reducing manual participation, improving data processing efficiency, and reducing costs. At the same time, due to the improvement of the accuracy and reliability of the labeling, the fault analysis and prediction results based on these labeled data are also more accurate and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, 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 some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 A schematic diagram of a flow chart of a method for automatically labeling fault data provided by an embodiment of the present disclosure;

[0022] Figure 2 A schematic diagram of a specific process of a method for automatically labeling fault data provided by an embodiment of the present disclosure;

[0023] Figure 3 A comparison chart of cleaning data of a wind turbine generator set A provided in one embodiment of the present disclosure;

[0024] Figure 4 A comparison chart of secondary cleaning data of wind turbine generator set A provided in one embodiment of the present disclosure;

[0025] Figure 5 A data comparison diagram of different marking methods for a wind turbine generator set A provided in an embodiment of the present disclosure;

[0026] Figure 6 A comparison chart of one cleaning data of a wind turbine generator set B provided in an embodiment of the present disclosure;

[0027] Figure 7 A comparison chart of secondary cleaning data of wind turbine generator set B provided in one embodiment of the present disclosure;

[0028] Figure 8 A data comparison diagram of different marking methods for a wind turbine generator set B provided in an embodiment of the present disclosure;

[0029] Fig. 9 A comparison chart of one cleaning data of a wind turbine generator set C provided in an embodiment of the present disclosure;

[0030] Fig.10 A comparison chart of secondary cleaning data of a wind turbine generator set C provided in an embodiment of the present disclosure;

[0031] Fig.11 A schematic diagram of fault data annotation of a wind turbine generator set C provided in an embodiment of the present disclosure;

[0032] Fig.12 A comparison chart of cleaning data of a wind turbine E provided in one embodiment of the present disclosure;

[0033] Fig.13 A comparison chart of secondary cleaning data of a wind turbine E provided in an embodiment of the present disclosure;

[0034] Fig.14 A schematic diagram of fault data annotation of a wind turbine E provided in an embodiment of the present disclosure;

[0035] Fig.15 A structural block diagram of a device for automatically marking fault data provided by an embodiment of the present disclosure;

[0036] Fig.16 A schematic block diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0037] In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present disclosure. However, it should be clear to those skilled in the art that the present disclosure may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obstructing the description of the present disclosure with unnecessary details.

[0038] The SCADA system contains hundreds of parameters related to the operating performance of wind turbines: parameters that reflect the performance of the unit, such as speed, power, voltage, and current; environmental parameters that have a direct impact on the operating conditions of the unit, such as wind speed and wind direction; setting and feedback parameters related to control, such as yaw angle, blade angle, and torque; status monitoring parameters such as temperature, oil level, and bearing vibration, as well as equipment start and stop status, fault alarm time, and type labels, etc. The high-dimensional mass data accumulated over a long period of time can provide a wealth of information.

[0039] When using neural networks, machine learning and other methods to establish a wind turbine fault diagnosis model, many fault samples are needed to train the model, and the effectiveness of the model depends on the accurate labeling of many fault data. Data labeling is a key link for most artificial intelligence algorithms to operate effectively. The more accurate the data labeling and the larger the amount of labeled data, the better the performance of the algorithm.

[0040] In the past, the annotation of wind turbine parameter over-limit fault data sets was mostly done by querying the information given in the fault manual, finding the corresponding alarm time point in the fault data, and manually annotating the fault type based on this point. Since SCADA data is extremely large, manual annotation is not only a huge workload, but also has low accuracy. In particular, since the wind turbine parameter over-limit fault is a gradual process, the fault time recorded by SCADA is often relatively lagging (consequences have already occurred). If the alarm point is used as the dividing point between normal and abnormal to label the fault samples, it will directly affect the timeliness and reliability of the built fault diagnosis model when applied in real time. Therefore, it is necessary to find an earlier and more reasonable fault starting point from the continuously changing fault data segment and accurately annotate the data. To this end, the present disclosure proposes a method and device for automatic fault data annotation.

[0041] In order to make the purpose, technical solutions and advantages of the present disclosure more clear, specific embodiments will be described below in conjunction with the accompanying drawings.

[0042] refer to Figures 1 to 5 ,The automatic annotation method for fault data includes:

[0043] S101: Cleaning a first data set to obtain a second data set; the first data set is a historical operation data set of a target device; the first data set includes a fault data set to be labeled.

[0044] In this embodiment, the first data set is a historical operation data set of the target device, which can be obtained by the SCADA system of the target device. The target device can be a device that can generate huge data, such as a wind turbine, a power distribution cabinet, etc. There is fault data in the first data set, and the data within a period of time before and after the fault occurs and the data when the fault occurs are taken to form the data set to be labeled. For example, a fault occurs between 8:10-8:20, and the data between 8:05-8:25 is intercepted as part of the data set to be labeled.

[0045] First, the first data set is cleaned (initial cleaning) to obtain a relatively normal wind turbine operation data set. The initial cleaning can use the quartile method and the Ordering points to identify the clustering structure (OPTICS) clustering algorithm. Quartile is an analysis method in statistics. It sorts the data in order of size and divides the data set into four equal values, namely the minimum value, the first quartile, the median, and the third quartile. The quartile method can detect outliers in the data. The OPTICS clustering algorithm is a density-based clustering algorithm that is mainly used to find the local density maximum in a spatial data set and divide the data set into different clusters. The OPTICS clustering algorithm can effectively process various complex data sets and discover the clustering structure therein.

[0046] For example, Figure 3 This is a statistical chart of the data obtained after cleaning the original data of wind turbine A (525,781 groups of data) using the quartile method and the OPTICS clustering algorithm. Figure 3 The horizontal axis is wind speed (m / s) and the vertical axis is wind turbine generator active power (kW).

[0047] S102: Determine key characteristic parameters according to the fault data set to be labeled, and perform secondary cleaning on the second data set based on the key characteristic parameters to obtain a third data set.

[0048] In this embodiment, the first data set includes fault data and normal data, and the data when the target device fails and before and after the failure are taken to form the fault data set to be labeled.

[0049] Determine the key characteristic parameters based on the fault data set to be marked, and segment the second data set according to the mean of the key characteristic parameters to obtain a characteristic data set A1 above the mean and a characteristic data set B1 below the mean. Perform secondary cleaning on the characteristic data set A1 or the characteristic data set B1 based on a machine learning algorithm to obtain a third data set containing the normal operation of the wind turbine under all working conditions. The machine learning algorithm may be a k-means algorithm, a spectral clustering algorithm, an isolation forest algorithm, or a support vector machine. In this embodiment, an isolation forest algorithm may be used to perform secondary cleaning on the characteristic data set A1 or the characteristic data set B1. The isolation forest algorithm is mainly used to identify abnormal points in a data set, and it can realize local detection.

[0050] For example, the feature data set A1 is cleaned twice using the isolation forest algorithm to obtain the cleaned data set A2, and the cleaned data set A2 is merged with the feature data set B1 to finally obtain the data set A3 containing the normal operation of the wind turbines under all working conditions after the second cleaning. The feature data set B1 is cleaned using the same method to obtain the cleaned data set B2, and the cleaned data set B2 is merged with the feature data set A1 to finally obtain the data set B3 containing the normal operation of the wind turbines under all working conditions after the second cleaning.

[0051] For example, refer to Figure 4 , the fault type is a fault below the lower limit, and the key characteristic parameter is "gearbox oil circuit inlet oil pressure". According to this parameter, the mean value is calculated, and the mean value of this parameter in the second data set is 2.152 bar. According to this parameter, the second data set is divided into a data set A11 above the characteristic mean and a data set B11 below the characteristic mean. The data set below the characteristic mean is cleaned twice by applying the local isolation forest to obtain the cleaned data set B12. The data set B12 is merged with the data set A11 to obtain the data set B13 that can be used for over-limit threshold calculation.

[0052] S103: Determine, according to the fault type, an over-limit threshold corresponding to the fault type based on the third data set.

[0053] In this embodiment, the fault types include multiple types, such as high temperature, low oil level, etc. Based on these real fault conditions, the faults are divided into two types, one is a fault type below a lower limit, and the other is a fault type above an upper limit.

[0054] According to the above-mentioned fault types, for fault types below the lower limit, based on the characteristic data set B3, the fault over-limit threshold can be calculated using statistical principles; for faults above the upper limit, based on the characteristic data set A3, the fault over-limit threshold can be calculated using statistical principles, thereby obtaining the over-limit threshold of the third data set.

[0055] S104: annotating the fault data set based on the over-limit threshold to obtain a fault annotated data set.

[0056] In this embodiment, a first labeled data set is labeled for the first time based on the exceeding threshold value to obtain a first labeled data set;

[0057] The first labeled data set is labeled for the second time to obtain a fault labeled data set.

[0058] In this embodiment, the fault data set to be labeled is labeled for the first time according to the over-limit threshold. For fault types below the lower limit, when the key characteristic parameter corresponding to the fault is lower than the lower limit threshold, it is determined that a fault below the lower limit occurs. For fault types above the upper limit, when the key characteristic parameter corresponding to the fault is higher than the upper limit threshold, it is determined that a fault above the upper limit occurs. According to the above calculation, the data set after the initial labeling is obtained.

[0059] The Gaussian distribution principle formula used to calculate the over-limit threshold is:

[0060]

[0061] Among them, σ is the scale parameter and μ is the location parameter.

[0062] In this embodiment, because the first labeling is based on the fault threshold, when the fault is in the hidden period and growth period, the labeled data may fluctuate from fault to normal to fault, so it is necessary to label the data set after the first labeling for the second time to obtain data consistent with the actual situation. The data set after the first labeling is labeled for the second time to obtain the fault labeling data set.

[0063] The present disclosure realizes the augmentation of a data set after a primary cleaning by constructing fusion features, and further realizes secondary cleaning of data and calculation of an over-limit threshold value based on the augmented data set; and annotates the fault data set according to the over-limit threshold value to obtain a fault annotated data set.

[0064] From the above, it can be concluded that, on the one hand, the automatic fault data labeling method proposed in the present disclosure is different from the existing idea of ​​data cleaning based only on the wind speed-power curve in the data cleaning stage. The present disclosure performs a secondary deep cleaning of the data based on the machine learning algorithm (isolation forest algorithm) for the key feature parameters corresponding to the fault data set to be labeled, further improving the data quality and obtaining a more accurate threshold.

[0065] On the other hand, the present disclosure can realize automatic labeling of fault data, greatly reducing manual participation, improving data processing efficiency, and reducing costs. At the same time, due to the improvement of the accuracy and reliability of the labeling, the fault analysis and prediction results based on these labeled data are also more accurate and reliable.

[0066] In one embodiment of the present disclosure, the first annotated data set includes a plurality of continuous first annotated data;

[0067] The first labeled data set is labeled for the second time to obtain a fault labeled data set, including:

[0068] All first labeled data between the second labeled data and the third labeled data are labeled as faulty labeled data sets, the second labeled data is the Nth first labeled data in the first labeled data set that exceeds the first threshold, the third labeled data is the Mth first labeled data in the first labeled data set that exceeds the first threshold, and N is less than M.

[0069] In this embodiment, the first annotated data set includes multiple continuous first annotated data, such as continuous operation data of a wind turbine over a period of time. The second annotated data is the Nth first annotated data in the first annotated data set that exceeds the first threshold, and the third annotated data is the Mth first annotated data in the first annotated data set that exceeds the first threshold, where N is less than M, and both N and M are positive integers. Because both the second annotated data and the third annotated data exceed the first threshold, both the second annotated data and the third annotated data are fault data, and on this basis, the first annotated data between the second annotated data and the third annotated data can be marked as fault annotated data.

[0070] For example, the first labeled data set is {a1, a2, a3, a4, a5, a6, a7, a8}. If a1, a2, a3, a4, a5, a6, a8 all exceed the first threshold, then N can be 2 and M can be 6, indicating that a2, a3, a4, a5, a6 are labeled as faulty data; N can be 3 and M can be 6, indicating that a3, a4, a5, a6 are labeled as faulty data.

[0071] For example, refer to Figure 5 , the key characteristic parameter is "gearbox oil circuit inlet oil pressure", and the over-limit threshold of the calculated data set B3 is 1.14 bar. According to this value, the first annotation of the fault data set to be annotated is completed, and according to the secondary annotation method mentioned in the present disclosure, the fault annotation data set is further processed to obtain.

[0072] By checking the fault manual, we get the alarm value of 0.8 bar, which can be used to label the fault data set. Comparing the labeling results of the two methods, the model output 0 indicates normal, and the output 1 indicates that the gearbox inlet oil pressure is below the lower limit.

[0073] Comparing the two labeling results, it can be seen that since the fault is in the hidden period and growth period, the parameter changes are unstable, and there are large pressure fluctuations in the minute-level data of SCADA. If the labeling is performed directly according to the fault manual, the "normal-fault" label will be repeated many times. According to the method of this embodiment, this situation will not occur because the second labeling is set, the stability of the labeling is maintained, which is conducive to the later training of the model.

[0074] In one embodiment of the present disclosure, the fault type includes a fault type below a lower limit and a fault type above an upper limit, and data in the fault type below a lower limit is different from data in the fault type above an upper limit;

[0075] Determining an over-limit threshold corresponding to the fault type based on the third data set includes:

[0076] A first over-limit threshold corresponding to a fault type lower than a lower limit is determined based on the third data set, and a second over-limit threshold corresponding to a fault type higher than an upper limit is determined based on the third data set.

[0077] In this embodiment, the fault type includes a fault type below a lower limit and a fault type above an upper limit. The fault type below the lower limit means that if there is data below a certain threshold in a certain data set, the data is determined to be faulty data. The fault type above the upper limit means that if there is data above a certain threshold in a certain data set, the data is determined to be faulty data. Based on the two fault types and the third data set, the over-limit threshold can be determined, and the faulty data set to be marked can be marked based on the over-limit threshold.

[0078] It can be concluded from the above that this method can significantly improve the accuracy of fault detection, so that the system can respond to different types of faults more quickly. This embodiment can optimize resource allocation, ensure timely handling of faults, reduce potential risks, and thus improve the stability and reliability of the entire system.

[0079] In one embodiment of the present disclosure, determining a first over-limit threshold based on a fault type below a lower limit includes:

[0080] Determine a first over-limit threshold based on the below-lower-limit fault type and a first formula;

[0081] The first formula includes:

[0082]

[0083] in, represents the first over-limit threshold, and s are the mean and standard deviation of the key characteristic parameter C when the target equipment is operating normally, The confidence level is The degrees of freedom are t distribution, n is the number of samples of the target device in normal operation state.

[0084] In this embodiment, the first over-limit threshold indicates being lower than the lower limit threshold.

[0085] Determining a second over-limit threshold based on a fault type that is higher than an upper limit includes:

[0086] Determining a second over-limit threshold based on the higher-than-upper-limit fault type and a second formula;

[0087] The second formula includes:

[0088]

[0089] in, Indicates the second over-limit threshold, and s are the mean and standard deviation of the key characteristic parameter C when the target equipment is operating normally, The confidence level is The degrees of freedom are t distribution, n is the number of samples of the target device in normal operation state.

[0090] In this embodiment, the second over-limit threshold indicates being higher than the upper limit threshold.

[0091] In one embodiment of the present disclosure, cleaning the first data set once to obtain the second data set includes:

[0092] Cleaning the first data set once based on the one-cleaning model to obtain a second data set;

[0093] The fault dataset to be labeled is labeled based on the exceeding threshold to obtain a fault labeling dataset, including:

[0094] Based on the exceeding threshold, the labeling model is used to label the fault dataset to be labeled to obtain the fault labeling dataset.

[0095] In this embodiment, the first data set is cleaned once based on the one-cleaning model to obtain the second data set; the second data set is cleaned twice based on the two-cleaning model to obtain the third data set. The parameters of the one-cleaning model include three quartile parameters, namely, upper and lower limit weights Para, row block step Step, and number of blocks per row row, and three OPTICS clustering parameters, namely, the minimum number of samples Min_samples, threshold parameter Xi, and minimum cluster size min_cluster_size. The parameters of the two-cleaning model include three parameters, namely, the number of decision trees n_estimators, the number of abnormal samples contamination, and the decision tree sample ratio max_samples.

[0096] For the primary cleaning model and the secondary cleaning model, a primary cleaning model with J SCADA key characteristic parameters (u1, u2, ..., uJ) and 2 output parameters (wind speed, active power) is established respectively; a secondary cleaning model with K SCADA key characteristic parameters (u1, u2, ..., uK) and 3 output parameters (wind speed, active power, key characteristics) is established. The SCADA data of the wind turbine is first cleaned by the primary cleaning model, and then combined with the key characteristics of the fault type, the secondary cleaning model is used to complete the secondary cleaning to obtain the third data set after cleaning. The over-limit threshold corresponding to the fault type is determined according to the third data set.

[0097] In this embodiment, based on the over-limit threshold, the fault data set to be labeled is labeled using the labeling model to obtain the fault labeling data set, including:

[0098] The labeling model includes labeling model 1 and labeling model 2. Based on the exceeding threshold, the labeling model 1 is used to label the fault data set to be labeled for the first time to obtain the first labeling data set; based on the abnormal situation of exceeding the threshold, the labeling model 2 is used to label the first labeling data set for the second time to obtain the fault labeling data set.

[0099] For example, first apply labeling model 1. Based on the over-limit threshold, if the fault type to be labeled is an over-limit fault, all data above the upper threshold are labeled as fault data, and the rest of the data are labeled as normal data. If the fault type to be labeled is a fault below the lower limit, all data below the lower threshold are labeled as fault data, and the rest of the data are labeled as normal data. The first labeled data set is obtained based on labeling model 1.

[0100] Applying labeling model 2, based on the data set after the first labeling, the counter counts the data exceeding the threshold value. When the threshold value is exceeded three times, the data is considered abnormal, and the data from the third exceeding threshold value to the fault shutdown is labeled as fault data. Finally, a labeled fault labeling data set is obtained for training the fault diagnosis model.

[0101] In one embodiment of the present disclosure, reference Figure 2 First, the wind turbine SCADA data is classified to obtain the full-year data set and the fault data set to be labeled. The full-year data set is cleaned once according to the quartile method and the OPTICS method (primary cleaning), and then the isolated forest is used to perform a secondary cleaning on the cleaned data in combination with the fault-related key features corresponding to the fault data set to be labeled, and finally a normal data set of wind turbines with fault-free operation is obtained.

[0102] According to the fault type, the over-limit threshold (lower than the lower threshold or higher than the upper threshold) is calculated based on the third data set by applying statistical principles. The fault data set to be labeled is labeled based on the over-limit threshold to obtain a fault labeling data set. For a type of non-over-limit fault, the data set is augmented by constructing fusion features, and then the above method is used to label the fault data set, which expands the scope of application of the method. The present disclosure fully considers the various influencing factors and the characteristics of the fault itself in the SCADA data preprocessing process of the wind turbine, which can effectively improve the accuracy of the annotation of the wind turbine over-limit fault data set, reduce the workload of data processing staff, and improve the timeliness and accuracy of wind turbine fault diagnosis.

[0103] In one embodiment of the present disclosure, reference Figures 6 to 8 , the characteristic parameter is "gearbox oil pool temperature is higher than the upper limit", and the first data set is the original data of wind turbine B. Firstly, the original data of wind turbine B is initially cleaned by using the quartile method and OPTICS clustering method according to the wind speed-power curve to obtain the second data set. Figure 6 The horizontal axis is wind speed (m / s) and the vertical axis is wind turbine generator active power (kW).

[0104] This fault is a fault above the upper limit. The key characteristic parameter is: "gearbox oil pool temperature". According to this parameter, the mean value is calculated, and the mean value of this feature in the second data set is 56.664℃. The second data set is divided into data set A21 above the feature mean and data set B21 below the feature mean. The data set A21 above the feature mean is cleaned twice using the local isolation forest to obtain data set A22. Data set A22 and data set B21 are merged to obtain data set A23 that can be used for threshold calculation.

[0105] The cleaned data set A23 is used to calculate the over-limit threshold. Since "gearbox oil pool temperature is higher than the upper limit" is a fault higher than the upper limit, the threshold calculation formula higher than the upper limit is used for threshold calculation, and α is taken as 0.005 and n is 367042 (the amount of normal data contained in A23 after secondary cleaning). The calculated threshold is 68.6℃. Based on this value, the initial labeling of the fault data set of wind turbine B is completed to obtain the first labeled data set; according to the secondary labeling method, further processing is performed to obtain the fault labeled data set. Data comparison reference before and after cleaning Figure 7 .

[0106] By checking the fault manual, we get the alarm value of 75°C. Based on this alarm value, we also complete the annotation of the fault data set of wind turbine B. Figure 8 As shown in the figure, the model output 0 is normal, and output 2 indicates that the gearbox oil pool temperature is higher than the upper limit fault.

[0107] It can be seen that the threshold calculated by this embodiment is lower than the threshold in the fault manual, and the fault time node marked according to this embodiment is much earlier than that marked according to the fault manual. The reason for this phenomenon is that the temperature of the gearbox oil pool changes relatively slowly, and the temperature gradient process takes a long time. The thresholds obtained by the two methods are both between the normal state and the fault state, but the threshold calculated by this embodiment is lower, which can achieve fault marking earlier under the premise of ensuring that the unit does not have a normal state misdiagnosis.

[0108] In one embodiment of the present disclosure, reference Figures 9 to 14 There is a type of fault in rotating machinery, where the recorded information of the fault is not directly below the lower limit or above the upper limit. This type of fault often cannot be directly labeled by the first labeling method, which is a more difficult problem in fault data preprocessing. To this end, the present disclosure further proposes a feature fusion data augmentation processing method, which converts this type of fault into an over-limit fault, and completes the fault labeling.

[0109] For example, when a wind turbine's pitch control system fails, faults such as "a pitch control motor is stalled" and "the angle difference between the two blades is 2 degrees repeatedly" occur frequently. When this type of fault occurs, the SCADA data parameters are mainly manifested as a continuous deviation between the actual pitch angle and the given pitch angle, while the parameters such as the pitch angle themselves are often still in the normal operating parameter range. This type of fault usually does not give specific parameter thresholds in the fault manual like ordinary over-limit faults. In order to solve the problem of labeling this type of fault, the key features are fused to achieve the augmentation of the original SCADA data. The fused feature column is then used to transform the fault into an over-limit fault, and the fault data is labeled using the labeling method disclosed in the present invention.

[0110] In view of the above fault problems of the variable pitch system, the difference method can be used to obtain the following two fusion features:

[0111]

[0112]

[0113] Among them, p1 is the angle of propeller blade No. 1, p s1 p is the angle of propeller blade No. 1, p2 is the angle of propeller blade No. 2, p s2 p is the angle of propeller blade No. 2, p3 is the angle of propeller blade No. 3, p s3 Set the angle for propeller blade number 3.

[0114] The above faults can be transformed into over-upper limit faults by using the fused feature P or feature Q, and the over-upper limit formula is used to calculate the threshold value, thereby completing the automatic fault labeling. For example, the fault data of "a certain variable pitch motor is blocked" is obtained when the wind turbine C fails. First, the original data of wind turbine C is cleaned according to the quartile method and OPTICS clustering method to obtain the cleaned second data set.

[0115] For the second data set, the fusion feature P calculation formula is used to obtain the value of the fusion feature P, and the augmented data set is constructed. The mean of the calculated feature P is 0.147°. According to this parameter, the augmented data set is divided into data sets A31 above the feature mean and data sets B31 below the feature mean. The data set A31 above the feature mean is cleaned by applying the local isolation forest to obtain the cleaned data set A32, which is then merged with the data set B31 to obtain the data set A33 for threshold calculation.

[0116] The cleaned dataset A33 is used to obtain the annotation threshold. Since the fusion feature is used to convert the "a certain variable pitch motor stall" fault into a fault above the upper limit, the upper limit threshold calculation formula is used to calculate the threshold, and α is taken as 0.005 and n is 88673 (the normal data volume contained in the dataset A33 after the second cleaning). The calculated threshold is 1.126°. According to the above method, the fault dataset of wind turbine C is annotated for the first time and the second time according to the fusion feature P. Fig.11 In the figure, the model output value 0 indicates normal, and the output 3 indicates that the fault of "a certain variable pitch motor is blocked" occurs. As can be seen from the figure, the fusion feature and automatic labeling model can be used to effectively label such fault data.

[0117] In one embodiment of the present disclosure, reference Figure 12 to Figure 14 ,For a wind turbine E in a wind farm, the fault data set “the angle difference between the two blades of the variable pitch is 2 degrees repeatedly occurred” is annotated, and the 1-638 and 639-1538 groups of data correspond to the minute-level data of the two faults.

[0118] First, we cleaned 527,001 sets of original data of wind turbine E according to the wind speed-power curve combined with the quartile method and the OPTICS clustering method to obtain the cleaned second data set. The wind speed-power curves before and after cleaning are shown in Figure 2. Fig.12 shown.

[0119] For the second data set, the fusion feature Q calculation formula is used to obtain the value of the fusion feature Q, and the augmented data set is constructed. The mean of the fusion feature Q is calculated to be 0.021°. According to this parameter, the second data set is divided into data sets A41 with values ​​higher than the feature mean and data sets B41 with values ​​lower than the feature mean. The data set A41 with values ​​higher than the feature mean is cleaned by applying the local isolation forest to obtain the cleaned data set A42, which is then merged with the data set B41 to obtain the data set A43 for threshold calculation. The data comparison before and after cleaning is as follows: Fig.13 shown.

[0120] The cleaned data is used to obtain the threshold. Since the fusion feature Q is used to convert the "two blades angle difference of 2 degrees repeatedly" fault into a fault above the upper limit, the upper limit threshold calculation formula is used to calculate the threshold. Taking α as 0.005 and n as 86012 (the normal data volume contained in the data set A43 after the second cleaning), the threshold is calculated to be 0.075°. The fault data set is annotated for the first time and the second time according to the above method combined with the threshold of the fusion feature Q. The results are as follows Fig.14 . Fig.14 In the figure, the model outputs 0 for normal, and outputs 4 for the fault "the angle difference between the two blades is 2 degrees and occurs repeatedly". It can be seen that the fusion feature and automatic labeling model can be used to effectively label this type of fault data.

[0121] From the above, it can be concluded that the present disclosure uses the idea of ​​feature fusion to transform a type of non-over-limit fault into an over-limit fault by constructing new fusion feature augmented data, thereby realizing fault labeling. The present disclosure expands the fault types applicable to the automatic labeling model and solves the problem that when such faults occur, the SCADA parameters are still in the normal operating range, making it difficult to label the faults.

[0122] Corresponding to the automatic fault data labeling method of the above embodiment, Fig.15 This is a structural block diagram of a fault data automatic labeling device provided by an embodiment of the present disclosure. For ease of explanation, only the parts related to the embodiment of the present disclosure are shown. Fig.15 The fault data automatic labeling device 20 includes: a first data cleaning unit 21, a second data cleaning unit 22, a fault threshold determination unit 23 and a data labeling unit 24.

[0123] The first data cleaning unit 21 is used to clean the first data set once to obtain the second data set; the first data set is a historical operation data set of the target device; the first data set includes a fault data set to be marked;

[0124] A second data cleaning unit 22, configured to determine key characteristic parameters according to the fault data set to be labeled, and perform secondary cleaning on the second data set based on the key characteristic parameters to obtain a third data set;

[0125] A fault threshold determination unit 23, configured to determine, according to the fault type, an over-limit threshold corresponding to the fault type based on the third data set;

[0126] The data labeling unit 24 is used to label the fault data set to be labeled based on the over-limit threshold value to obtain a fault labeling data set.

[0127] In one embodiment of the present disclosure, the data annotation unit 24 is specifically used to:

[0128] Based on the exceeding threshold, the to-be-annotated fault data set is annotated for the first time to obtain a first annotated data set;

[0129] The first labeled data set is labeled for the second time to obtain a fault labeled data set.

[0130] In one embodiment of the present disclosure, the data annotation unit 24 is specifically used to:

[0131] The first annotated data set includes a plurality of continuous first annotated data;

[0132] All first labeled data between the second labeled data and the third labeled data are labeled as faulty labeled data sets, the second labeled data is the Nth first labeled data in the first labeled data set that exceeds the first threshold, the third labeled data is the Mth first labeled data in the first labeled data set that exceeds the first threshold, and N is less than M.

[0133] In one embodiment of the present disclosure, the fault type includes a fault type below a lower limit and a fault type above an upper limit, and data in the fault type below a lower limit is different from data in the fault type above an upper limit;

[0134] The fault threshold unit 23 is specifically used for:

[0135] A first exceeding-limit threshold is determined based on a fault type that is below a lower limit, and a second exceeding-limit threshold is determined based on a fault type that is above an upper limit.

[0136] In one embodiment of the present disclosure, the fault threshold determination unit 23 is specifically configured to:

[0137] Determine a first over-limit threshold based on the below-lower-limit fault type and a first formula;

[0138] The first formula includes:

[0139]

[0140] in, represents the first over-limit threshold, and s are the mean and standard deviation of the key characteristic parameter C when the target equipment is operating normally, The confidence level is The degrees of freedom are t distribution, n is the number of samples of the target device in normal operation state.

[0141] In one embodiment of the present disclosure, the fault threshold determination unit 23 is specifically configured to:

[0142] Determining a second over-limit threshold based on the higher-than-upper-limit fault type and a second formula;

[0143]

[0144] in, Indicates the second over-limit threshold, and s are the mean and standard deviation of the key characteristic parameter C when the target equipment is operating normally, The confidence level is The degrees of freedom are t distribution, n is the number of samples of the target device in normal operation state.

[0145] In an embodiment of the present disclosure, the first data cleaning unit 21 is specifically used for:

[0146] Cleaning the first data set once based on the one-cleaning model to obtain a second data set;

[0147] The data annotation unit 24 is specifically used for:

[0148] Based on the exceeding threshold, the labeling model is used to label the fault dataset to be labeled to obtain the fault labeling dataset.

[0149] See also Fig.16 , Fig.16 A schematic block diagram of an electronic device provided by an embodiment of the present disclosure. Fig.16 The electronic device 300 in the embodiment shown may include: one or more processors 301, one or more input devices 302, one or more output devices 303 and one or more memories 304. The processors 301, input devices 302, output devices 303 and memories 304 communicate with each other via a communication bus 305. The memory 304 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. The processor 301 is configured to call the program instructions to execute the functions of each module / unit in the above-mentioned device embodiments, such as Fig.15 The functions of units 21 to 24 are shown.

[0150] It should be understood that in the embodiment of the present disclosure, the processor 301 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0151] The input device 302 may include a touch panel, a fingerprint sensor (for collecting fingerprint information of the user and fingerprint direction information), a microphone, etc. The output device 303 may include a display (LCD, etc.), a speaker, etc.

[0152] The memory 304 may include a read-only memory and a random access memory, and provide instructions and data to the processor 301. A portion of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store information about the device type.

[0153] In a specific implementation, the processor 301, input device 302, and output device 303 described in the embodiments of the present disclosure can execute the implementation methods described in the first and second embodiments of the method for automatic labeling of fault data provided in the embodiments of the present disclosure, and can also execute the implementation methods of the electronic device described in the embodiments of the present disclosure, which will not be repeated here.

[0154] In another embodiment of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by the processor, all or part of the processes in the above-mentioned embodiment method are implemented, and the computer program can also be completed by instructing the relevant hardware through the computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, the steps of each of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0155] The computer-readable storage medium may be an internal storage unit of the electronic device of any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0156] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.

[0157] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the electronic devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0158] In the several embodiments provided in the present application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces or units, or it can be an electrical, mechanical or other form of connection.

[0159] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present disclosure.

[0160] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0161] The above are only specific embodiments of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present disclosure, and these modifications or replacements should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A method for automatic fault data labeling, characterized in that: include: The first data set is cleaned once to obtain a second data set; the first data set is a historical operation data set of the target device; the first data set includes a fault data set to be annotated; Determine key characteristic parameters according to the fault data set to be annotated, and perform secondary cleaning on the second data set based on the key characteristic parameters to obtain a third data set; According to the fault type, determining an over-limit threshold corresponding to the fault type based on the third data set; The fault type includes a fault type below a lower limit and a fault type above an upper limit, and the data in the fault type below a lower limit is different from the data in the fault type above an upper limit; The determining, based on the third data set, an over-limit threshold corresponding to the fault type includes: Determine a first over-limit threshold corresponding to the fault type below a lower limit based on the third data set, and determine a second over-limit threshold corresponding to the fault type above an upper limit based on the third data set; Determine a first over-limit threshold based on the below-lower-limit fault type and a first formula; The first formula includes: in, represents the first over-limit threshold, and s are the mean and standard deviation of the key characteristic parameter C when the target equipment is operating normally, The confidence level is The degrees of freedom are t distribution, n is the number of samples of the target device in normal operation state; The to-be-annotated fault data set is annotated based on the over-limit threshold to obtain a fault annotated data set.

2. The method for automatic fault data labeling according to claim 1, characterized in that: The step of labeling the to-be-labeled fault data set based on the over-limit threshold to obtain a fault labeling data set includes: Performing a first labeling on the to-be-labeled fault data set based on the over-limit threshold to obtain a first labeled data set; The first labeled data set is labeled for a second time to obtain a fault labeled data set.

3. The automatic fault data labeling method according to claim 2, characterized in that: The first annotated data set includes a plurality of continuous first annotated data; Performing a second annotation on the first annotated data set to obtain a fault annotated data set includes: All first labeled data between the second labeled data and the third labeled data are labeled as faulty labeled data sets, where the second labeled data is the Nth first labeled data in the first labeled data set that exceeds the first threshold, and the third labeled data is the Mth first labeled data in the first labeled data set that exceeds the first threshold, where N is less than M.

4. The method for automatic fault data labeling according to claim 1, characterized in that: The determining, based on the third data set, a second over-limit threshold corresponding to the fault type that is higher than the upper limit, comprises: Determining a second over-limit threshold based on the higher-than-upper-limit fault type and a second formula; in, Indicates the second over-limit threshold, and s are the mean and standard deviation of the key characteristic parameter C when the target equipment is operating normally, The confidence level is The degrees of freedom are t distribution, n is the number of samples of the target device in normal operation state.

5. The method for automatic fault data labeling according to claim 1, characterized in that: The step of cleaning the first data set to obtain the second data set includes: Cleaning the first data set once based on a one-cleaning model to obtain a second data set; The step of labeling the to-be-labeled fault data set based on the over-limit threshold to obtain a fault labeling data set includes: Based on the over-limit threshold, the fault data set to be labeled is labeled using a labeling model to obtain a fault labeling data set.

6. A device for automatically marking fault data, characterized in that: include: A first data cleaning unit, used for cleaning the first data set once to obtain a second data set; The first data set is a historical operation data set of the target device; the first data set includes a fault data set to be annotated; A second data cleaning unit, configured to determine key characteristic parameters according to the fault data set to be labeled, and perform secondary cleaning on the second data set based on the key characteristic parameters to obtain a third data set; a fault threshold determination unit, configured to determine, according to the fault type, an over-limit threshold corresponding to the fault type based on the third data set; The fault type includes a fault type below a lower limit and a fault type above an upper limit, and the data in the fault type below a lower limit is different from the data in the fault type above an upper limit; The determining, based on the third data set, an over-limit threshold corresponding to the fault type includes: Determine a first over-limit threshold corresponding to the fault type below a lower limit based on the third data set, and determine a second over-limit threshold corresponding to the fault type above an upper limit based on the third data set; Determine a first over-limit threshold based on the below-lower-limit fault type and a first formula; The first formula includes: in, represents the first over-limit threshold, and s are the mean and standard deviation of the key characteristic parameter C when the target equipment is operating normally, The confidence level is The degrees of freedom are t distribution, n is the number of samples of the target device in normal operation state; The data labeling unit is used to label the to-be-labeled fault data set based on the over-limit threshold to obtain a fault labeling data set.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images