CN115204220A - Blade state detection method and device - Google Patents
Blade state detection method and device Download PDFInfo
- Publication number
- CN115204220A CN115204220A CN202210742692.1A CN202210742692A CN115204220A CN 115204220 A CN115204220 A CN 115204220A CN 202210742692 A CN202210742692 A CN 202210742692A CN 115204220 A CN115204220 A CN 115204220A
- Authority
- CN
- China
- Prior art keywords
- data
- representation
- health
- real
- blade
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- 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
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/70—Wind energy
- Y02E10/72—Wind turbines with rotation axis in wind direction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application provides a blade state detection method and device, relates to the field of operation and maintenance of wind driven generators, and aims to accurately monitor and evaluate the state of a blade in real time. The method comprises the following steps: acquiring real-time measurement data, wherein one real-time measurement data comprises data of each dimension sampled at each sampling time point in a sampling time range; inputting the real-time measurement data into a mapping network to obtain data representation of the real-time measurement data, wherein the mapping network is obtained by training with healthy data as a training sample, and the healthy data refers to measurement data acquired when the blade is in a healthy state; obtaining a data representation of the health data, the data representation of the health data being obtained using the mapping network; determining a state of the blade based on a similarity between the data representation of the real-time measurement data and the data representation of the health data.
Description
Technical Field
The application relates to the field of operation and maintenance of wind driven generators, in particular to a blade state detection method and device.
Background
Wind power generation is one of the renewable energy utilization modes with mature technology and wide prospect at present, and the vigorous development of wind power generation has important significance for relieving energy crisis and solving environmental problems. The wind driven generator is important equipment for wind power generation, and comprises a blade, a tower, a traditional system, a power generation system, a variable pitch system, a yaw system and the like. Among all the components of the wind driven generator, the blade can be regarded as an important pilot component, which is directly contacted with external wind, and converts wind energy into mechanical energy by utilizing the aerodynamic shape, so as to drive a transmission system and a power generation system to operate, and finally converts the mechanical energy into electric energy, therefore, the service state of the blade has a decisive influence on the operation condition of the whole wind driven generator.
The blades are large in size, complex in structure and high in failure rate, and the shutdown time caused by the blade failure is long, so that the great loss of generated energy can be caused, and the blade fracture can be even caused by the serious blade failure, so that casualties are caused, and the important threat of production safety is realized. At present, fault prevention of the wind driven generator blade still depends on manual regular inspection, but the mode is high in labor intensity and dangerous in working environment, and the regular inspection mode cannot be used for monitoring the state of the blade in real time during operation, so that the hysteresis is high. Therefore, a method for detecting the state of the blade in real time is needed, which can detect the symptom of the blade fault as soon as possible, avoid the catastrophic consequences caused by serious faults and ensure the efficiency and safety of wind power generation.
At present, a supervisory control and data acquisition (SCADA) system is installed in most wind driven generators, the operating conditions of the wind driven generators are monitored through various sensors such as temperature, pressure and vibration, and control links such as pitch control and yaw are remotely implemented, so that SCADA measurement data provide a data base for monitoring the state of the wind driven generator blades, the health conditions of the blades can be identified through a data-driven mode identification method, the data driving method based on a large amount of historical measurement data can reduce the dependence on expert knowledge related to a failure mechanism, and the SCADA system is beneficial to mastering and application of wind driven generation technicians. However, the wind turbine with a blade fault runs for a long time, so that the SCADA measurement data under the fault condition is rare relative to the health measurement data, and the lack of the fault measurement data causes that the data driving model is difficult to learn information related to the fault state in the training process, and the separability of the measurement data under different health states cannot be ensured, thereby affecting the judgment effect of the health condition of the blade. Therefore, the interference of the scarcity of the blade fault measurement data on the measurement data is not considered in the related technology, and the health condition of the wind driven generator blade cannot be accurately monitored and evaluated in real time according to the SCADA measurement data.
Disclosure of Invention
In view of the above, embodiments of the present invention provide a blade condition detection method and apparatus to overcome the above problems or at least partially solve the above problems.
In a first aspect of the embodiments of the present invention, a method for detecting a blade state is provided, including:
acquiring real-time measurement data, wherein one real-time measurement data comprises data of each dimension sampled at each sampling time point in a sampling time range;
inputting the real-time measurement data into a mapping network to obtain data representation of the real-time measurement data, wherein the mapping network is obtained by training with healthy data as a training sample, and the healthy data refers to measurement data acquired when the blade is in a healthy state;
obtaining a data representation of the health data, the data representation of the health data being obtained using the mapping network;
determining a state of the blade based on a similarity between the data representation of the real-time measurement data and the data representation of the health data.
Optionally, said determining the state of the blade from the similarity between the data representation of the real-time measurement data and the data representation of the health data comprises:
calculating a mean vector and a covariance matrix of a data representation of the health data;
calculating a first distance of a data representation of the real-time measurement data and a data representation of the health data from the data representation of the real-time measurement data, and the mean vector and the covariance matrix;
determining a state of the blade based on the first distance.
Optionally, the method further comprises:
obtaining a second distance between each of a plurality of data representations of historical measurement data and a data representation of the health data;
establishing an exponential weighted moving average control chart according to a plurality of second distances;
determining the early warning distance according to the exponentially weighted moving average control chart;
the determining the state of the blade according to the first distance comprises:
determining that the blade is in a healthy state if the first distance is not greater than the early warning distance;
determining that the blade is in an unhealthy state if the first distance is greater than the early warning distance.
Optionally, the mapping network is trained by the following steps:
acquiring a plurality of said health data;
performing two random data enhancement treatments on each health data to obtain two different enhancement data corresponding to each health data;
inputting two different enhancement data corresponding to each health data into a mapping network to be trained to obtain data representation of each enhancement data;
and carrying out self-supervision training on the mapping network to be trained according to the data representation of each piece of enhancement data to obtain the mapping network.
Optionally, the obtaining a plurality of the health data comprises:
while the blade is in a healthy state, acquiring, for each of the dimensions, sampled-to-data for each of a plurality of sampling time points within the sampling time range;
filtering invalid data in the data to obtain a plurality of residual data of each dimension in a plurality of sampling time ranges;
eliminating the influence of the data magnitude of a plurality of residual data of each dimension in a plurality of sampling time ranges to obtain a plurality of target data of each dimension in the plurality of sampling time ranges;
and stacking a plurality of target data of each dimension in each sampling time range according to each sampling time point in each sampling time range to obtain a plurality of health data.
Optionally, the performing, according to the data representation of each enhanced data, an auto-supervised training on the mapping network to be trained to obtain the mapping network includes:
inputting the data representation of each enhancement data into a projection network to obtain a projection data representation of each enhancement data;
inputting the projection data representation of each enhanced data into a prediction network to obtain a prediction result of whether each two enhanced data are enhanced data of the same health data;
establishing a loss function based on a plurality of said predictions and a projection data representation of each said enhancement data;
and training the mapping network to be trained based on the loss function to obtain the mapping network.
Optionally, the establishing a loss function according to a plurality of the prediction results and the projection data representation of each of the enhancement data comprises:
calculating the similarity of negative cosine according to a plurality of prediction results and the projection data representation of each enhancement data;
and establishing the loss function according to the negative cosine similarity.
Optionally, the random two data enhancement processes include, but are not limited to, any two of: amplitude axis overturning, variable axis overturning, cutting, noise adding, cutting adjustment, time packaging and smoothing.
In a second aspect of the embodiments of the present invention, there is provided a blade state detection apparatus, including:
the data acquisition module is used for acquiring real-time measurement data, and one piece of real-time measurement data comprises data of each dimension sampled at each sampling time point in a sampling time range;
the first data representation module is used for inputting the real-time measurement data into a mapping network to obtain data representation of the real-time measurement data, wherein the mapping network is obtained by training with healthy data as a training sample, and the healthy data refers to measurement data acquired when the blade is in a healthy state;
a second data representation module for obtaining a data representation of the health data, the data representation of the health data being obtained using the mapping network;
a state determination module to determine a state of the blade based on a similarity between the data representation of the real-time measurement data and the data representation of the health data.
Optionally, the mapping network is trained by the following steps:
acquiring a plurality of said health data;
performing two random data enhancement treatments on each health data to obtain two different enhancement data corresponding to each health data;
inputting two different enhancement data corresponding to each health data into a mapping network to be trained to obtain data representation of each enhancement data;
and performing self-supervision training on the mapping network to be trained according to the data representation of each enhanced data to obtain the mapping network.
The embodiment of the invention has the following advantages:
in this embodiment, real-time measurement data may be acquired, where one real-time measurement data includes data of each dimension sampled at each sampling time point within a sampling time range; inputting the real-time measurement data into a mapping network to obtain data representation of the real-time measurement data, wherein the mapping network is obtained by training with healthy data as a training sample, and the healthy data refers to measurement data acquired when the blade is in a healthy state; obtaining a data representation of the health data, the data representation of the health data being obtained using the mapping network; determining a state of the blade based on a similarity between the data representation of the real-time measurement data and the data representation of the health data. Therefore, the mapping network only needs to output data representation of the data without directly judging the state of the blade, so that the mapping network can be trained by only utilizing the health data, and the defect that the information related to the fault state is difficult to learn in the training process of a model due to scarcity of fault measurement data is avoided; by comparing the similarity of the data representation of the real-time measurement data and the data representation of the health data generated by the same mapping network, the similarity of the current state and the health state of the blade, and thus the state of the blade, may be determined. In addition, the real-time measurement data can be directly obtained without manual regular inspection, and the data representation of the real-time measurement data is rapidly generated through the mapping network, so that the state of the blade can be rapidly determined, and the state of the blade can be accurately monitored and evaluated in real time.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the description of the embodiments of the present application will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without inventive exercise.
FIG. 1 is a flow chart of steps of a method of detecting a blade condition in accordance with an embodiment of the present invention;
FIG. 2 is a block diagram of the components of a mapping network, a projection network, and a prediction network in an embodiment of the present invention;
FIG. 3 is a schematic flow chart of a blade condition detection method according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a blade state detection device according to an embodiment of the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present application more comprehensible, the present application is described in further detail with reference to the accompanying drawings and the detailed description.
Referring to fig. 1, a flowchart illustrating steps of a blade state detection method according to an embodiment of the present invention is shown, and as shown in fig. 1, the blade state detection method may specifically include the following steps:
step S11: real-time measurement data is acquired, and one real-time measurement data comprises data of all dimensions sampled at all sampling time points in a sampling time range.
The real-time measurement data is data collected by the SCADA system, and the data collected by the SCADA system can be stored. One sampling channel of the SCADA system samples data of one dimension. The dimensions include at least any one or more of: a wind speed dimension, a wind direction dimension, an ambient temperature dimension, a generator speed dimension, a generated power dimension, a lubricant temperature dimension, and the like. A real-time measurement data includes data for each dimension sampled at each sampling time point within a last sampling time range. Optionally, data of each dimension sampled at each sampling time point in a sampling time range may be stacked to obtain a two-dimensional matrix, and the two-dimensional matrix is a real-time measurement data. One of the row dimension and the column dimension of the two-dimensional matrix may characterize the data sampled at each sampling time point, and the other may characterize the data in each dimension.
For example, one sampling time range is half an hour, and 10 minutes are respectively separated between sampling time points, i.e., 10 th, 20 th and 30 th minutes in one sampling time range are respectively sampled once; at one sampling time point, each sampling channel of the SCADA system will take one sample. If there are i channels, 3 × i data can be collected within a sampling time range, and the 3 × i data together form a real-time measurement data. The state of the blade is usually gradually changed, and the change of the state of the blade in a sampling time range is very little, so that although a real-time measurement data is actually data from the end of the last sampling time range (the start of the current sampling time range) to the end of the current sampling time range in the past 30 minutes, the real-time measurement data can be regarded as real-time data. Optionally, the size of the sampling time range and the interval of the sampling time points may be set according to actual requirements.
Because the data magnitude of each dimension is different, normal normalization processing can be performed on the data of each dimension in order to eliminate the influence of the data magnitude. The normal normalization process can be performed by the following equation:
wherein x is i Data representing an ith dimension; i =1,2, … … I; i represents the total dimension number; the superscript "0" represents the raw data prior to normal normalization, μ represents the mean, and σ represents the standard deviation.
Step S12: inputting the real-time measurement data into a mapping network to obtain data representation of the real-time measurement data, wherein the mapping network is obtained by training with healthy data as a training sample, and the healthy data refers to measurement data acquired when the blade is in a healthy state.
The health data also includes data for each dimension sampled at each sampling time point within a sampling time range. The health data may be collected via a SCADA system. The mapping network obtained by training with the health data as the training sample can accurately output the data representation of each measurement data according to the input measurement data. Real-time measurement data is input into a pre-trained mapping network, and the mapping network can directly output data representation of the real-time measurement data. Alternatively, the data representation of the data may be a vector.
Step S13: obtaining a data representation of the health data, the data representation of the health data being obtained using the mapping network.
The data representation of the health data may be derived by inputting the health data into the mapping network after training the mapping network. After the data representation of the health data is obtained, it may be stored. When the state of the blade needs to be judged according to the real-time measurement data, the data representation of the stored health data is directly obtained.
Step S14: determining a state of the blade based on a similarity between the data representation of the real-time measurement data and the data representation of the health data.
The similarity of the representation of the real-time measurement data and the data representation of the health data may be determined by calculating the distance between the two. The higher the similarity between the data representation of the real-time measurement data and the data representation of the health data, the higher the similarity of the state of the blade to the health state is demonstrated, and thus the state of the blade can be determined from the similarity. The blade may be a blade of a wind turbine.
By adopting the technical scheme of the embodiment of the application, the mapping network only needs to output data representation of the data without directly judging the state of the blade, so that the mapping network can be trained only by utilizing the health data, and the defect that the information related to the fault state is difficult to learn in the training process of the model due to the scarcity of fault measurement data is avoided; by comparing the similarity of the data representation of the real-time measurement data and the data representation of the health data generated by the same mapping network, the similarity of the current state and the health state of the blade, and thus the state of the blade, may be determined. In addition, the real-time measurement data can be directly acquired without manual regular inspection, and the data representation of the real-time measurement data is rapidly generated through a mapping network, so that the state of the blade can be rapidly determined, and the state of the blade can be accurately monitored and evaluated in real time.
Alternatively, a mean vector and a covariance matrix of the data representation of the health data may be calculated, and the first distance of the data representation of the real-time measurement data and the data representation of the health data may be calculated based on the data representation of the real-time measurement data and the mean vector of the data representation of the health data and the covariance matrix of the data representation of the health data. The first distance may be a mahalanobis distance. The similarity between the data representation of the real-time measurement data and the data representation of the health data, and thus the state of the blade, may be determined based on the first distance of the data representation of the real-time measurement data and the data representation of the health data.
The first distance of the data representation of the real-time measurement data and the data representation of the health data may be determined by the following formula:
wherein S is a first distance, y r Data representation characterizing real-time measurement data, T being the total number of sampling time points, μ h Mean vector of data representation, Σ, characterizing health data h A covariance matrix of data representations characterizing the health data.
In this way, the mean vector of the data representation of the healthy data may integrate the features of each data, and the covariance matrix of the data representation of the healthy data may characterize the correlation of the data of each dimension. Therefore, the similarity between the data representation of the real-time measurement data and the data representation of the health data can be accurately measured according to the first distance calculated by the mean vector and the covariance matrix of the data representation of the health data, and the state of the blade can be accurately determined.
Optionally, on the basis of the above technical solution, when determining whether the blade is healthy according to the first distance, the first distance may be compared with the warning distance. Determining that the blade is in a healthy state under the condition that the first distance is not greater than the early warning distance; and determining that the blade is in an unhealthy state under the condition that the first distance is greater than the early warning distance.
The forewarning distance may be an exponentially weighted moving average control map established from a plurality of second distances, wherein the second distances are distances between the data representation of the historic measurement data and the data representation of the health data. The method for obtaining the second distance between the data representation of the historical data and the data representation of the health data may refer to the method for obtaining the first distance, and is not described herein again.
An exponentially weighted moving average control map can be established by the following equation:
EWMA(t)＝γS(t)+(1-γ)EWMA(t-1)
wherein t =1,2, … …; s (t) characterizing a second distance between the data representation of the tth historical measurement data and the data representation of the health data; EWMA (t) characterizes the statistics of the exponentially weighted moving average control chart, EWMA (0) represents the average of S (t); gamma is 0< gamma.ltoreq.1 as a memory parameter for controlling the short-term and long-term memory weights.
The upper control limit of the early warning distance can be determined by the following formula:
UCL＝EWMA(0)+3σ EWMA
wherein, UCL represents the upper control limit of the early warning distance, sigma EWMA Representing the standard deviation of the variance in the EWMA statistic.
In this way, the information of all historical measurement data can be combined through the exponentially weighted moving average control chart, and the exponentially weighted moving average control chart is not seriously influenced when a smaller value or a larger value enters the calculation. By varying the weights used and the amount of historical measurement data that control limits, a control map can be constructed that can detect deviations of almost any magnitude in the process, and therefore can detect smaller deviations, and thus accurately determine the state of the blade.
Optionally, on the basis of the above technical solution, the mapping network may be obtained by training through the following steps:
step S21: a plurality of the health data is acquired.
The plurality of health data may be obtained by: acquiring, for each of the dimensions, sample-to-data for each of a plurality of sampling time points within the sampling time range while the blade is in a healthy state; filtering invalid data in the data to obtain a plurality of residual data of each dimension in a plurality of sampling time ranges; eliminating the influence of the data magnitude of a plurality of residual data of each dimension in a plurality of sampling time ranges to obtain a plurality of target data of each dimension in the plurality of sampling time ranges; and stacking a plurality of target data of each dimension in each sampling time range according to each sampling time point in each sampling time range to obtain a plurality of health data.
Invalid data refers to data when the wind driven generator is operated at limited power and is outside a cut-in-cut-out wind speed range.
The influence of the data magnitude means that the data of each dimension may have different value ranges, for example, the value range of data of one dimension may be 0 to 10, and the value range of data of another dimension may be 0 to 1000. In order to eliminate the influence of the data magnitude, normal normalization processing can be performed on the data of each dimension, so that the value range of the data of each dimension is 0 to 1, the processed data of each dimension conforms to normal distribution, and the influence of the data of different data magnitudes on the data representation of the health data is avoided.
After the influence of the data magnitude is eliminated, target data can be obtained. And stacking a plurality of target data of each dimension in a sampling time range according to sampling time points to obtain the health data. The health data may be a two-dimensional matrix, one of a row dimension and a column dimension of which may characterize the data sampled at each sampling time point and the other may characterize the data in each dimension.
Step S22: and performing random two-kind data enhancement processing on each health data to obtain two different enhancement data corresponding to each health data.
After the health data are obtained, two different data enhancement treatments are carried out on each health data, the data enhancement treatments carried out on each health data are random, and therefore two enhancement data corresponding to each health data are obtained, and the two enhancement data are different because the two enhancement data are carried out on different data enhancement treatments.
Alternatively, the data enhancement processing may be amplitude axis inversion, variable axis inversion, truncation, adding noise, cropping adjustments, time packing, and smoothing. Which two data enhancement processes are to be performed on each health data may be determined by generating two random integers of 1 through 7, one for each data enhancement process.
Wherein, the amplitude axis inversion is the inversion of the health data from left to right, and for the data x, the enhanced data with amplitude axis inversion is performed
Can be expressed as:
wherein A is 1 ＝[a m,n ] I×I Representing element a m,n The transformation matrix of (3), wherein:
the variable axis flipping is realized by changing labels of elements in data x, and the enhanced data is expressed as:
wherein
Represents a diagonal matrix and the diagonal elements are-1.
The purpose of truncation is to randomly set the elements in the data x to zero, with the proportion of the number of truncated elements to the total number being 0<η3<1, the enhancement data can be expressed as
Wherein
Represents a diagonal matrix, where int (I η) 3 ) The diagonal element is 0, [ I-int (I eta) ] 3 )]The diagonal element at 1,int () represents an integer function.
The noise addition is to add white Gaussian noise to the data x
The enhancement data may be represented as
Wherein σ a ＝σ x η 4 Standard deviation, σ, representing additive noise x Denotes the standard deviation, η, of the data x 4 >0 represents the relative noise intensity.
The clipping adjustment refers to the data x occupying the total element 0<η 5 <Part of 1 is cut off randomly, then the cut data is sampled again by linear interpolation, so that the enhanced data
The same length as the original data x.
Time packing is a special resampling method, where elements in the data x are randomly discarded or interpolated. And the number of discarded data points and interpolated data points is the same in order to ensure that the length of the augmented sample is constant. In particular, it can be said that one decision vector is generated by the element-1,0,1
The proportion of the elements 0 and 1 to the total number of the elements is 0<η 6 <0.5. Augmented data
Medium decision vector a 6 The element in data x corresponding to element-1 in (a) is discarded 6 Element 0 in (b) is reserved for the element in x corresponding to element 0 in (b)To a, a 6 The element 1 in (b) is linearly interpolated with the element in (x). Its enhancement data can be expressed as:
and
the smoothing process is to filter out the high frequency components in the data x and then to combine the data x with the length eta 7 Sliding window a 7 By performing a convolution operation, enhanced data can be obtained, represented as
For this case, the data is augmented
The transient portion of (a) is suppressed and the general form of the data x is preserved.
Step S23: and inputting two different enhancement data corresponding to each health data into a mapping network to be trained to obtain data representation of each enhancement data.
3 convolutional neural networks containing convolutional layers, batch Normalization layers and nonlinear activation functions are constructed under a PyTorch (open source machine learning library) platform, wherein one convolutional neural network is a mapping network to be trained, and the other two convolutional neural networks can be a projection network and a prediction network respectively.
For each data input, the mapping network to be trained may derive a data representation of the data.
Step S24: and carrying out self-supervision training on the mapping network to be trained according to the data representation of each piece of enhancement data to obtain the mapping network.
The time-varying operating condition and the variable operating environment of the wind driven generator cause interference to the measured data, so that the measured data under the same health condition is also variable, the health condition of the wind driven generator blade is difficult to reflect by the measured data, and the accuracy of the state monitoring result of the wind driven generator blade is damaged.
According to the embodiment of the application, the mapping network is obtained through the enhanced data training of the data enhancement processing, so that the mapping network has stronger robustness. The mapping network can therefore overcome the interference caused by operating conditions and operating environments to the measured data, thereby generating an accurate data representation. Therefore, the data obtained based on the mapping network represents the judged state of the blade, and the method has high accuracy.
Optionally, on the basis of the above technical solution, in order to perform self-supervision training on the mapping network, the mapping network may be assisted by using a projection network and a prediction network for training.
Fig. 2 shows a structural diagram of the mapping network, the projection network and the prediction network, and the mapping network, the projection network and the prediction network are connected in series.
And after the data representation of the enhanced data output by the mapping network to be trained is obtained, inputting the data representation of the enhanced data into the projection network to obtain the projection data representation of each enhanced data. And (3) the projection data of each piece of enhancement data is input into a prediction network, and the prediction network can predict whether any two pieces of enhancement data are enhancement data of the same health data to obtain a prediction result.
Based on the prediction, and the projection data representation of each enhancement data, a loss function may be established. And training the mapping network to be trained based on the loss function to obtain the mapping network. The projection network and the prediction network are trained simultaneously while the mapping network to be trained is trained based on the loss function.
Alternatively, the negative cosine similarity may be calculated from the plurality of prediction results and the projection data representation of each enhancement data; and establishing a loss function according to the negative cosine similarity. The loss function L can be calculated by the following formula:
wherein, the first and the second end of the pipe are connected with each other,
is a negative cosine similarity, p 1 And p 2 A projection data representation of two enhancement data, q 1 And q is 2 Respectively outputting prediction results of the prediction network; τ () represents the gradient stopping counter-propagation during training.
Negative cosine similarity
Can be calculated by the following formula:
where p is the projection data representation of the enhancement data and q is the prediction result.
Based on a loss function, a small batch of gradient descent is adopted to carry out iterative training on a neural network so as to carry out self-supervision learning, the prediction among samples is enhanced through health data, information related to the health condition of the blades of the wind driven generator is extracted, and whether the training is finished or not is judged according to a preset condition. The training may be stopped when a preset number of training times is reached, or the training converges.
Optionally, 2-norm normalization processing may be performed on the data representation of the enhanced data obtained by the mapping network; the projection network projects the data representation of the enhanced data subjected to the 2-norm normalization processing to obtain the projection data representation of the enhanced data; performing 2 norm normalization processing on the projection data representation of the enhancement data; and (3) the projection data of the enhanced data subjected to the 2-norm normalization processing is input into a prediction network, and the prediction network determines whether the two enhanced data are the enhanced data of the same health data or not by calculating the cosine similarity of the data representation of the two input enhanced data.
Therefore, the mapping network can output accurate data representation, and the projection network can isolate the mapping network from the prediction network so as to prevent the mapping network from being coupled with the prediction network; the predictive network may generate predictive results to complete the self-supervised training.
Fig. 3 shows a flow chart of a blade condition detection method. The training of the mapping network can be performed off-line, and the detection of the blade state can be performed on-line in real time.
According to the embodiment of the application, the self-supervision learning mode is adopted in the training stage to extract effective representation from the SCADA measurement data to reflect the health condition of the wind driven generator blade, a state monitoring reference model can be established only by the measurement data from the healthy wind driven generator, the measurement data under the blade fault state is not needed, the problem that the measurement data under the blade fault state of the wind driven generator is scarce can be solved, the data-driven state monitoring can be favorably developed under the condition that the measurement data are limited, and therefore important reference information is provided for the operation and maintenance planning of the wind driven generator.
According to the embodiment of the application, the measurement data from the healthy wind driven generator blade is predicted by executing the self-supervision agent task, the distance of the healthy measurement data sample in the characteristic space is shortened, the separability of the healthy measurement data and the fault measurement data in the learned data representation is enhanced, the interference of working condition change and environmental noise to the measurement data is inhibited, the accuracy of the state monitoring result of the wind driven generator blade is improved, and the occurrence of catastrophic accidents caused by serious faults of the blade is avoided.
It should be noted that, for simplicity of description, the method embodiments are described as a series of acts or combination of acts, but those skilled in the art will recognize that the present invention is not limited by the illustrated order of acts, as some steps may occur in other orders or concurrently in accordance with the embodiments of the present invention. Further, those skilled in the art will appreciate that the embodiments described in the specification are presently preferred and that no particular act is required to implement the invention.
Fig. 4 is a schematic structural diagram of a blade state detection apparatus according to an embodiment of the present invention, and as shown in fig. 4, the blade state detection apparatus includes a data acquisition module, a first data representation module, a second data representation module, and a state determination module, where:
the data acquisition module is used for acquiring real-time measurement data, and one piece of real-time measurement data comprises data of each dimension sampled at each sampling time point in a sampling time range;
the first data representation module is used for inputting the real-time measurement data into a mapping network to obtain data representation of the real-time measurement data, wherein the mapping network is obtained by training by taking health data as a training sample, and the health data refers to the measurement data acquired when the blade is in a healthy state;
a second data representation module for obtaining a data representation of the health data, the data representation of the health data being obtained using the mapping network;
a state determination module to determine a state of the blade based on a similarity between the data representation of the real-time measurement data and the data representation of the health data.
Optionally, the state determination module includes:
a calculation unit for calculating a mean vector and a covariance matrix of a data representation of the health data;
a first distance calculation unit for calculating a first distance of the data representation of the real-time measurement data and the data representation of the health data according to the data representation of the real-time measurement data, and the mean vector and the covariance matrix;
and the state determining unit is used for determining the state of the blade according to the first distance.
Optionally, the method further comprises:
a second distance acquisition module for acquiring a second distance between each of the data representations of the plurality of historical measurement data and the data representation of the health data;
the control chart establishing module is used for establishing an exponential weighted moving average control chart according to the plurality of second distances;
the early warning distance determining module is used for determining the early warning distance according to the exponentially weighted moving average control chart;
the state determination unit includes:
the first determining subunit is used for determining that the blade is in a healthy state under the condition that the first distance is not greater than the early warning distance;
and the second determining subunit is used for determining that the blade is in an unhealthy state under the condition that the first distance is greater than the early warning distance.
Optionally, the mapping network is trained by the following steps:
acquiring a plurality of said health data;
performing two random data enhancement treatments on each health data to obtain two different enhancement data corresponding to each health data;
inputting two different enhancement data corresponding to each health data into a mapping network to be trained to obtain data representation of each enhancement data;
and carrying out self-supervision training on the mapping network to be trained according to the data representation of each piece of enhancement data to obtain the mapping network.
Optionally, the obtaining a plurality of the health data comprises:
while the blade is in a healthy state, acquiring, for each of the dimensions, sampled-to-data for each of a plurality of sampling time points within the sampling time range;
filtering invalid data in the data to obtain a plurality of residual data of each dimension in a plurality of sampling time ranges;
eliminating the influence of the data magnitude of a plurality of residual data of each dimension in a plurality of sampling time ranges to obtain a plurality of target data of each dimension in the plurality of sampling time ranges;
and stacking a plurality of target data of each dimension in each sampling time range according to each sampling time point in each sampling time range to obtain a plurality of health data.
Optionally, the performing, according to the data representation of each enhanced data, an auto-supervised training on the mapping network to be trained to obtain the mapping network includes:
inputting the data representation of each enhancement data into a projection network to obtain a projection data representation of each enhancement data;
inputting the projection data representation of each enhancement data into a prediction network to obtain a prediction result of whether each two enhancement data are enhancement data of the same health data;
establishing a loss function based on a plurality of said prediction results and a projection data representation of each said enhancement data;
and training the mapping network to be trained based on the loss function to obtain the mapping network.
Optionally, the establishing a loss function according to a plurality of the prediction results and the projection data representation of each of the enhancement data comprises:
calculating the similarity of negative cosine according to a plurality of prediction results and the projection data representation of each enhancement data;
and establishing the loss function according to the negative cosine similarity.
Optionally, the random two data enhancement processes include, but are not limited to, any two of: amplitude axis overturning, variable axis overturning, cutting, noise adding, cutting adjustment, time packaging and smoothing.
It should be noted that the apparatus embodiments are similar to the method embodiments, and therefore the description is simple, and reference may be made to the method embodiments for relevant points.
The embodiment of the invention also provides electronic equipment, which comprises a processor, a memory and a computer program which is stored on the memory and can run on the processor, wherein the processor executes the program to realize the blade state detection method disclosed by the embodiment of the application.
The embodiment of the invention also provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed, the blade state detection method disclosed by the embodiment of the application is realized.
Embodiments of the present invention further provide a computer program product, which includes a computer program or computer instructions, and when the computer program or the computer instructions are executed by a processor, the method for detecting a blade state as disclosed in the embodiments of the present application is implemented.
The embodiments in the present specification are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, apparatus or computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
Embodiments of the present invention are described with reference to flowchart illustrations and/or block diagrams of methods, apparatus, electronic devices and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing terminal to cause a series of operational steps to be performed on the computer or other programmable terminal to produce a computer implemented process such that the instructions which execute on the computer or other programmable terminal provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications of these embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the embodiments of the invention.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or terminal apparatus that comprises the element.
The method and the device for detecting the blade state provided by the present application are introduced in detail, and specific examples are applied in the present application to explain the principle and the implementation of the present application, and the description of the above embodiments is only used to help understand the method and the core idea of the present application; meanwhile, for a person skilled in the art, according to the idea of the present application, the specific implementation manner and the application scope may be changed, and in summary, the content of the present specification should not be construed as a limitation to the present application.
Claims (10)
1. A method of detecting a state of a blade, comprising:
acquiring real-time measurement data, wherein one real-time measurement data comprises data of each dimensionality sampled at each sampling time point in a sampling time range;
inputting the real-time measurement data into a mapping network to obtain data representation of the real-time measurement data, wherein the mapping network is obtained by training with healthy data as a training sample, and the healthy data refers to measurement data acquired when the blade is in a healthy state;
obtaining a data representation of the health data, the data representation of the health data being obtained using the mapping network;
determining a state of the blade based on a similarity between the data representation of the real-time measurement data and the data representation of the health data.
2. The method of claim 1, wherein determining the state of the blade based on the similarity between the data representation of the real-time measurement data and the data representation of the health data comprises:
calculating a mean vector and a covariance matrix of a data representation of the health data;
calculating a first distance of a data representation of the real-time measurement data and a data representation of the health data from the data representation of the real-time measurement data, and the mean vector and the covariance matrix;
determining a state of the blade based on the first distance.
3. The method of claim 2, further comprising:
obtaining a second distance between each of a plurality of data representations of historical measurement data and a data representation of the health data;
establishing an exponential weighted moving average control chart according to a plurality of second distances;
determining the early warning distance according to the exponentially weighted moving average control chart;
the determining the state of the blade according to the first distance comprises:
determining that the blade is in a healthy state when the first distance is not greater than the early warning distance;
determining that the blade is in an unhealthy state if the first distance is greater than the early warning distance.
4. The method of claim 1, wherein the mapping network is trained by:
acquiring a plurality of said health data;
performing two random data enhancement treatments on each health data to obtain two different enhancement data corresponding to each health data;
inputting two different enhancement data corresponding to each health data into a mapping network to be trained to obtain data representation of each enhancement data;
and performing self-supervision training on the mapping network to be trained according to the data representation of each enhanced data to obtain the mapping network.
5. The method of claim 4, wherein said obtaining a plurality of said health data comprises:
acquiring, for each of the dimensions, sample-to-data for each of a plurality of sampling time points within the sampling time range while the blade is in a healthy state;
filtering invalid data in the data to obtain a plurality of residual data of each dimension in a plurality of sampling time ranges;
eliminating the influence of the data magnitude of a plurality of residual data of each dimension in a plurality of sampling time ranges to obtain a plurality of target data of each dimension in the plurality of sampling time ranges;
and stacking a plurality of target data of each dimensionality in each sampling time range according to each sampling time point in each sampling time range to obtain a plurality of health data.
6. The method of claim 4, wherein the performing an auto-supervised training of the mapping network to be trained according to the data representation of each of the enhanced data to obtain the mapping network comprises:
inputting the data representation of each enhancement data into a projection network to obtain the projection data representation of each enhancement data;
inputting the projection data representation of each enhancement data into a prediction network to obtain a prediction result of whether each two enhancement data are enhancement data of the same health data;
establishing a loss function based on a plurality of said prediction results and a projection data representation of each said enhancement data;
and training the mapping network to be trained based on the loss function to obtain the mapping network.
7. The method of claim 6, wherein establishing a loss function based on the plurality of predictors and the projection data representation of each of the enhancement data comprises:
calculating the similarity of negative cosine according to a plurality of prediction results and the projection data representation of each enhancement data;
and establishing the loss function according to the negative cosine similarity.
8. The method according to any one of claims 4-7, wherein the random two data enhancement processes include, but are not limited to, any two of: amplitude axis overturning, variable axis overturning, cutting, noise adding, cutting adjustment, time packaging and smoothing.
9. A vane condition detection apparatus, comprising:
the data acquisition module is used for acquiring real-time measurement data, and one piece of real-time measurement data comprises data of each dimension sampled at each sampling time point in a sampling time range;
the first data representation module is used for inputting the real-time measurement data into a mapping network to obtain data representation of the real-time measurement data, wherein the mapping network is obtained by training with healthy data as a training sample, and the healthy data refers to measurement data acquired when the blade is in a healthy state;
a second data representation module for obtaining a data representation of the health data, the data representation of the health data being obtained using the mapping network;
a state determination module to determine a state of the blade based on a similarity between the data representation of the real-time measurement data and the data representation of the health data.
10. The apparatus of claim 9, wherein the mapping network is trained by:
acquiring a plurality of said health data;
performing two random data enhancement treatments on each health data to obtain two different enhancement data corresponding to each health data;
inputting two different enhancement data corresponding to each health data into a mapping network to be trained to obtain data representation of each enhancement data;
and carrying out self-supervision training on the mapping network to be trained according to the data representation of each piece of enhancement data to obtain the mapping network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210742692.1A CN115204220A (en) | 2022-06-28 | 2022-06-28 | Blade state detection method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210742692.1A CN115204220A (en) | 2022-06-28 | 2022-06-28 | Blade state detection method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115204220A true CN115204220A (en) | 2022-10-18 |
Family
ID=83578625
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210742692.1A Pending CN115204220A (en) | 2022-06-28 | 2022-06-28 | Blade state detection method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115204220A (en) |
-
2022
- 2022-06-28 CN CN202210742692.1A patent/CN115204220A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zare et al. | Simultaneous fault diagnosis of wind turbine using multichannel convolutional neural networks | |
| CN110276416B (en) | Rolling bearing fault prediction method | |
| Chen et al. | Diagnosing planetary gear faults using the fuzzy entropy of LMD and ANFIS | |
| CN112733283A (en) | Wind turbine generator component fault prediction method | |
| CN111209934A (en) | Fan fault prediction and alarm method and system | |
| CN111191838B (en) | Industrial equipment state management and control method and device integrating artificial intelligence algorithm | |
| Wu et al. | Design a degradation condition monitoring system scheme for rolling bearing using EMD and PCA | |
| CN114841580A (en) | Generator fault detection method based on hybrid attention mechanism | |
| CN115329986A (en) | Wind turbine generator anomaly detection and positioning method based on interpretable graph neural network | |
| CN114708885A (en) | Fan fault early warning method based on sound signals | |
| Wang et al. | Wind turbine fault detection and identification through self-attention-based mechanism embedded with a multivariable query pattern | |
| CN117473411A (en) | Bearing life prediction method based on improved transducer model | |
| CN117664558A (en) | Generator gear box abnormality detection method, device, equipment and storage medium | |
| CN111914490B (en) | Pump station unit state evaluation method based on depth convolution random forest self-coding | |
| CN117458955A (en) | Operation control method and system of motor | |
| CN115204220A (en) | Blade state detection method and device | |
| Song et al. | Framework of designing an adaptive and multi-regime prognostics and health management for wind turbine reliability and efficiency improvement | |
| Sekhon et al. | A comparison of two trending strategies for gas turbine performance prediction | |
| KR20230102431A (en) | Oil gas plant equipment failure prediction and diagnosis system based on artificial intelligence | |
| CN110441081B (en) | Intelligent diagnosis method and intelligent diagnosis system for faults of rotating machinery | |
| Ayman et al. | Fault Detection in Wind Turbines using Deep Learning | |
| Cross et al. | Model-based condition monitoring for wind turbines | |
| CN111878323A (en) | Wind generating set fault early warning method based on frequency spectrum autocorrelation function | |
| Wei et al. | An Early Fault Detection Method of the Induced Draft Fan Based on Long-Short Term Memory Network and Double Warning Thresholds | |
| Bejaoui et al. | Remaining Useful Life Prediction based on Degradation Model: Application to a Scale Replica Assembly Plant |
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 |