Method and device for monitoring and early warning of abnormal temperature of wind turbine generator bearing

Abstract

The application relates to the field of wind turbine fault diagnosis, and discloses a wind turbine generator bearing temperature abnormality monitoring and early warning method and device. The method comprises the following steps: acquiring and preprocessing online SCADA data of a wind turbine generator; extracting bearing temperature related variables, inputting a preset model to obtain a temperature theoretical value; extracting a temperature actual value, calculating a temperature difference between the temperature actual value and the temperature theoretical value, and judging abnormality when the temperature difference exceeds a preset value; and diagnosing and issuing an early warning based on operation data within a certain time. The application improves the efficiency and accuracy of wind turbine fault diagnosis, and realizes accurate monitoring and early warning of wind turbine generator bearing temperature abnormality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind turbine fault diagnosis, in particular to a wind turbine generator bearing temperature anomaly monitoring and early warning method and device. BACKGROUND

[0002] With the rapid development of renewable energy, wind power as a clean energy, its proportion in the global energy structure increases year by year. The core components of wind power equipment include wind wheel, blade, tower, base, transmission system and control system, etc. With the rapid development of wind power industry, the single machine capacity is increasing, and the stability and reliability of wind turbine become particularly important. The failure of wind turbine not only increases the maintenance cost, but also has a negative impact on the healthy development of wind power industry. During the operation of wind turbine, generator bearing temperature anomaly is one of the common faults, if not found and handled in time, it may cause system failure, reduce wind energy utilization rate, and even cause significant economic loss. Therefore, accurately identifying generator bearing temperature anomaly and realizing early warning of fault has important significance for reducing safety risk and power loss.

[0003] Similar prior art is Chinese patent application No. CN112577739A, which discloses a wind turbine engine drive end bearing over-temperature fault diagnosis and early warning method. Based on multivariate regression analysis method, the nonlinear causal relationship between bearing temperature and related variables is established, and the reference curve of engine drive end bearing temperature under wind turbine health state is obtained. The deviation degree of real-time regression curve of bearing temperature relative to health reference curve is quantified to obtain a health index reflecting potential over-temperature fault of the bearing. Based on the obtained health index reflecting potential over-temperature fault of the bearing, the bearing over-temperature fault is monitored and warned by setting threshold. This method depends on the accuracy of the reference curve, if the reference curve is not accurate, it will directly affect the reliability of the health index, and its quantification method does not consider the abnormal values or noises in the data, when the bearing temperature fluctuates abnormally, the accuracy of the health index is affected. Chinese patent application No. CN112598172A discloses a wind turbine bearing temperature early warning method, acquires wind turbine SCADA temperature data set; preprocesses the wind turbine SCADA temperature data set, and selects model input variables; divides the data set, and establishes a wind turbine bearing temperature prediction model based on GS-LightGBM; predicts wind turbine bearing operation data according to the prediction model, and analyzes the residual error, calculates the EWMA control chart parameters, obtains the alarm threshold, and analyzes the working state. This method only performs abnormal early warning according to temperature data, lacks comprehensive consideration of abnormal factors of bearing temperature, and is not comprehensive

[0004] Therefore, the application provides an abnormal monitoring and early warning method and device for generator bearing temperature of a wind turbine to improve the effectiveness and accuracy of abnormal monitoring of the wind turbine. SUMMARY

[0005] The application provides an abnormal monitoring and early warning method and device for generator bearing temperature of a wind turbine to improve the efficiency and accuracy of abnormal monitoring and early warning of the generator bearing temperature of the wind turbine.

[0006] In a first aspect, the application provides an abnormal monitoring and early warning method for generator bearing temperature of a wind turbine, which comprises the following steps:

[0007] Step 1: obtaining online SCADA data of a generator of the wind turbine and pre-processing the online SCADA data;

[0008] Step 2: extracting a bearing temperature related variable from the pre-processed online SCADA data, inputting the related variable into a preset model, and obtaining a bearing temperature theoretical value;

[0009] Step 3: extracting an actual bearing temperature value from the pre-processed online SCADA data, calculating a temperature difference between the actual bearing temperature value and the bearing temperature theoretical value, and determining that there is an abnormality when the absolute value of the temperature difference is greater than a first preset value;

[0010] Step 4: obtaining first generator operation data within a first preset time, diagnosing the generator bearing based on the first generator operation data, and issuing a warning information.

[0011] In combination with the first aspect, in a first implementation manner of the first aspect of the application, the pre-processing of the online SCADA data comprises the following steps:

[0012] Step 11: determining whether there is missing data or abnormal data in the online SCADA data, and if so, defining a variable corresponding to the missing data or abnormal data as a first variable;

[0013] Step 12: extracting any first variable, obtaining a theoretical collection number of any first variable within a second preset time, and an invalid number of data corresponding to any first variable being missing data or abnormal data, calculating a ratio of the invalid number to the theoretical collection number, and defining the ratio as an invalid rate;

[0014] Step 13: after traversing all first variables, dividing all first variables into a preset number of variable groups according to a preset rule based on the invalid rate;

[0015] Step 14, setting a corresponding data generation strategy for each variable group, extracting any variable group, generating a replacement value for each first variable in any variable group using the data generation strategy corresponding to any variable group, and completing the online SCADA data using the replacement value.

[0016] In combination with the first aspect, in a second implementation manner of the first aspect of the application, the preset number is 3, the first variable with the invalidity less than the second preset value is divided into the first variable group, the first variable with the invalidity greater than or equal to the second preset value and less than or equal to the third preset value is divided into the second variable group, and the first variable with the invalidity greater than the third preset value is divided into the third variable group, and step 14 includes:

[0017] For the first variable group, the variable in the first variable group is defined as a second variable, a data sequence of any second variable within a third preset time is obtained, an estimated value of any second variable at each time point is calculated based on Formula One, and the estimated value of any second variable at T-1 moment is taken as the replacement value of any second variable at T moment, and Formula One is:

[0018]

[0019] Wherein, E(t) is the estimated value at t moment, a is a parameter weight coefficient, D(t) is the actual value at t moment, and T is the current moment;

[0020] For the third variable group, the variable in the third variable group is defined as a third variable, and the data obtained at T-1 moment of any third variable is taken as the replacement value of any third variable at T moment.

[0021] For the second variable group, the variable in the second variable group is defined as a fourth variable, a first replacement value of any fourth variable at T moment is generated based on the data generation strategy corresponding to the third variable group, the first replacement value is taken as the tentative value of any fourth variable at T moment, then an estimated value of any fourth variable at T moment is generated based on the data generation strategy corresponding to the first variable group, and is defined as a second replacement value, and the average value of the first replacement value and the second replacement value is set as the replacement value of any first variable at T moment.

[0022] In combination with the first aspect, in a third implementation manner of the first aspect of the application, the establishment method of the preset model is:

[0023] Step 21, obtaining historical SCADA data of the wind turbine bearing, the historical SCADA data including N1 data sequences, each data sequence including second generator operation data, wherein N1 is a positive integer.

[0024] Step 22, traversing the historical SCADA data, when there is a missing data corresponding to any characteristic variable in the second generator operation data, defining any characteristic variable as a first characteristic variable, and dividing the historical SCADA data into a training data set and a validation data set according to a preset proportion;

[0025] Step 23, combining all the first characteristic variables in any manner to obtain N2 characteristic variable combinations, wherein N2 is a positive integer;

[0026] Step 24, extracting any characteristic variable combination, defining the variables in any characteristic variable combination as third characteristic variables, defining the variables outside any characteristic variable combination as second characteristic variables, deleting all data corresponding to any second characteristic variable in the training data set, and deleting data sequences lacking data corresponding to the third characteristic variables, to generate a first training data set, repeating step 24 until all characteristic variable combinations are traversed, generating N3 first training data sets, wherein N3 is a positive integer less than or equal to N2;

[0027] Step 25, performing the same operation on the validation data set as step 24 to generate N4 first validation data sets, wherein N4 is a positive integer less than or equal to N2;

[0028] Step 26, performing machine learning on any first training data set to obtain a first preset model corresponding to any first training data set, and generating N3 first preset models after traversing all first training data sets;

[0029] Step 27, verifying each first preset model using the first validation data set to obtain the accuracy of each first preset model, and taking the first preset model corresponding to the maximum accuracy as the preset model.

[0030] In combination with the first aspect, in a fourth implementation manner of the first aspect of the application, step 27 includes:

[0031] extracting any first preset model, and defining the characteristic variables used for training any first preset model as fourth characteristic variables;

[0032] extracting any first validation data set, and defining the characteristic variables contained in any first validation data set as fifth characteristic variables, when the fifth characteristic variables contain the fourth characteristic variables, defining any first validation data set as a second validation data set;

[0033] after traversing all first validation data sets, verifying any first preset model using each second validation data set to obtain a plurality of model accuracies, calculating a first statistical value of all model accuracies, and taking the first statistical value as the accuracy of any first preset model.

[0034] In a fifth implementation form of the first aspect, in the step 2, the bearing temperature related variables are extracted from the pre-processed online SCADA data, and the bearing temperature related variables include:

[0035] The feature variables used in the preset model training are defined as sixth feature variables, and the related variables are extracted from the first SCADA data based on the sixth feature variables.

[0036] In a sixth implementation form of the first aspect, the second generator operation data includes bearing temperature data, cabin temperature data, ambient temperature data, wind speed, wind direction, and power, lubrication state data, rotation speed, torque and power of each component of the bearing.

[0037] In a seventh implementation form of the first aspect, the step 4 includes:

[0038] The first generator operation data is divided into a plurality of variable data sequences based on variable types, the second statistical values of each variable data sequence are calculated respectively, and the first feature vector corresponding to the first generator operation data is generated based on the second statistical values;

[0039] The second feature vector corresponding to each fault event within the third preset time is extracted, and the Euclidean distance between any second feature vector and the first feature vector is calculated, wherein the second feature vector of any fault event is generated based on the historical generator operation data within the first preset time before a specific time, and the specific time is a time point that is a specific time away from the time point when any fault event occurs.

[0040] It is determined whether the minimum Euclidean distance is less than a fourth preset value, and if yes, it is determined that there is a fault risk.

[0041] In an eighth implementation form of the first aspect, after it is determined that there is a fault risk, the method further includes the following steps:

[0042] The statistical value in the second feature vector corresponding to the minimum Euclidean distance is defined as a third statistical value, and the sensitivity corresponding to any variable type is calculated based on Formula Two, and Formula Two is as follows:

[0043]

[0044] wherein G i is the sensitivity corresponding to the i-th variable type, ST i,2 is the second statistical value of the i-th variable type, ST i,3 is the third statistical value of the i-th variable type, and d is the minimum Euclidean distance.

[0045] When the absolute value of the sensitivity corresponding to any variable type is greater than a preset threshold, the any variable type is determined as an abnormal influencing factor.

[0046] After traversing all variable types, generate early warning information based on all abnormal influencing factors.

[0047] In a second aspect, the application provides an abnormal monitoring and early warning device for a wind turbine generator bearing temperature, comprising a preprocessing module, a temperature analysis module, an abnormality judgment module and an abnormality diagnosis module.

[0048] The preprocessing module is configured to acquire online SCADA data of the wind turbine generator and preprocess the online SCADA data.

[0049] The temperature analysis module is configured to extract bearing temperature related variables from the preprocessed online SCADA data, input the related variables into a preset model, and acquire a bearing temperature theoretical value.

[0050] The abnormality judgment module is configured to extract a bearing temperature actual value from the preprocessed online SCADA data, calculate a temperature difference between the bearing temperature actual value and the bearing temperature theoretical value, and determine that an abnormality exists when the absolute value of the temperature difference is greater than a first preset value.

[0051] The abnormality diagnosis module is configured to acquire first generator operation data within a first preset time, diagnose the generator bearing based on the first generator operation data, and issue an early warning information.

[0052] The technical scheme provided by the application has at least the following beneficial effects:

[0053] 1. After acquiring the online SCADA data, the missing or abnormal data is identified and processed, and through variable grouping and data generation strategies, these data can be effectively completed or replaced, reducing the impact of data missing on the analysis results; different data generation strategies are adopted for different invalid data, ensuring the rationality, accuracy and integrity of data completion, providing a reliable basis for subsequent analysis.

[0054] 2. The establishment method of the preset model considers the combination of multiple characteristic variables and is trained and verified through machine learning method, which can filter out the optimal model and improve the prediction accuracy.

[0055] 3. The fault diagnosis method based on the generator operation data comprehensively considers the influence of multiple variable types, accurately judges the fault risk through calculation of characteristic vectors and Euclidean distances, and the calculation of sensitivity can further identify abnormal influencing factors, providing strong support for fault reason finding and early warning information generation; warning can be issued at the early stage of the fault, so that maintenance personnel can take timely measures to reduce downtime and maintenance costs caused by the fault. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort based on these drawings.

[0057] Figure 1 An embodiment of the method for monitoring and early warning of abnormal temperature of a generator bearing of a wind turbine generator in the present application;

[0058] Figure 2 An embodiment of the device for monitoring and early warning of abnormal temperature of a generator bearing of a wind turbine generator in the present application. DETAILED DESCRIPTION

[0059] The embodiments of the present application provide a method and device for monitoring and early warning of abnormal temperature of a generator bearing of a wind turbine generator. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variation thereof is intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0060] For the convenience of understanding, the specific process of the embodiments of the present application will be described below. Please refer to Figure 1 An embodiment of the method for monitoring and early warning of abnormal temperature of a generator bearing of a wind turbine generator in the present application includes:

[0061] Step 1, obtaining online SCADA data of a wind turbine generator and pre-processing the online SCADA data.

[0062] Specifically, the online SCADA data includes but is not limited to data collection time, wind turbine generator identification, bearing temperature data, cabin temperature data, environmental temperature data, wind speed, wind direction, and power, lubrication state data, rotating speed, torque and power, bearing vibration frequency, generator shaft torque of each component of the bearing, etc.

[0063] In a specific embodiment, in step 1, the pre-processing of the online SCADA data includes:

[0064] Step 11, judging whether there is missing data or abnormal data in the online SCADA data, if so, defining the variable corresponding to the missing data or abnormal data as the first variable.

[0065] Step 12, extracting any first variable, obtaining the theoretical collection times of any first variable within the second preset time, and the invalid times of the data corresponding to any first variable being missing data or abnormal data, calculating the ratio of the invalid times to the theoretical collection times, and defining it as the invalid rate.

[0066] Step 13, after traversing all first variables, based on the invalid rate, all first variables are divided into a preset number of variable groups according to a preset rule.

[0067] Step 14, setting a corresponding data generation strategy for each variable group, extracting any variable group, using the data generation strategy corresponding to any variable group, generating a replacement value for each first variable in any variable group, and using the replacement value to complete the online SCADA data.

[0068] The second preset time and the preset number are set according to the experience of those skilled in the art or according to the actual application scene, and the embodiments of the present application are not limited thereto.

[0069] Specifically, the abnormal data described above is abnormal data caused by data acquisition equipment failure, transmission error, etc. For example, the temperature is 4000℃, the wind speed is -5m / s, these values obviously do not conform to the actual situation, and can be identified as noise or outliers. Further, the above abnormal data can also be identified based on statistical quantities such as standard deviation or interquartile range (IQR).

[0070] The invalid rate is an important indicator to measure the quality of data, which can help us understand the data integrity and reliability of each variable. By calculating the invalid rate, we can have a quantitative understanding of the quality of the data. The invalid rates of different variables may differ greatly, and dividing all first variables into a preset number of variable groups according to the invalid rate can be more targeted. For example, variables with low invalid rates may only need simple processing, while variables with high invalid rates may need more complex data generation strategies.

[0071] The abnormal data is replaced by the replacement value or the missing data is supplemented by the replacement value to complete the SCADA data.

[0072] In a specific embodiment, the preset number is 3, the first variable with an invalid rate less than a second preset value is divided into a first variable group, the first variable with an invalid rate greater than or equal to a second preset value and less than or equal to a third preset value is divided into a second variable group, and the first variable with an invalid rate greater than a third preset value is divided into a third variable group, and step 14 comprises:

[0073] (1) For the first variable group, define the variables in the first variable group as the second variables, obtain the data sequence of any second variable within the third preset time, calculate the estimated value of any second variable at each time point based on Formula One, and take the estimated value of any second variable at T-1 time as the replacement value of any second variable at T time, and Formula One is:

[0074]

[0075] wherein E(t) is the estimated value at t time, a is the parameter weight coefficient, D(t) is the actual value at t time, and T is the current time.

[0076] (2) For the third variable group, define the variables in the third variable group as the third variables, and take the data obtained at T-1 time of any third variable as the replacement value of any third variable at T time.

[0077] (3) For the second variable group, define the variables in the second variable group as the fourth variables, first generate a first replacement value of any fourth variable at T time based on the data generation strategy corresponding to the third variable group, take the first replacement value as the tentative value of any fourth variable at T time, then generate an estimated value of any fourth variable at T time based on the data generation strategy corresponding to the first variable group, and define it as a second replacement value, and set the average of the first replacement value and the second replacement value as the replacement value of any first variable at T time.

[0078] The second preset value, the third preset value and the third preset time are set according to the experience of those skilled in the art or according to the actual application scene, and the embodiments of the present application are not limited thereto.

[0079] Exemplarily, the data sequence A of temperature within the third preset time is [20, 21, 22,], wherein a1 is 20, a2 is 21, a3 is 22, and a4 is an abnormal value, and a replacement value is generated for a4. When the data sequence A belongs to the first variable group, if a is 0.2 and a1 is the initial value, the corresponding estimated value is 20, the corresponding estimated value of a2 is 20.2, the corresponding estimated value of a3 is 20.56, and the actual value of a4 is missing, then 20.56 is used as the replacement value thereof. When the data sequence A belongs to the third variable group, the replacement value of a4 is 22. When the data sequence A belongs to the second variable group, first generate a first replacement value for a4 based on the data generation strategy corresponding to the third variable group, which is 22, and then use the data generation strategy corresponding to the first variable group to generate a second replacement value for a4 based on taking 22 as the tentative value of a4, which is 20.848, and take the average 21.424 of the first replacement value and the second replacement value as the replacement value of a4.

[0080] The invalidity less than or equal to the second preset value indicates that the availability of the variable is high, and the invalidity greater than the third preset value indicates that the availability of the variable is low. The variables in the first variable group have high availability, less data missing, and contain sufficient information to estimate trends and patterns. The replacement values are generated according to the method of Formula One, the data of multiple time points can be considered, the continuity and consistency of the data can be maintained, the data noise caused by random fluctuations can be reduced, and the accuracy of the generated replacement values can be improved. The variables in the third variable group have low availability, more data missing, and the parameter changes infrequently or the change trend is not obvious. If a complex interpolation method is used, it may cause overfitting and introduce too much estimation error. By simply continuing the last observation, the risk of overfitting is reduced, and the data can be quickly and simply processed, reducing the consumption of computing resources. The variables in the second variable group have a certain number of missing values, and the availability is not very high. The integrity of the data set is affected to a certain extent, but there is still enough data for analysis. Such variables may not change as frequently as variables in the first variable group, nor change slowly or stably as variables in the third variable group. They may be between the two, with a certain degree of dynamics. By combining the results of the two different data generation strategies, the bias that may be introduced by a single method can be reduced, and the accuracy of the generated replacement values can be improved.

[0081] By customizing the data generation strategy for each variable group, different characteristics of the data can be processed more accurately and effectively, the quality of data processing can be improved, the accuracy and robustness of the data can be enhanced, the performance fluctuations caused by data quality problems can be reduced, and the accuracy of subsequent analysis can be improved. Reliable early warning information is provided.

[0082] Step 2, extracting bearing temperature related variables from the preprocessed online SCADA data, inputting the related variables into a preset model to obtain a bearing temperature theoretical value.

[0083] Specifically, the above-mentioned preset model is a model for obtaining a bearing temperature theoretical value by training temperature related variables.

[0084] In a specific embodiment, the method for establishing the preset model is:

[0085] Step 21, obtaining historical SCADA data of a wind turbine bearing, the historical SCADA data including N1 data sequences, each data sequence including second generator operation data, wherein N1 is a positive integer.

[0086] Step 22, traversing the historical SCADA data, when there is missing data corresponding to any feature variable in the second generator operation data, defining any feature variable as a first feature variable, and dividing the historical SCADA data into a training data set and a validation data set according to a preset proportion.

[0087] Step 23, any combination of all first characteristic variables is performed to obtain N2 characteristic variable combinations, wherein N2 is a positive integer.

[0088] Step 24, any characteristic variable combination is extracted, the variable in any characteristic variable combination is defined as a third characteristic variable, the variable outside any characteristic variable combination is defined as a second characteristic variable, all data corresponding to any second characteristic variable in the training data set is deleted, and the data sequence lacking data corresponding to the third characteristic variable is deleted, to generate a first training data set, step 24 is repeated until all characteristic variable combinations are traversed, N3 first training data sets are generated, wherein N3 is a positive integer less than or equal to N2.

[0089] Step 25, the same operation as step 24 is performed on the verification data set to generate N4 first verification data sets, wherein N4 is a positive integer less than or equal to N2.

[0090] Step 26, machine learning is performed on any first training data set to obtain a first preset model corresponding to any first training data set, after all first training data sets are traversed, N3 first preset models are generated.

[0091] Step 27, each first preset model is verified using the first verification data set, the accuracy of each first preset model is obtained respectively, the first preset model corresponding to the maximum accuracy is taken as the preset model.

[0092] Specifically, the second generator operating data includes but is not limited to bearing temperature data, cabin temperature data, environmental temperature data, wind speed, wind direction, and power, lubrication state data, rotating speed, torque and power of each component of the bearing. The second generator operating data includes various equipment operating data related to the bearing temperature.

[0093] Exemplarily, the historical SCADA data includes x1, x2,..., x11, 11 data sequences, each data sequence has b1, b2,..., b9, 9 variables, wherein the data corresponding to the b1 variable in the x1 sequence is missing, and the data corresponding to the b5 variable in the x9 sequence is missing, and b1 and b5 are defined as the first characteristic variables. There are two first characteristic variables b1 and b5, and any combination of the two first characteristic variables exists, and there are 3 combination modes, b1, b5, (b1, b5). When b1 is used as the third characteristic variable, b5 is used as the second characteristic variable, and if the training data set is (x1, x2,..., x7) and the verification data set is (x8, x9, x10, x11), for the training data set, the variable b5 is deleted from the 7 data sequences (remaining b1, b2, b3, b4, b6,..., b9, 8 variables), and the data sequence x1 in which the third characteristic variable b1 is missing is deleted (remaining x2,..., x7, 6 data sequences); for the verification data set, the variable b5 is deleted from the 4 data sequences (remaining b1, b2, b3, b4, b6,..., b9, 8 variables), and since there is no data sequence in which the third characteristic variable b1 is missing in the verification data set, no further operation is required.

[0094] The technical scheme of the present application provides flexibility and adjustability for model training in the presence of data missing, and maximizes the use of available data for effective learning and prediction. By creating multiple training data sets, training multiple first preset models, further evaluating and comparing the multiple first preset models, and selecting the best preset model, the influence of different characteristic variable combinations on the prediction ability of the model can be explored, which helps to reduce the influence of variable noise that does not contribute to the bearing theoretical temperature prediction on the preset model, thereby improving the robustness and accuracy of the model.

[0095] In a specific embodiment, step 27 includes:

[0096] (1) Extract any first preset model, and define the characteristic variables used when training any first preset model as fourth characteristic variables.

[0097] (2) Extract any first verification data set, and define the characteristic variables contained in any first verification data set as fifth characteristic variables. When the fifth characteristic variables contain the fourth characteristic variables, any first verification data set is defined as a second verification data set.

[0098] (3) After traversing all first verification data sets, each second verification data set is used to verify any first preset model, a plurality of model accuracies are obtained, a first statistical value of all model accuracies is calculated, and the first statistical value is used as the accuracy of any first preset model.

[0099] Specifically, the first statistical value is the average of the accuracy of all models.

[0100] Using the validation dataset containing the variables used to train the first preset model, the first preset model is validated, which can ensure the pertinence, accuracy and efficiency of model validation, and help improve the accuracy and reliability of model selection.

[0101] In a specific embodiment, in step 2, the bearing temperature related variables are extracted from the preprocessed online SCADA data, including:

[0102] The feature variables used in the training of the preset model are defined as the sixth feature variables, and the related variables are extracted from the first SCADA data based on the sixth feature variables.

[0103] Specifically, according to the sixth feature variables, the variables most relevant to bearing temperature prediction are extracted more pertinently, the performance fluctuation of the model caused by inconsistent feature variables is reduced, the influence of non-contributing variable noise on model prediction is reduced, the robustness of the model and the accuracy of data processing are improved, and the data processing speed and efficiency are improved.

[0104] Step 3, extract the actual value of bearing temperature from the preprocessed online SCADA data, calculate the temperature difference between the actual value of bearing temperature and the theoretical value of bearing temperature, and judge that there is an anomaly when the absolute value of the temperature difference is greater than the first preset value.

[0105] The first preset value is set according to the experience of those skilled in the art or according to the actual application scene, and the embodiments of the present application are not limited thereto.

[0106] The theoretical value of bearing temperature is calculated based on the preset model, which represents the state of bearing temperature under normal circumstances and provides a benchmark for temperature evaluation. The actual temperature value of the bearing is affected by various factors, including environmental changes, equipment aging, operating condition changes, etc. These factors may cause the actual temperature to deviate from the theoretical temperature. By comparing the two, unknown abnormal conditions can be detected.

[0107] The anomaly of bearing temperature may be a precursor of equipment failure. By comparing the actual temperature and the theoretical temperature, problems can be found before failure occurs, so that preventive maintenance measures can be taken to ensure the stable operation and safety of the wind turbine generator.

[0108] Step 4, obtain the first generator operating data within the first preset time, diagnose the generator bearing based on the first generator operating data, and issue a warning message.

[0109] The first preset time is set according to experience of a person skilled in the art or according to an actual application scenario, and embodiments of the present application do not limit this.

[0110] The first generator operation data includes, but is not limited to, bearing temperature data, cabin temperature data, ambient temperature data, wind speed, wind direction, and power, lubrication state data, rotation speed, torque and power of each component of the bearing. The first generator operation data includes various equipment state data related to bearing abnormal diagnosis.

[0111] In a specific embodiment, step 4 includes:

[0112] (1) The first generator operation data is divided into a plurality of variable data sequences based on variable types, the second statistical value of each variable data sequence is calculated respectively, and the first feature vector corresponding to the first generator operation data is generated based on the second statistical value.

[0113] (2) The second feature vector corresponding to each fault event within the third preset time is extracted, and the Euclidean distance between any second feature vector and the first feature vector is calculated, wherein the second feature vector of any fault event is generated based on the historical generator operation data within the first preset time before a specific time, and the specific time is a time point at a specific time from the time point when any fault event occurs.

[0114] (3) It is determined whether the minimum value of the Euclidean distance is less than the fourth preset value, and if so, it is determined that there is a fault risk.

[0115] The third preset time and the fourth preset value are set according to experience of a person skilled in the art or according to an actual application scenario, and embodiments of the present application do not limit this.

[0116] Specifically, the second statistical value is the average value. The variable type is the type of each variable in the first generator operation data, including but not limited to pressure, temperature, humidity, and rotation speed. For example, the first generator operation data includes three variables c1, c2 and c3, the first generator operation data is divided into three variable data sequences, the variable data sequence y1 corresponding to the variable c1 is [c11,..., c1m], the variable data sequence y2 corresponding to the variable c2 is [c21,..., c2m], the variable data sequence y3 corresponding to the variable c3 is [c31,..., c3m], the second statistical value of the variable data sequence y1 is E1, the second statistical value of the variable data sequence y2 is E2, and the second statistical value of the variable data sequence y3 is E3. The first feature vector corresponding to the first generator operation data is [E1, E2, E3].

[0117] The second feature vector is a feature vector corresponding to the generator operation data in a period of time before the occurrence of the fault event, and the second feature vector is obtained using the same method as the first feature vector. For example, the specific time is the time of the last acquisition of the generator operation data before the occurrence of the fault event, if the first preset time is 30 minutes, the occurrence time of the fault event P is 10:00, and the time of the last acquisition of the generator operation data before the occurrence of the fault event P is 9:55, then the second feature vector corresponding to the fault event P is a feature vector corresponding to the generator operation data in the period of 9:25-9:55.

[0118] Preferably, the specific time can also be the time of the first acquisition of the abnormal generator operation data before the occurrence of the fault event.

[0119] By calculating the Euclidean distance between the second feature vector and the first feature vector, the similarity between the first feature vector and the second feature vector corresponding to the fault event is compared, whether the current operation of the generator is similar to the past fault condition can be quickly identified, whether a fault is about to occur can be determined, a fault event can be identified in advance, and timely measures can be taken for intervention.

[0120] If the minimum Euclidean distance is less than the fourth preset value, it indicates that the current generator operation state has a high similarity to the feature vector in the historical fault data, and can be in a high-risk state of fault occurrence. By identifying the second feature vector corresponding to the minimum Euclidean distance, the cause of the fault can be analyzed, the potential factors leading to the fault can be understood, a basis for fault elimination can be provided, the decision-making process can be simplified, and an explicit fault warning and recommended action plan can be provided.

[0121] Preferably, the warning information further includes a fault prediction time, which is generated based on the time from the specific time of the fault event corresponding to the minimum Euclidean distance to the time of the fault occurrence.

[0122] By predicting the time point at which the generator can possibly occur a fault in the future, maintenance activities can be effectively planned, and the operation and maintenance efficiency can be improved.

[0123] In a specific embodiment, after it is judged that there is a fault risk, the following steps are further included:

[0124] (1) defining the statistical value in the second feature vector corresponding to the minimum Euclidean distance as a third statistical value, and calculating the sensitivity corresponding to any variable type based on Formula Two, wherein Formula Two is:

[0125]

[0126] wherein G i is the sensitivity of the i-th variable type, ST i,2 is the second statistical value of the i-th variable type, and ST i,3The third statistical value is for the i-th variable type, and d is the minimum value of the Euclidean distance.

[0127] (2) When the absolute value of the sensitivity corresponding to any variable type is greater than a preset threshold, the any variable type is determined as an abnormal influencing factor.

[0128] (3) After traversing all variable types, the warning information is generated based on all abnormal influencing factors.

[0129] The preset threshold is set according to the experience of a person skilled in the art or according to an actual application scenario, and the embodiments of the present application do not limit this.

[0130] By calculating the sensitivity corresponding to any variable type, the change rate of the Euclidean distance with respect to each variable data can be quantitatively described, and it can be identified which variables have a greater contribution to the fault risk. These variables may be key factors leading to failure, thereby reducing false positives caused by non-key factors, performing more refined fault analysis, and improving the accuracy of the warning system. At the same time, after identifying the specific abnormal influencing factors, more targeted warning information is generated to help maintenance personnel quickly locate the problem.

[0131] The above describes the abnormal monitoring and warning method for the generator bearing temperature of the wind turbine generator in the embodiments of the present application. The following describes the abnormal monitoring and warning device for the generator bearing temperature of the wind turbine generator in the embodiments of the present application. Please refer to Figure 2 The abnormal monitoring and warning device for the generator bearing temperature of the wind turbine generator in the embodiments of the present application includes a preprocessing module 10, a temperature analysis module 20, an abnormal judgment module 30, and an abnormal diagnosis module 40.

[0132] The preprocessing module 10 is configured to obtain online SCADA data of a wind turbine generator and pre-process the online SCADA data.

[0133] The temperature analysis module 20 is configured to extract bearing temperature related variables from the pre-processed online SCADA data, input the related variables into a preset model, and obtain a bearing temperature theoretical value.

[0134] The abnormal judgment module 30 is configured to extract a bearing temperature actual value from the pre-processed online SCADA data, calculate a temperature difference between the bearing temperature actual value and the bearing temperature theoretical value, and determine that an abnormality exists when the absolute value of the temperature difference is greater than a first preset value.

[0135] The abnormal diagnosis module 40 is configured to obtain first generator operation data within a first preset time, diagnose the generator bearing based on the first generator operation data, and issue a warning information.

[0136] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0137] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0138] The above-described and the above embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for abnormal monitoring and early warning of wind turbine generator bearing temperature, characterized in that, The abnormal monitoring and early warning method of the wind turbine generator bearing temperature comprises: Step 1, obtaining online SCADA data of a wind turbine generator, and preprocessing the online SCADA data; Step 2, extracting bearing temperature related variables from the preprocessed online SCADA data, inputting the related variables into a preset model to obtain a bearing temperature theoretical value; Step 3, extracting a bearing temperature actual value from the preprocessed online SCADA data, calculating a temperature difference between the bearing temperature actual value and the bearing temperature theoretical value, and determining that there is an abnormality when the absolute value of the temperature difference is greater than a first preset value; Step 4, obtaining first generator operation data within a first preset time, diagnosing the generator bearing based on the first generator operation data, and issuing a warning information; In the step 1, the preprocessing of the online SCADA data comprises: Step 11, determining whether there is missing data or abnormal data in the online SCADA data, and defining a first variable corresponding to the missing data or the abnormal data if there is; Step 12, extracting any first variable, obtaining a theoretical collection number of any first variable within a second preset time, and a data corresponding to any first variable as an invalid number of the missing data or the abnormal data, calculating a ratio of the invalid number to the theoretical collection number, and defining the ratio as an invalid rate; Step 13, after traversing all first variables, dividing all first variables into a preset number of variable groups based on the invalid rate according to a preset rule; Step 14, setting a corresponding data generation strategy for each variable group, extracting any variable group, generating a replacement value for each first variable in any variable group using the data generation strategy corresponding to any variable group, and using the replacement value to complete the online SCADA data; The preset number is 3, the first variable with an invalid rate less than a second preset value is divided into a first variable group, the first variable with an invalid rate greater than or equal to the second preset value and less than or equal to a third preset value is divided into a second variable group, and the first variable with an invalid rate greater than the third preset value is divided into a third variable group, and the step 14 comprises: For the first variable group, defining variables in the first variable group as second variables, obtaining a data sequence of any second variable within a third preset time, calculating an estimated value of any second variable at each time point based on formula one, and taking the estimated value of any second variable at T-1 moment as the replacement value of any second variable at T moment, and the formula one is: ， wherein E(t) is the estimation value at time t, is a parameter weight coefficient, D(t) is the actual value at time t, and T is the current time. For the third variable group, defining variables in the third variable group as third variables, and taking data obtained at T-1 moment of any third variable as the replacement value of any third variable at T moment; For the second variable group, the variables in the second variable group are defined as fourth variables, the first replacement value of any fourth variable at T time is generated based on the data generation strategy corresponding to the third variable group, the first replacement value is taken as the tentative value of any fourth variable at T time, then the estimated value of any fourth variable at T time is generated based on the data generation strategy corresponding to the first variable group, and is defined as a second replacement value, and the average value of the first replacement value and the second replacement value is set as the replacement value of any first variable at T time; The method for establishing the preset model is: Step 21, obtaining historical SCADA data of a wind turbine bearing, the historical SCADA data comprising N1 data sequences, each data sequence comprising second generator operating data, wherein N1 is a positive integer; Step 22, traversing the historical SCADA data, when there is missing data corresponding to any characteristic variable in the second generator operating data, defining any characteristic variable as a first characteristic variable, and dividing the historical SCADA data into a training data set and a validation data set according to a preset proportion; Step 23, combining all first characteristic variables in any combination to obtain N2 characteristic variable combinations, wherein N2 is a positive integer; Step 24, extracting any characteristic variable combination, defining the variables in any characteristic variable combination as third characteristic variables, defining the variables other than any characteristic variable combination as second characteristic variables, deleting all data corresponding to any second characteristic variable in the training data set, and deleting data sequences lacking data corresponding to third characteristic variables, to generate a first training data set, and repeating step 24 until all characteristic variable combinations are traversed, to generate N3 first training data sets, wherein N3 is a positive integer less than or equal to N2; Step 25, performing the same operation on the validation data set as step 24 to generate N4 first validation data sets, wherein N4 is a positive integer less than or equal to N2; Step 26, machine learning on any first training data set to obtain a first preset model corresponding to any first training data set, and generating N3 first preset models after traversing all first training data sets; Step 27, verifying each first preset model using the first validation data set to obtain the accuracy of each first preset model, and taking the first preset model corresponding to the maximum accuracy as the preset model; The step 27 comprises: Extracting any first preset model, and defining the characteristic variables used to train any first preset model as fourth characteristic variables; Extracting any first validation data set, and defining the characteristic variables contained in any first validation data set as fifth characteristic variables, when the fifth characteristic variables contain the fourth characteristic variables, defining any first validation data set as a second validation data set; After traversing all the first verification data sets, each second verification data set is used to verify any of the first preset models, a plurality of model accuracies are obtained, a first statistical value of all model accuracies is calculated, and the first statistical value is taken as the accuracy of any of the first preset models.

2. The method of abnormal monitoring and early warning of wind turbine generator bearing temperature according to claim 1, characterized in that, In the step 2, the bearing temperature related variable is extracted from the preprocessed online SCADA data. The feature variable used in the preset model training is defined as a sixth feature variable, and the related variable is extracted from the preprocessed online SCADA data based on the sixth feature variable.

3. The method of abnormal monitoring and early warning of wind turbine generator bearing temperature according to claim 1, characterized in that, The second generator operation data includes bearing temperature data, cabin temperature data, ambient temperature data, wind speed, wind direction, and power, lubrication state data, rotating speed, torque and power of each component of the bearing.

4. The method of abnormal monitoring and early warning of wind turbine generator bearing temperature according to claim 1, characterized in that, The step 4 includes: The first generator operation data is divided into a plurality of variable data sequences based on variable types, a second statistical value of each variable data sequence is calculated respectively, and a first feature vector corresponding to the first generator operation data is generated based on the second statistical value; A second feature vector corresponding to each fault event in a third preset time is extracted, and an Euclidean distance between any second feature vector and the first feature vector is calculated, wherein the second feature vector of any fault event is generated based on historical generator operation data in the first preset time before a specific time, and the specific time is a time point at a specific time from a time point of occurrence of any fault event; It is judged whether the minimum Euclidean distance is less than a fourth preset value, and if so, it is judged that there is a fault risk.

5. The method of abnormal monitoring and early warning of wind turbine generator bearing temperature according to claim 4, characterized in that, After judging that there is a fault risk, the following steps are further included: A statistical value in the second feature vector corresponding to the minimum Euclidean distance is defined as a third statistical value, and a sensitivity corresponding to any variable type is calculated based on formula two, and the formula two is: ， wherein G i is the sensitivity of the ith variable type, ST i,2 is the second statistical value of the ith variable type, ST i,3 is the third statistical value of the ith variable type, d is the Euclidean distance minimum value; When the absolute value of the sensitivity corresponding to any variable type is greater than a preset threshold, any variable type is determined as an abnormal influencing factor; After traversing all variable types, the prewarning information is generated based on all abnormal influencing factors.

6. An abnormal monitoring and early warning device for the temperature of a generator bearing of a wind turbine generator unit, characterized in that, The abnormal monitoring and prewarning device for the generator bearing temperature of the wind turbine generator includes a preprocessing module, a temperature analysis module, an abnormality judgment module and an abnormality diagnosis module. The preprocessing module is used to obtain online SCADA data of a generator of a wind turbine generator, and pre-process the online SCADA data. The temperature analysis module is used to extract a bearing temperature related variable from the preprocessed online SCADA data, input the related variable into a preset model, and obtain a bearing temperature theoretical value. The abnormality judgment module is used to extract a bearing temperature actual value from the preprocessed online SCADA data, calculate a temperature difference between the bearing temperature actual value and the bearing temperature theoretical value, and judge that there is an abnormality when the absolute value of the temperature difference is greater than a first preset value. The abnormality diagnosis module is used to obtain first generator operation data in a first preset time, diagnose a generator bearing based on the first generator operation data, and issue a prewarning information. The preprocessing of the online SCADA data comprises: Step 11, judging whether there is missing data or abnormal data in the online SCADA data, if so, defining the variable corresponding to the missing data or the abnormal data as a first variable; Step 12, extracting any first variable, obtaining the theoretical collection number of any first variable within a second preset time, and the invalid number of data corresponding to any first variable being the missing data or the abnormal data, calculating the ratio of the invalid number to the theoretical collection number, and defining it as an invalid rate; Step 13, after traversing all first variables, based on the invalid rate, all first variables are divided into a preset number of variable groups according to a preset rule; Step 14, setting a corresponding data generation strategy for each variable group, extracting any variable group, using the data generation strategy corresponding to any variable group to generate a replacement value for each first variable in any variable group, and using the replacement value to complete the online SCADA data; The preset number is 3, the first variable with an invalid rate less than a second preset value is divided into a first variable group, the first variable with an invalid rate greater than or equal to the second preset value and less than or equal to a third preset value is divided into a second variable group, and the first variable with an invalid rate greater than the third preset value is divided into a third variable group, and the step 14 comprises: For the first variable group, defining the variables in the first variable group as second variables, obtaining the data sequence of any second variable within a third preset time, calculating the estimated value of any second variable at each time point based on formula one, and taking the estimated value of any second variable at T-1 moment as the replacement value of any second variable at T moment, the formula one is: ， wherein E(t) is the estimation value at time t, is a parameter weight coefficient, D(t) is the actual value at time t, and T is the current time. For the third variable group, defining the variables in the third variable group as third variables, and taking the data obtained at T-1 moment of any third variable as the replacement value of any third variable at T moment; For the second variable group, defining the variables in the second variable group as fourth variables, first generating a first replacement value of any fourth variable at T moment based on the data generation strategy corresponding to the third variable group, taking the first replacement value as the tentative value of any fourth variable at T moment, and then generating the estimated value of any fourth variable at T moment based on the data generation strategy corresponding to the first variable group, and defining it as a second replacement value, and setting the average value of the first replacement value and the second replacement value as the replacement value of any first variable at T moment; The establishment method of the preset model comprises: Step 21, obtaining historical SCADA data of a wind turbine bearing, the historical SCADA data comprising N1 data sequences, each data sequence comprising second generator operation data, wherein N1 is a positive integer; Step 22, traversing the historical SCADA data, when any characteristic variable in the second generator operation data corresponds to missing data, defining any characteristic variable as a first characteristic variable, and dividing the historical SCADA data into a training data set and a validation data set according to a preset proportion; Step 23, combining all first characteristic variables in any combination to obtain N2 characteristic variable combinations, wherein N2 is a positive integer; Step 24, extracting any characteristic variable combination, defining variables in any characteristic variable combination as third characteristic variables, defining variables other than any characteristic variable combination as second characteristic variables, deleting all data corresponding to any second characteristic variable in the training data set, and deleting data sequences lacking data corresponding to third characteristic variables, to generate a first training data set, repeating step 24 until all characteristic variable combinations are traversed, generating N3 first training data sets, wherein N3 is a positive integer less than or equal to N2; Step 25, performing the same operation on the validation data set as step 24 to generate N4 first validation data sets, wherein N4 is a positive integer less than or equal to N2; Step 26, performing machine learning on any first training data set to obtain a first preset model corresponding to any first training data set, and generating N3 first preset models after traversing all first training data sets; Step 27, verifying each first preset model using the first validation data set to obtain the accuracy of each first preset model, and selecting the first preset model with the maximum accuracy as the preset model; The step 27 comprises: extracting any first preset model, and defining the characteristic variable used to train any first preset model as a fourth characteristic variable; extracting any first validation data set, and defining the characteristic variable contained in any first validation data set as a fifth characteristic variable, when the fifth characteristic variable contains the fourth characteristic variable, defining any first validation data set as a second validation data set; after traversing all first validation data sets, verifying any first preset model using each second validation data set to obtain multiple model accuracies, calculating a first statistical value of all model accuracies, and taking the first statistical value as the accuracy of any first preset model.

Images