CN110414155B - Fan component temperature abnormity detection and alarm method with single measuring point - Google Patents
Fan component temperature abnormity detection and alarm method with single measuring point Download PDFInfo
- Publication number
- CN110414155B CN110414155B CN201910702600.5A CN201910702600A CN110414155B CN 110414155 B CN110414155 B CN 110414155B CN 201910702600 A CN201910702600 A CN 201910702600A CN 110414155 B CN110414155 B CN 110414155B
- Authority
- CN
- China
- Prior art keywords
- temperature
- data
- model
- component
- alarm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F03—MACHINES OR ENGINES FOR LIQUIDS; WIND, SPRING, OR WEIGHT MOTORS; PRODUCING MECHANICAL POWER OR A REACTIVE PROPULSIVE THRUST, NOT OTHERWISE PROVIDED FOR
- F03D—WIND MOTORS
- F03D17/00—Monitoring or testing of wind motors, e.g. diagnostics
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02B—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO BUILDINGS, e.g. HOUSING, HOUSE APPLIANCES OR RELATED END-USER APPLICATIONS
- Y02B10/00—Integration of renewable energy sources in buildings
- Y02B10/30—Wind power
Landscapes
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Sustainable Development (AREA)
- Sustainable Energy (AREA)
- Chemical & Material Sciences (AREA)
- Combustion & Propulsion (AREA)
- Mechanical Engineering (AREA)
- General Engineering & Computer Science (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The application relates to a fan component temperature anomaly detection and alarm method with single measuring points, which comprises the steps of utilizing training set data in an early fault-free period of equipment to fit data of temperature measuring points of a component to be measured to obtain coefficients of a multivariate linear equation, and constructing a temperature fitting model; predicting temperature measuring points of the test set data by using a fitting model, and using a difference value between the actual temperature of the test data and the model predicted temperature as a predicted residual sequence; during online operation, collecting real-time operation data of the wind turbine generator with a fixed time length T, and predicting a temperature measuring point of the online data by using the model through the extracted features to obtain a model predicted temperature; and (3) taking the difference value between the real-time temperature of the part to be measured and the model predicted temperature as a predicted residual sequence, and entering a part temperature early warning program based on the statistical analysis of a group of obtained online residual values.
Description
Technical Field
The application relates to a fan component temperature anomaly detection and alarm method with a single measuring point, which is suitable for the technical field of fan anomaly detection.
Background
The existing partial technologies, including the main control system of the wind turbine generator, are often methods for limiting threshold values for early warning logics of temperature abnormality, namely, it is found that the temperature is lower than or higher than a certain preset threshold value, and then an alarm is generated. Taking the early warning logic of the main bearing temperature as an example, one or two temperature measuring points are usually arranged on a fan component to be measured, the operating state of the main bearing is monitored by a wind power plant SCADA system (data acquisition and monitoring control system) through a method of simply setting a main bearing temperature threshold, and when the main bearing temperature is higher than the set threshold, the SCADA sends an alarm to a wind power plant owner. The early warning mode is too simple, the false alarm rate is extremely high, the fault of the component is usually an accumulation and progressive process, the alarm of the SCADA master control system is not advanced with proper length, and the predictive maintenance of an owner cannot be assisted. The detection and evaluation criteria for temperature anomaly in the prior art mainly comprise the following criteria:
(1) and an early warning mechanism based on the component temperature characteristic quantity. In chinese patent application 201410355897.X, 201810698450.0, a method of dividing working conditions is used to extract temperature characteristic quantities from subsets divided from unit data, and a statistical method that real-time temperature exceeds a threshold value is used to perform fault early warning according to a determined threshold value. The early warning mode is more detailed, the temperature difference of the wind turbine generator under different operation conditions is considered, but the method still belongs to an absolute threshold value method, the defects of low accuracy and unstable alarm still exist, and the method greatly depends on manually determined threshold values.
(2) And a residual error early warning mechanism based on a normal behavior model. In chinese patent applications 201810146979.1, 201711477916.6, and 201710853212.8, a neural network model is trained by obtaining variables related to the temperature of a component to be measured in fan history data, the temperature of the component to be measured is predicted, and a fault is determined by residual deviation between an actual temperature and a predicted temperature after a real-time temperature is obtained on line. The early warning mode is rigorous, the physical characteristics and the historical operating condition of the component operation are fully considered, the residual error can generally embody the fault characteristics when the fault occurs, but due to the dependence on the integrity of the working condition of the training data and the coverage rate of the working condition, the method only depends on the judgment of the absolute value of the residual error when the working condition changes, the false alarm is easy to occur, and the stability is poor.
(3) And (4) a cluster-based benchmarking early warning mechanism. In the chinese patent application 201910027340.6, by setting the monitoring unit and the reference unit to the full wind farm fan, the temperature data obtained by the component to be measured of the monitoring unit is compared with the average value of the temperature data sets of the remaining reference units each time, and if the temperature difference exceeds a certain proportion and has an obvious rising trend, the temperature fault of the monitoring unit is determined. The early warning method is visual, but the requirements on the consistency of the model characteristics, the health state assumption and the physical characteristics during operation of the full wind field fan are high, and once the consistency of the running conditions of the fan cannot be met, the early warning accuracy cannot be guaranteed. In addition, especially in mountain wind farms, the climatic conditions of adjacent fans may even differ due to the effect of turbulence.
In addition, the existing method cannot form a system closed loop, and is difficult to systematically monitor and alarm the temperature abnormity of the wind turbine generator component. Experience shows that once a wind turbine generator component has a temperature abnormal fault, the fault can be continued until the fault is ended, the characteristic has extremely high requirements on an alarm mechanism, the temperature monitoring systems of the components are required to be accurately and stably early-warned before the fault occurs, and the early-warning level is correspondingly improved due to the fact that the temperature monitoring systems are gradually close to the fault, which is the biggest challenge in the prior art.
In summary, the problems in the prior art include: whether the judgment method is based on the temperature threshold or the residual error model, the accuracy, the predictability and the stability of the alarm are often difficult to meet at the same time; at present, the existing temperature anomaly diagnosis technology is not generally transited to the alarm logic which is finally oriented to a user terminal, the requirement and the requirement on the technology are higher, and for a wind farm owner, on the premise of reducing the false alarm rate and the missing alarm rate, an alarm signal needs to be stably output as much as possible to play the roles of preventing diseases and reminding.
Disclosure of Invention
The method and the device overcome the irritability of absolute threshold discrimination, improve the accuracy and stability compared with a simple residual discrimination model, reduce the false alarm rate on the premise of keeping sensitivity to abnormality, finally form a reasonable alarm with grade discrimination, and facilitate the predictive maintenance of wind farm owners.
According to the method for detecting and alarming the temperature abnormity of the fan component with the single measuring point, the fan component is provided with the single temperature measuring point, and the method for detecting and alarming comprises a mechanism driving data modeling program and a component temperature early warning program based on residual errors; the mechanism driven data modeling program includes the steps of:
(1) acquiring monitoring data of a fan in operation, and selecting data of a group of equipment in an early stage without faults of a component to be tested as training data;
(2) screening out original variables related to the temperature of the component according to the physical characteristics of the component to be tested;
(3) preprocessing the original variable to obtain a new processing variable, and constructing a multivariate linear equation for describing the temperature of the component according to the original variable and the processing variable;
(4) dividing the acquired historical data into a training set and a test set, wherein the training set is used for fitting a temperature prediction model, and the test set is used for determining a threshold value of a subsequent step;
(5) fitting data of temperature measuring points of the component to be measured by using a training set to obtain coefficients of a multivariate linear equation, and constructing a temperature fitting model to obtain a fitting model of the measured values of the temperature measuring points of the component;
(6) predicting the temperature measuring points of the test set data by using the fitting model to obtain model predicted temperature; meanwhile, the fitting precision of the model is judged, and the fitting model is stored when the accuracy requirement is met;
(7) using the difference value of the actual temperature of the test data and the model predicted temperature as a predicted residual sequence;
(8) for the single-point component, obtaining a group of residual errors, and storing the residual errors for online prediction;
(9) during online operation, collecting real-time operation data of the wind turbine generator with a fixed time length T, and extracting features of the online data according to the method in the step (3);
(10) predicting the temperature measuring point of the online data through the extracted features by using the model to obtain the model predicted temperature;
(11) using the difference value between the real-time temperature of the part to be measured and the model predicted temperature as a predicted residual sequence to obtain a group of online residuals;
(12) and entering the part temperature early warning program based on the statistical analysis of the obtained group of on-line residual error values.
Preferably, the residual error-based component temperature early warning program comprises the following steps:
(1) recording the total number N of sample points of the online data in the time period T; for a single-point component, firstly, normal fitting is carried out on the residual errors of a test set through a statistical method to obtain the variance estimation sigma after fitting 2 ；
(2) Adding a weight term lambda to an n sigma interval of the fitted normal probability density curve, wherein the closer the weight is to the center of the mean value, the smaller the weight is, and the larger the weight is, the more the weight deviates from the center of the mean value; interactively constructing a health index HI according to the weight items and the number of probability intervals of the online residual error samples falling to the right
λ k Representing a set weight term, n k The number of probability intervals when the online residual error samples fall to the right side is represented, and i is 1, 2, 3 and 4;
(3) processing the online residual error by sliding window, calculating four HIs for the sample points in each small window, and counting the HIs 1 ，HI 2 ，HI 3 ，HI 4 A value of (d); if HI is in the time window 3 +HI 4 Greater than HI 1 +HI 2 Then record W for the time window 1 Otherwise, it is denoted as W 0 ；
For W in all time windows 0 And W 1 Making statistics if most of the classes are W 1 If so, judging that the temperature abnormal fault occurs in the fan component within the time T, and outputting an alarm identification parameter as true.
Preferably, all warning identification parameters of the measuring points of the component to be measured are output to an algorithm operation result database for storage, and warning identification parameter values stored in history are called for logic judgment in each algorithm operation; if the warning identification parameters output by the algorithm operation for three times continuously are true or the warning identification parameters in the historical n-time operation are true and exceed half, the operation triggers an alarm; wherein the size of n is determined according to the running frequency of the algorithm.
Preferably, when an alarm is triggered, setting the alarm identification parameters of corresponding measuring points to be true, outputting the alarm identification parameters corresponding to the measuring points to an algorithm operation result database for storage, calling historically stored alarm identification parameter values for logic judgment in each algorithm operation, and determining the risk value of the alarm by taking the number of the alarm identification parameters in the historical k operations as true if the alarm identification parameters of the corresponding measuring points in the current operation are true, and mapping the proportion to be the alarm level; the size of k is determined according to the running frequency of the algorithm, and k > n.
After the fan component is overhauled, resetting a historical early warning library of the corresponding component; the original variables in the step (2) of the mechanism-driven data modeling program comprise at least one of wind speed, active power, generator rotating speed, measured values of temperature measuring points and cabin temperature; in the step (6) of the mechanism-driven data modeling program, the judgment standard of the fitting accuracy of the model is carried out through residual analysis and fitting accuracy analysis; and if the fitting precision is less than 10% of the overall data mean value, and the residual error obeys the normality distribution through the QQ-Norm test and the Jarqe-Bera test, the requirement of the accuracy is considered to be met.
The beneficial effect of this application does:
1. the method provides a temperature abnormity monitoring and alarming method for a component with a temperature measuring point of the wind turbine generator, and can be suitable for monitoring the temperature of the component with a single measuring point;
2. according to the method, according to the physical laws of operation of different components, a normal behavior model is adopted for modeling to obtain a predicted temperature residual error and an actual temperature residual error, the characteristics and the historical operating condition of each single fan are fully considered, for the fan component with a single measuring point, a statistical distribution analysis method is used on the basis of the obtained one-dimensional residual error, the alarm accuracy is fully improved, and a certain lead is provided compared with the absolute threshold early warning;
3. on the basis of a normal behavior model, the method fully considers the historical alarm condition of each fan independently and brings the historical alarm condition into the alarm logic of real-time monitoring, fully improves the stability of alarm, and meets the requirement of increasing the alarm grade of the occurrence of the adjacent fault.
Drawings
FIG. 1 is a schematic flow diagram of a method for detecting and warning temperature anomalies in a fan component with a single measuring point according to the present application.
FIG. 2 is a probability density curve of a test set residual normal fit in the present application.
Detailed Description
To make the objects, technical solutions and advantages of the present application more apparent, embodiments of the present application will be described in detail below with reference to the accompanying drawings. It should be noted that the embodiments and features of the embodiments in the present application may be arbitrarily combined with each other without conflict.
As shown in FIG. 1, according to the method for detecting and alarming the temperature abnormality of the fan component with the single measuring point, the fan component is provided with the temperature measuring point, the method for detecting and alarming the temperature abnormality of the fan component comprises two processes of mechanism driving data modeling and residual error-based component temperature early warning,
the mechanism-driven data modeling comprises the following processing steps:
1. acquiring monitoring data of an SCADA system when a fan operates, and selecting data of a group of equipment in an early stage and without failure of a component to be tested as training data;
2. screening out original variables related to the temperature of the component according to the physical characteristics of the component to be tested, wherein the original variables comprise wind speed, active power, the rotating speed of a generator, the measured value of a temperature measuring point, cabin temperature and a timestamp of data;
3. preprocessing the original variable to obtain a new processing variable, and constructing a multivariate linear equation for describing the temperature of the component according to the original variable and the processing variable;
4. according to the number of acquired history data, the data is acquired according to, for example, 7: 3, dividing the data into a training set and a test set according to the proportion, wherein the training set is used for fitting a temperature prediction model, and the test set is used for determining a threshold value in the subsequent step;
5. fitting the temperature measuring points of the component to be measured by using a regression method (including but not limited to linear regression, Ridge regression, LASSO regression and the like) to obtain coefficients of a multiple linear equation, constructing a temperature fitting model by using a tree model-based regression method (including but not limited to random forest algorithm, XGboost algorithm and the like) and a neural network-based regression method, and obtaining a model of the measured values of the temperature measuring points of the component;
6. predicting the temperature measuring points of the test data by using the model through the extracted features to obtain the predicted temperature of the model; meanwhile, the fitting precision of the model is judged, and the model is saved when the accuracy requirement is met; specific judgment criteria can be analyzed by, for example, residual analysis and fitting accuracy analysis, if the fitting accuracy (RMSE) is less than 10% of the overall data mean, and the residual follows normal distribution by QQ-Norm test and Jarque-Bera test;
7. using the difference value of the actual temperature of the test data and the model predicted temperature as a predicted residual sequence;
8. for a single-point component, obtaining a group of residual errors RES, and storing the residual errors RES for online prediction;
9. during online operation, collecting real-time operation data of the wind turbine generator with a fixed time length T, and extracting features of the online data according to the method in the step 3;
10. predicting the temperature measuring point of the online data through the extracted features by using the model to obtain the model predicted temperature;
11. and (3) using the difference value between the real-time temperature of the part to be measured and the model predicted temperature as a predicted residual sequence (hereinafter abbreviated as online residual), and obtaining a group of online residuals at a single measuring point.
The residual error-based part temperature early warning comprises the following processing steps:
1. recording the total number N of sample points of the online data in the time period T; for a one-point part, first goCarrying out normal fitting on the residual errors of the test set by an over-statistical method to obtain fitted variance estimation sigma 2 ；
Var(X)＝σ 2 ＝∫(x-μ) 2 f(x)dx＝∫x 2 f(x)dx-μ 2
2. Adding a weight term lambda to an n sigma interval of the fitted normal probability density curve, wherein the weight closer to the center of the mean value is smaller, the weight deviated from the center of the mean value is larger, the curve in fig. 2 is the probability density curve of the test set residual normal fitting, sample points in four probability intervals 1, 2, 3 and 4 respectively correspond to weight terms from small to large, and only samples falling on the right side (namely the right half part) of the mean value are concerned; interactively constructing a health index HI according to the weight items and the number of probability intervals of the online residual error samples falling to the right side
λ k Weight term, n, representing a setting k The number of probability intervals when the online residual error samples fall to the right side is represented, and i is 1, 2, 3 and 4;
3. processing the online residual error by sliding window, calculating four HIs for the sample points in each small window, and counting the HIs 1 ，HI 2 ，HI 3 ，HI 4 A value of (d); if HI in the time window 3 +HI 4 Greater than HI 1 +HI 2 Then, the time window is marked as W 1 Otherwise, it is recorded as W 0 ；
For W in all time windows 0 And W 1 Making statistics if most of the classes are W 1 If the temperature of the fan component is abnormal, judging that the temperature of the fan component is abnormal in the time T, and outputting a warning identification parameter WarningFlag to be True;
4. outputting all warning identification parameters WarningFlag of the measuring points of the component to be measured to an algorithm operation result database for storage, and calling the historically stored WarningFlag value for each algorithm operation to perform logic judgment; if WarningFlag output by three times of continuous algorithm operation is True or WarningFlag output by n times of historical operation exceeds half, the operation triggers an alarm, and an alarm identification parameter AlarmFlag of a corresponding measuring point is set as True, wherein the size of n depends on the operation frequency of the algorithm;
outputting an alarm identification parameter AlarmFlag corresponding to a measuring point to an algorithm operation result database for storage, calling a historically stored AlarmFlag value for logic judgment in each algorithm operation, determining the risk value of the alarm by taking the number of AlarmFlag as True in the historical k operations to determine if the AlarmFlag of the measuring point corresponding to the current operation is True, and mapping the ratio to an alarm level AlarmLevel, namely, the more the AlarmFlag is as True in the historical k operations, the higher the alarm level is; the size of k depends on the algorithm running frequency, usually k > > n;
5. if the algorithm does not trigger the alarm identification parameter WarningFlag or the alarm identification parameter AlarmFlag, no alarm is given;
6. and after the wind farm owner overhauls, resetting the historical early warning library of the corresponding part.
In the application, based on mechanism and data hybrid driving, data preprocessing and feature construction methods such as an energy conservation equation and the like are adopted to establish temperature fitting models corresponding to different measuring points, namely normal behavior models, so that the model prediction quality is improved, and the aim of predictive maintenance is fulfilled; aiming at a large single-point component, a targeted post-processing mechanism is adopted for the temperature prediction residual error, wherein: by means of probability density interval segmentation and sliding window, the accuracy of alarming is guaranteed, and the false alarm rate is effectively reduced; a self-feedback alarm mechanism is adopted, an algorithm model and a result database are connected, the operation alarm condition of a historical algorithm is fully considered and is included in the logic of early warning alarm, the stability of alarm is improved, and alarm feedback with gradually increasing grades is realized when a fault occurs.
Examples
The method is characterized in that fan operation data of a certain wind field in 2018 are used for carrying out algorithm alarm testing, the temperature of a stator winding of a fan generator is measured as a single measuring point of an experiment, and data simulation is adopted in the experiment and compared with an original data operation result because abnormal temperature faults do not occur.
A verification step:
1. collecting 10-month operation data of the wind field; adding random noise and trend noise into the generator stator winding temperature collected in the last month manually, and taking the random noise and the trend noise as abnormal data simulation;
2. the method comprises the following steps of (1) manually constructing a normal behavior model for predicting the temperature of a generator stator winding by taking load parameters, environmental parameters, temperatures of other reference measuring points and the like as input parameters;
3. for a generator stator winding, a regression algorithm based on a tree model is directly adopted, feature screening is carried out, the used input features mainly comprise engine bearing temperature, generator rotating speed, current, generator active power and the like, grid searching is used for optimizing the super parameters of the XGboost model, and the optimal feature weight, the maximum depth of the tree, the learning rate and the like are determined.
4. Selecting the variables by adopting data before the last month, and performing regression fitting based on a model on the equation in the step 3;
5. and observing the fitting precision, enabling the residual errors of the test set to obey normal distribution, and storing the model. Determining a probability density function (p.d.f) estimated by the residual distribution of the stator winding of the generator according to the distribution condition of the residual of the test set, and obtaining a mean value 0 and an estimated variance 0.2 of the residual; dividing n _ sigma (n sigma) intervals of residual distribution according to the steps, and setting gradient weight terms for different intervals to represent the sensitivity degree to each level of faults;
6. carrying out online verification by using data of the fan in the first half year of failure;
7. taking one day as a unit, running the algorithm once every day, inputting data of the day once, and outputting a risk value and an alarm grade of the running once per running;
8. for the stator winding of the generator, the sampling frequency is 5 seconds, the number of sample points input at one time is 17280, the window width of each sample point is 360 points, the step length of each sample point is 100 points, and 170 small windows are constructed by sliding windows. Counting the number of sample points falling into each residual division area in all samples for the samples in each small window, constructing four levels of health values HI in the corresponding small window by combining the weight items, and judging to obtain an output result of each small window (0/1);
9. counting the health values of all the small windows, and if the health index output result of the whole window exceeding half of the window is 1, meeting the triggering early warning condition;
10. if the early warning is generated in three continuous operations or more than half of the ten operations are generated in the last operation, the last operation triggers the alarm;
11. determining the risk value of the alarm according to the alarm frequency accumulated in history, and mapping the risk value to the alarm level, wherein the risk value is more than 40 serious;
12. in the online verification, the time length is two months. In the last month of fault data simulation, the algorithm continuously generates alarms with more than 40 levels which are continuously increased from the first simulation day, the control group test 1 is used for early warning by directly using an n _ sigma threshold value for residual errors, and a plurality of false alarms occur in the first month without faults; control test 2 was an algorithm executed using raw data and no alarm occurred within two months. The method is proved to predict the fault in advance, accurately and stably, and achieve the effect of preventing the fault in the bud.
Although the embodiments disclosed in the present application are described above, the descriptions are only for the convenience of understanding the present application, and are not intended to limit the present application. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the disclosure as defined by the appended claims.
Claims (7)
1. A fan component temperature anomaly detection and alarm method with single measuring point, the said fan component has single temperature measuring point, characterized by that, the said detection and alarm method includes mechanism driving data modeling procedure and part temperature early warning procedure based on residual error;
the mechanism driven data modeling program includes the steps of:
(1) acquiring monitoring data of a fan in operation, and selecting data of a group of equipment in an early stage without faults of a component to be tested as training data;
(2) screening out original variables related to the temperature of the component according to the physical characteristics of the component to be tested;
(3) preprocessing the original variable to obtain a new processing variable, and constructing a multivariate linear equation for describing the temperature of the component according to the original variable and the processing variable;
(4) dividing the acquired historical data into a training set and a test set, wherein the training set is used for fitting a temperature prediction model, and the test set is used for determining a threshold value of a subsequent step;
(5) fitting the data of the temperature measuring points of the component to be measured by utilizing the training set to obtain the coefficient of a multivariate linear equation, and constructing a temperature fitting model to obtain a fitting model of the measured values of the temperature measuring points of the component;
(6) predicting the temperature measuring points of the test set data by using the fitting model to obtain model predicted temperature; meanwhile, the fitting precision of the model is judged, and the fitting model is stored when the accuracy requirement is met;
(7) using the difference value of the actual temperature of the test data and the model predicted temperature as a predicted residual sequence;
(8) for the single-measuring-point component, obtaining a group of residual errors, and storing the residual errors for online prediction;
(9) during online operation, collecting real-time operation data of the wind turbine generator with a fixed time length T, and extracting features of the online data according to the method in the step (3);
(10) predicting the temperature measuring point of the online data through the extracted features by using the model to obtain the predicted temperature of the model;
(11) using the difference value between the real-time temperature of the part to be measured and the model predicted temperature as a predicted residual sequence to obtain a group of online residuals;
(12) entering the part temperature early warning program based on the statistical analysis of the obtained group of online residual error values;
wherein the residual-based component temperature early warning program comprises the following steps:
(1) recording the total number N of sample points of the online data in the time period T; for one-point components, firstFirstly, carrying out normal fitting on the residual errors of the test set by a statistical method to obtain the variance estimation sigma after fitting 2 ；
(2) Adding a weight term lambda to an n sigma interval of the fitted normal probability density curve, wherein the closer the weight is to the center of the mean value, the smaller the weight is, and the more the weight deviates from the center of the mean value, the larger the weight is; interactively constructing a health index HI according to the weight items and the number of probability intervals of the online residual error samples falling to the right
λ k Representing a set weight term, n k The number of probability intervals when the online residual error samples fall to the right side is represented, and i is 1, 2, 3 and 4;
(3) processing the online residual error by sliding window, calculating four HIs for the sample points in each small window, and counting the HIs 1 ，HI 2 ，HI 3 ，HI 4 A value of (d); if HI is in the time window 3 +HI 4 Greater than HI 1 +HI 2 Then, the time window is marked as W 1 Otherwise, it is recorded as W 0 ；
For W in all time windows 0 And W 1 Making statistics if most of the classes are W 1 If so, judging that the temperature abnormal fault occurs in the fan component within the time T, and outputting an alarm identification parameter as true.
2. The method for detecting and alarming the temperature abnormality of the fan component with the single measuring point as claimed in claim 1, wherein all warning identification parameters of the measuring point of the component to be measured are output to an algorithm operation result database for storage, and the warning identification parameter values stored in history are called for logic judgment in each algorithm operation; if the warning identification parameters output by the algorithm operation for three times continuously are true or the warning identification parameters in the historical n-time operation are true and exceed half, the operation triggers an alarm; wherein the size of n is determined according to the running frequency of the algorithm.
3. The method for detecting and alarming the temperature abnormality of the fan component with the single measuring point as claimed in claim 2, wherein when an alarm is triggered, the alarm identification parameter of the corresponding measuring point is set to be true, the alarm identification parameter corresponding to the measuring point is output to an algorithm operation result database for storage, the historically stored alarm identification parameter value is called for logic judgment in each algorithm operation, if the alarm identification parameter of the corresponding measuring point in the current operation is true, the number of the alarm identification parameters in the historical k operations is taken to be true to determine the risk value of the current alarm, and the proportion is mapped to the alarm level; the size of k is determined according to the running frequency of the algorithm, and k > n.
4. The method for detecting and alarming the temperature abnormality of the fan component with the single-point according to claim 3, wherein after the fan component is overhauled, a historical early warning library of the corresponding component is cleared.
5. The method for detecting and alarming temperature abnormality of fan component with single-measuring point according to any one of claims 1-4, characterized in that the original variables in the step (2) of the mechanism-driven data modeling program include at least one of wind speed, active power, generator speed, measured values of temperature measuring point and cabin temperature.
6. The method for detecting and alarming temperature abnormality of fan component with single-measuring point according to any one of claims 1-4, characterized in that in the step (6) of the mechanism-driven data modeling program, the judgment standard of the fitting accuracy of the model is performed by residual analysis and fitting accuracy analysis; and if the fitting precision is less than 10% of the overall data mean value, and the residual error obeys the normality distribution through the QQ-Norm test and the Jarqe-Bera test, the requirement of the accuracy is considered to be met.
7. The method for detecting and alarming temperature abnormality of fan component with single-point according to claim 5, characterized in that in the step (6) of the mechanism driven data modeling program, the judgment criterion of the fitting accuracy of the model is performed by residual analysis and fitting accuracy analysis; and if the fitting precision is less than 10% of the average value of the whole data, and the residual error is subjected to the normality distribution through a QQ-Norm test and a Jarqe-Bera test, the requirement of accuracy is met.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910702600.5A CN110414155B (en) | 2019-07-31 | 2019-07-31 | Fan component temperature abnormity detection and alarm method with single measuring point |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910702600.5A CN110414155B (en) | 2019-07-31 | 2019-07-31 | Fan component temperature abnormity detection and alarm method with single measuring point |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110414155A CN110414155A (en) | 2019-11-05 |
| CN110414155B true CN110414155B (en) | 2022-09-30 |
Family
ID=68364808
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910702600.5A Active CN110414155B (en) | 2019-07-31 | 2019-07-31 | Fan component temperature abnormity detection and alarm method with single measuring point |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110414155B (en) |
Families Citing this family (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110991666B (en) * | 2019-11-25 | 2023-09-15 | 远景智能国际私人投资有限公司 | Fault detection method, training device, training equipment and training equipment for model, and storage medium |
| CN110992205A (en) * | 2019-11-28 | 2020-04-10 | 中国船舶重工集团海装风电股份有限公司 | State detection method and system for generator winding of wind turbine generator and related components |
| CN111075661B (en) * | 2019-12-25 | 2021-11-09 | 明阳智慧能源集团股份公司 | Method for judging health condition of main shaft bearing of wind turbine generator based on temperature change trend |
| CN111581072B (en) * | 2020-05-12 | 2023-08-15 | 国网安徽省电力有限公司信息通信分公司 | Disk fault prediction method based on SMART and performance log |
| US11920562B2 (en) | 2020-06-04 | 2024-03-05 | Vestas Wind Systems A/S | Temperature estimation in a wind turbine |
| CN111766514B (en) * | 2020-06-19 | 2023-03-14 | 南方电网调峰调频发电有限公司 | Data analysis method for equipment detection point |
| CN111781498B (en) * | 2020-06-19 | 2023-03-14 | 南方电网调峰调频发电有限公司 | Data analysis system of equipment detection point |
| CN111814848B (en) * | 2020-06-22 | 2024-04-09 | 浙江大学 | Self-adaptive early warning strategy design method for temperature faults of wind turbine generator |
| CN111878320B (en) * | 2020-07-15 | 2021-07-30 | 上海电气风电集团股份有限公司 | Monitoring method and system of wind generating set and computer readable storage medium |
| CN112211794B (en) * | 2020-09-02 | 2023-01-06 | 五凌电力有限公司新能源分公司 | Cabin temperature abnormity early warning method, device, equipment and medium of wind turbine generator |
| CN112556130A (en) * | 2020-12-11 | 2021-03-26 | 青岛海尔空调器有限总公司 | Air conditioner alarm control method and device, electronic equipment and storage medium |
| CN112734130B (en) * | 2021-01-21 | 2022-06-10 | 河北工业大学 | Fault early warning method for double-fed fan main shaft |
| CN113221453A (en) * | 2021-04-30 | 2021-08-06 | 华风数据(深圳)有限公司 | Fault monitoring and early warning method for output shaft of gearbox of wind turbine generator |
| CN113325824B (en) * | 2021-06-02 | 2022-10-25 | 三门核电有限公司 | Regulating valve abnormity identification method and system based on threshold monitoring |
| CN113588123B (en) * | 2021-07-29 | 2023-01-06 | 东方电气集团东方电机有限公司 | Stator winding temperature early warning method |
| CN114151291B (en) * | 2021-11-18 | 2024-06-18 | 华能新能源股份有限公司 | Early fault monitoring method for wind turbine generator |
| CN114186666B (en) * | 2021-11-29 | 2023-10-13 | 中电华创(苏州)电力技术研究有限公司 | Generator coil temperature anomaly monitoring method based on self-standardized encoding and decoding |
| CN114251238A (en) * | 2021-11-30 | 2022-03-29 | 北京金风慧能技术有限公司 | Variable pitch motor temperature anomaly detection method and equipment |
| CN116167250B (en) * | 2023-04-23 | 2023-07-07 | 南京群顶科技股份有限公司 | Machine room environment assessment method based on temperature difference weighting and time sequence algorithm |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107977508A (en) * | 2017-11-29 | 2018-05-01 | 北京优利康达科技股份有限公司 | A kind of dynamo bearing failure prediction method |
| CN108680358A (en) * | 2018-03-23 | 2018-10-19 | 河海大学 | A kind of Wind turbines failure prediction method based on bearing temperature model |
| WO2018228648A1 (en) * | 2017-06-14 | 2018-12-20 | Kk Wind Solutions A/S | Independent monitoring system for a wind turbine |
-
2019
- 2019-07-31 CN CN201910702600.5A patent/CN110414155B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018228648A1 (en) * | 2017-06-14 | 2018-12-20 | Kk Wind Solutions A/S | Independent monitoring system for a wind turbine |
| CN107977508A (en) * | 2017-11-29 | 2018-05-01 | 北京优利康达科技股份有限公司 | A kind of dynamo bearing failure prediction method |
| CN108680358A (en) * | 2018-03-23 | 2018-10-19 | 河海大学 | A kind of Wind turbines failure prediction method based on bearing temperature model |
Non-Patent Citations (5)
| Title |
|---|
| Anomaly Detection and Fault Prognosis for Bearings;Xiaohang Jin等;《 IEEE Transactions on Instrumentation and Measurement》;20160606;全文 * |
| 基于Hadoop平台的风机群落故障预警;姚万业等;《电力科学与工程》;20180628(第06期);全文 * |
| 基于相关主成分分析和极限学习机的风电机组主轴承状态监测研究;何群等;《计量学报》;20180122(第01期);全文 * |
| 基于轴承温度模型的风电机组故障预测研究;丁佳煜等;《可再生能源》;20180220(第02期);全文 * |
| 采用预测模型与模糊理论的风电机组状态参数异常辨识方法;孙鹏等;《电力自动化设备》;20170810(第08期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110414155A (en) | 2019-11-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110414155B (en) | Fan component temperature abnormity detection and alarm method with single measuring point | |
| CN110414154B (en) | Fan component temperature abnormity detection and alarm method with double measuring points | |
| CN109298697B (en) | Method for evaluating working state of each part of thermal power plant system based on dynamic baseline model | |
| CN108763729A (en) | Process industry Mechatronic Systems couple state appraisal procedure based on network structure entropy | |
| CN107844067B (en) | A kind of gate of hydropower station on-line condition monitoring control method and monitoring system | |
| CN115409131B (en) | Production line abnormity detection method based on SPC process control system | |
| CN112785091A (en) | Method for performing fault prediction and health management on oil field electric submersible pump | |
| CN114239377A (en) | Method and system for evaluating health state of urban rail electromechanical equipment and storage medium | |
| CN111103137A (en) | Wind turbine gearbox fault diagnosis method based on deep neural network | |
| CN116914917A (en) | Big data-based monitoring and management system for operation state of power distribution cabinet | |
| CN110580492A (en) | Track circuit fault precursor discovery method based on small fluctuation detection | |
| CN110969185A (en) | Equipment abnormal state detection method based on data reconstruction | |
| CN112393906A (en) | Method for diagnosing, classifying and evaluating health of weak signal fault of bogie bearing of metro vehicle | |
| CN105910806A (en) | Filling pump early health status monitoring method | |
| CN117365869A (en) | Self-adaptive early warning strategy design method for wind turbine blade tower sweeping faults | |
| CN109299201B (en) | Power plant production subsystem abnormity monitoring method and device based on two-stage clustering | |
| CN112016193B (en) | Online prediction method and system for lubrication failure of shield tunneling machine system | |
| CN112633331A (en) | Self-encoder-based thermal power plant auxiliary machine health state assessment method | |
| CN114492636B (en) | Transformer winding state signal acquisition system | |
| Zhang | Comparison of data-driven and model-based methodologies of wind turbine fault detection with SCADA data | |
| CN115326393A (en) | Wind turbine generator bearing pair fault diagnosis method based on temperature information | |
| CN112308391A (en) | Real-time monitoring and anomaly detection method for equipment state based on neural network | |
| RU2648413C1 (en) | Method of the modes control based on the neural network diagnostics of faults and technical condition of the electric-drive gas-compressor unit | |
| Bingjie et al. | An improved similarity-based prognostics method for remaining useful life estimation of aero-engine | |
| CN114151291B (en) | Early fault monitoring method for wind turbine generator |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |