CN114841061A - Wind power gear box health assessment method and system integrating timing sequence information - Google Patents
Wind power gear box health assessment method and system integrating timing sequence information Download PDFInfo
- Publication number
- CN114841061A CN114841061A CN202210423006.4A CN202210423006A CN114841061A CN 114841061 A CN114841061 A CN 114841061A CN 202210423006 A CN202210423006 A CN 202210423006A CN 114841061 A CN114841061 A CN 114841061A
- Authority
- CN
- China
- Prior art keywords
- network
- health
- time sequence
- temperature
- model
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/06—Wind turbines or wind farms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/08—Thermal analysis or thermal optimisation
-
- 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
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Wind Motors (AREA)
Abstract
The invention discloses a wind power gearbox health assessment method and system integrating time sequence information, belonging to the field of wind power generator state monitoring, and comprising S1, multivariate time sequence acquisition and generation of sensor characteristics and time sequence characteristics; s2, constructing a convolution self-coding Gaussian mixed mode network; s3, training a network, setting a health degree index based on the energy value of the input signal and setting a threshold; s4, online evaluation; according to the method, time sequence information is obtained by encoding the time information and is combined with sensor information, a convolution self-encoding Gaussian mixture model is used for learning sensor characteristics and time sequence characteristics, and a health degree index based on original signal energy value distribution is provided and is used for evaluating the running health degree of the wind power gear box.
Description
Technical Field
The invention belongs to the field of state monitoring of wind driven generators, and relates to a wind power gearbox health assessment method and system integrating time sequence information.
Background
Wind power as a clean energy has been increasingly focused in recent years, and the installation amount of fans is also increasing. A large number of in-service wind turbine generators are in the middle stage of the working life cycle, wind power operators expect the fan fault early warning and pay more and more attention to the health state of the wind turbine generators, faults are eliminated when the health state of the fan is reduced, and major loss and casualties are avoided.
The gearbox is a key component of a transmission system of the wind driven generator, and the operation state of the gearbox directly influences the working state and the operation efficiency of the whole fan. The internal structure of the gear box is complex, the operation environment is severe, and the gear box is easily broken down when being under complex and variable alternating load for a long time. Since the units are usually installed in remote, infrequently located areas, the unit is shut down in the event of a failure, which results in high maintenance costs. Therefore, the method has important practical significance and application value for timely and accurate health assessment of the gear box.
The Data of a Supervisory Control and Data Acquisition (SCADA) system of a wind turbine generator is typical multi-source sensor Data, and the SCADA system has important significance in applying the multi-variable time sequence to performance evaluation and state monitoring of the wind turbine generator. Problems of information redundancy, large time sequence correlation and the like caused by coupling of wind turbine components in the SCADA data are not considered sufficiently, false alarm can be caused, and unnecessary maintenance cost is caused. Therefore, it is necessary to apply deep learning techniques to improve the accuracy of the health assessment.
Most of the existing researches establish a residual error model for the SCADA signal only based on a deep learning network. The subsequent construction of the performance evaluation indexes for the unit is easily influenced by the linear coupling among the variables. In reality, under the working condition that the running of the wind turbine is complex, the situation that the data of the wind turbine contains health state information is mostly not considered in the past, and false alarm is easily caused by the influence of a complex environment.
Disclosure of Invention
The invention aims to solve the defects of the problems and provides a wind power gearbox health assessment method and system integrating time sequence information.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a wind power gear box health assessment method fusing time sequence information comprises the following steps:
s1, feature generation: selecting characteristic variables to perform data preprocessing through multivariate time signals acquired by a wind turbine generator monitoring control and data acquisition system, performing One-Hot coding on time stamp information of SCADA data to serve as a time sequence characteristic, and taking the time sequence characteristic and a sensor characteristic as input signals for constructing a deep learning network model;
s2, network construction: constructing a convolution self-coding network module as a compression network, and inputting the time sequence characteristics and the sensor characteristics obtained in the step S1 into the compression network for characteristic learning and compression; constructing a Gaussian mixture model as an evaluation network, and fusing and inputting the learned compression characteristics, time sequence characteristics and reconstructed residual error characteristics into the evaluation network for learning;
s3, network training: and designing a health degree index based on the SCADA sample probability distribution output by the compression network in the step S2 network according to the Gaussian mixture model, and evaluating the health condition of the wind power gear box. And (4) inputting the data of the wind turbine generator set in the healthy state into the convolutional self-coding network constructed in the step S2, training the model to obtain a health evaluation model for learning the energy value distribution of the wind turbine generator set in the healthy state, and setting a threshold and storing the model.
S4, online evaluation: and inputting the SCADA data obtained through online monitoring into the trained health assessment model in the step S3, and outputting the health degree index of the wind power gear box.
The technical scheme of the invention is further improved as follows: in the step S1, the method includes the following steps:
s11, performing outlier detection on the original SCADA data through an isolated forest algorithm, and removing data which do not accord with physical significance;
s12, performing resampling and normalization processing on selected characteristic variables and eliminating the coupling of temperature-related variables through multivariate time signals acquired by a wind turbine generator monitoring control and data acquisition system; the multivariable comprises a working condition variable and a temperature state variable, and the temperature difference processing is carried out on the temperature state variable to eliminate the influence of the working condition factors on the temperature variable;
s13, carrying out One-Hot coding on the time stamp information of the SCADA data according to month, day, time and minute to obtain binary time sequence characteristic information of 12, 31, 24 and 60 dimensions. And respectively inputting the time sequence characteristics into a characteristic layer and an encoding layer of the network in the network training process.
S14, selecting the health data of the unit under the normal operation state to construct the input of the network model, and the training model does not need extra fault data.
The technical scheme of the invention is further improved as follows: in the step S2, the method includes the following steps:
and S21, learning the correlation between the time sequence of each feature and the dimension of the sensor by the model input through a convolution self-coding network, and extracting the low-dimensional features of the network input. The convolution self-encoder adopts an asymmetric structure, and the encoder and the decoder comprise multilayer convolution and pooling operations; and connecting the two layers by using a one-dimensional convolution layer, performing low-dimensional compression on the characteristic dimension by using a maximum pooling layer, and taking a ReLU layer as an activation function.
S22, constructing a Gaussian Mixture Model (GMM) as an evaluation network, and performing energy value evaluation on probability distribution information of input signals to obtain the health degree of the wind power gearbox during operation. The estimation network can directly estimate the parameters of the GMM and evaluate the likelihood of the sample through the unknown mixed component distribution, the mixed mean and the mixed covariance.
S23, the original network input can well distinguish partial sample abnormity under low-dimensional representation, in order to learn more space time sequence characteristics by the network, the compression network is connected with the evaluation network, so that the compression network can learn the evaluation information of the evaluation network during characteristic extraction, and the low-dimensional and time sequence characteristics of the multivariate time sequence and the reconstruction error of the convolution self-encoder are fused into a group of new characteristic variables to be input to the evaluation network.
z c =h(x;θ e )
x′=g(z c ;θ d )
z r =f(x,x′)
z=[z c ,z r ]
Constructing SCADA data samples X, z for step S1 c Representing the constructed reconstruction error theta e And theta d For convolutional autocoder parameters, x' is the SCADA sample reconstruction input, z r For the low-dimensional characterization learned by the deep compression network, z is the reconstruction error of the feature. h () denotes an encoding function, g () denotes a decoding function, and f () denotes a function that calculates the reconstruction error characteristics.
The technical scheme of the invention is further improved as follows: in the step S3, the method includes the following steps:
s31, constructing a health degree index based on a Gaussian mixture model, and evaluating the running health degree of the gearbox;
given the feature z and the number feature K of the mixture components, the probability that a sample belongs to each of the Gaussian mixture distributions is
Wherein the content of the first and second substances,the prediction result is represented as a K-dimensional vector. p is the output of the multilayer perceptron.
Wherein the content of the first and second substances,for hidden layer representation, Z i Probability of belonging to the i-th component, andthen the mixing probability, mean and variance of the kth component in the GMM are represented, respectively. Then, the Health Indicator (Health Indicator) of the gearbox may be defined as:
and S32, weighting the health indexes of the signals on a time scale and outputting the weighted health indexes of the signals by considering the time sequence characteristics of the signals, so that the health indexes of the signals at a certain moment are influenced by the health indexes of the previous time sequences.
And S33, inputting the data of the set constructed in the step S1 in the healthy state into a convolution self-coding network, training a model to obtain an energy value distribution model under the condition that the wind turbine gearbox is in the normal operation state, and storing and setting a threshold value.
The technical scheme of the invention is further improved as follows: the operating condition variables include wind speed, power, and rotor speed.
The technical scheme of the invention is further improved as follows: the temperature state variables include ambient temperature, N temperatures of different positions of the generator bearings, N temperatures of different positions of the generator stator windings, hydraulic group oil temperature, gearbox bearing temperature on the high speed shaft, nacelle temperature, N temperatures of different positions of the high speed transformer, grid side inverter temperature, top nacelle controller temperature, hub controller temperature, VCP board temperature, separator ring room temperature, head cone temperature, VCP blocking coil temperature, N temperatures of different positions of the IGBT-drivers on the rotor side inverter, VCP cooling water temperature, bus section temperature.
A wind power gearbox health assessment system fusing timing sequence information comprises:
the characteristic generating module is used for acquiring multivariate time signals acquired by a wind turbine monitoring control and data acquisition system, selecting characteristic variables for data preprocessing, and performing One-Hot coding on the timestamp information of the SCADA data to obtain time sequence characteristics and sensor characteristics;
the network construction module is used for constructing the convolution self-coding network module as a compression network and inputting the time sequence characteristics and the sensor characteristics into the compression network for characteristic learning and compression; constructing a Gaussian mixture model as an evaluation network, and fusing and inputting the learned compression characteristics, time sequence characteristics and reconstructed residual error characteristics into the evaluation network for learning;
the training module is used for designing a health degree index based on SCADA sample probability distribution output by a compression network according to a Gaussian mixture model and evaluating the health condition of the wind power gear box; inputting the data of the wind turbine generator set in the healthy state into the convolutional self-coding network constructed in the step S2, training a model to obtain a health evaluation model for learning the energy value distribution of the wind turbine generator set in the healthy state, and setting a threshold and storing the model;
and the evaluation output module is used for inputting the SCADA data obtained by online monitoring into the trained health evaluation model and outputting the health degree index of the wind power gear box.
Due to the adoption of the technical scheme, the invention has the technical progress that:
the invention provides a wind power gear box health assessment method and system integrating time sequence information, wherein multi-variable time sequences acquired by a data acquisition and monitoring system (SCADA system) are utilized to obtain multi-dimensional information including evaluation health degree during the running period of a fan, and the multi-dimensional information is subjected to data preprocessing and one-hot coding on timestamp information; and designing a deep convolutional self-coding network as a compression network to obtain a low-dimensional representation of the multivariate time sequence. A Gaussian mixture model is designed as an evaluation network for analyzing the health degree of the wind turbine. Designing an integral network to connect a compression network and an evaluation network, so that the compression network learns the evaluation information of the evaluation network during feature extraction, and inputting the low-dimensional representation of a multivariate time sequence and the reconstruction error of a depth self-encoder into the evaluation network; an energy value evaluation index is designed to evaluate the performance of the fan during operation, and compared with the prior art, the method has the following beneficial effects:
1. according to the invention, a mode of encoding the SCADA signal time is adopted, and compared with a simple modeling of a single SCADA signal, the time-space characteristics of a multivariate time sequence are fused, so that the performance evaluation effect and efficiency can be improved.
2. The invention adaptively learns the time-space characteristics of the SCADA signal, does not depend on a complex signal processing or transformation method, belongs to an unsupervised learning process, does not need label information of data, reduces the manual marking cost and improves the generalization capability of a model;
3. according to the method, the time sequence characteristics of the multivariate time sequence, the low-dimensional characteristics and the reconstruction error of the self-encoder are used for calculation to obtain the energy value based on the data characteristics of the signal, and the energy value is used as the health index of the wind turbine generator, so that the accuracy of fault early warning and health assessment can be effectively improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is a schematic time code of the SCADA signal of the present invention;
FIG. 3 is a schematic diagram of the structure of the convolution self-coding Gaussian mixture model of the present invention;
FIG. 4 is a schematic diagram of the performance evaluation off-line training process and the on-line diagnostic testing process of the present invention;
FIG. 5 is the results of an on-line diagnostic test of the performance assessment method of the present invention;
FIG. 6 is a result of an on-line diagnostic test of the performance evaluation method for unfused timing information when the compression network is a self-encoder;
FIG. 7 shows the on-line diagnostic test results of the fusion timing information performance evaluation method when the compression network is a self-encoder;
FIG. 8 shows the results of the online diagnostic test of the performance evaluation method of the un-fused time sequence information of the convolution self-coding Gaussian mixture model.
Detailed Description
The present invention will be described in further detail with reference to the following examples:
FIG. 1 shows a wind power gear box health assessment method fusing time sequence information, which comprises five parts, namely obtaining a multivariate time sequence and preprocessing data, fusing time stamp information of SCADA data, obtaining sensor characteristics and time sequence characteristics, inputting processed health data into a convolution self-coding Gaussian mixed network, extracting time-space correlation characteristics, setting a threshold value according to a verification set, inputting online data into a model, calculating a health index, and judging whether fault early warning is performed or not according to a valve group.
The time sequence coding schematic diagram is shown in fig. 2, the time stamp information of the SCADA data is one-hot coded according to month, day, hour, minute and second to obtain information of 12, 31, 24, 60 and 60 dimensions as the input of the network, and the time dimension characteristic is obtained.
The original data of the experimental cases are derived from data of 2016, namely data from 1 month and 1 day to 7 months and 18 days, and 25493 pieces of data exist. The data set contains the SCADA system 43-dimensional sensor parameters. The 10.00-minute gearbox of the wind turbine generator fails when the year 2016, 07, 18 and 02. The selected sensor characteristics were chosen as follows:
wherein the operating condition variables comprise wind speed, power and rotor speed, and the temperature state variables comprise ambient temperature generator bearing temperature 1, generator bearing temperature 2, generator stator winding temperature 1, generator stator winding temperature 2, generator stator winding temperature 3, hydraulic group oil temperature, gearbox bearing temperature on the high speed shaft, nacelle temperature, high speed transformer temperature 1, high speed transformer temperature 2, high speed transformer temperature 3, grid side inverter temperature, top nacelle controller temperature, hub controller temperature, VCP board temperature, separator ring chamber temperature, nose cone temperature, VCP blocker coil temperature, IGBT-driver temperature on the rotor side inverter 1, IGBT-driver temperature on the rotor side inverter 2, IGBT-driver temperature on the rotor side inverter 3, VCP cooling water temperature, water temperature, Bus-section temperature. And the temperature difference treatment is carried out on the temperature variable, so that the influence of the working condition factors on the temperature variable is eliminated.
Fig. 3 is a schematic structural diagram of a convolution self-coding gaussian mixture model. Inputting training set data X into a convolution self-encoder in an asymmetric structure, wherein the encoder and a decoder comprise multilayer convolution and pooling operations; and connecting the two layers by using a one-dimensional convolution layer, performing low-dimensional compression on the characteristic dimension by using a maximum pooling layer, and taking a ReLU layer as an activation function.
z c =h(x;θ e )
x′=g(z c ;θ d )
z r =f(x,x′)
z=[z c ,z r ]
Constructing SCADA data samples X, z for step S1 c Representing the constructed reconstruction error theta e And theta d For convolutional autocoder parameters, x' is the SCADA sample reconstruction input, z r For the low-dimensional characterization learned by the deep compression network, z is the reconstruction error of the feature. h () denotes an encoding function, g () denotes a decoding function, and f () denotes a function that calculates the reconstruction error characteristics. Given the feature z and the number feature K of the mixture components, the probability that a sample belongs to each of the Gaussian mixture distributions is
Wherein the content of the first and second substances,the prediction result is represented as a K-dimensional vector. p is the output of the multilayer perceptron.
WhereinFor hidden layer representation, Z i Probability of belonging to the i-th component, andthen the mixing probability, mean and variance of the kth component in the GMM are represented, respectively. Then, the Health Indicator (Health Indicator) of the gearbox may be defined as:
after the health data samples are subjected to network feature extraction and feature learning, the health degree index of the gearbox is calculated, and 0.5% quantile point of the health index of the gearbox in a healthy state is used as a threshold value.
In the online fault early warning part, online real-time monitoring multivariable time data are acquired from a wind turbine generator, are input into a model, are calculated to obtain health indexes, and are compared with a preset threshold value; when the abnormal score is larger than the threshold value, a fault early warning is sent out, as shown in fig. 4.
On-line fault early warning is shown in fig. 5, a transverse line is a threshold value, a health index is a sampling value of the health degree of the gearbox every day in a test set, and a fault occurs in the last day in the test set. It can be seen that the 2016 month 7 and day 8 anomaly score exceeds a threshold and continues to rise, so that gearbox failure can be effectively forewarned 10 days in advance.
Comparing the results of fig. 5 with those of fig. 6, fig. 7 and fig. 8, it can be seen that the convolution self-coding gaussian mixture model with the fused timing sequence information can reduce false alarms and reflect the health condition of the wind power gearbox.
The system based on the wind driven generator gearbox component health assessment method comprises a feature generation module, a network construction module, a training module and an assessment output module.
The characteristic generation module is used for acquiring multivariate time signals acquired by a wind turbine monitoring control and data acquisition system, selecting characteristic variables for data preprocessing, and performing One-Hot coding on timestamp information of SCADA data to obtain time sequence characteristics and sensor characteristics; the network construction module is used for constructing a convolution self-coding network module as a compression network and inputting the time sequence characteristics and the sensor characteristics into the compression network for characteristic learning and compression; constructing a Gaussian mixture model as an evaluation network, and fusing and inputting the learned compression characteristics, time sequence characteristics and reconstructed residual error characteristics into the evaluation network for learning; the training module is used for designing a health degree index based on SCADA sample probability distribution output by a compression network according to a Gaussian mixture model and evaluating the health condition of the wind power gear box; inputting the data of the wind turbine generator set in the healthy state into the convolutional self-coding network constructed in the step S2, training a model to obtain a health evaluation model for learning the energy value distribution of the wind turbine generator set in the healthy state, and setting a threshold and storing the model; and the evaluation output module is used for inputting the SCADA data obtained by online monitoring into the trained health evaluation model and outputting the health degree index of the wind power gear box.
The above-mentioned embodiments are merely illustrative of the preferred embodiments of the present invention, and do not limit the scope of the present invention, and various modifications and improvements of the technical solution of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the appended claims.
Claims (7)
1. A wind power gear box health assessment method integrating time sequence information is characterized by comprising the following steps:
s1, feature generation: selecting characteristic variables to perform data preprocessing through multivariate time signals acquired by a wind turbine generator monitoring control and data acquisition system, performing One-Hot coding on time stamp information of SCADA data to serve as a time sequence characteristic, and taking the time sequence characteristic and a sensor characteristic as input signals for constructing a deep learning network model;
s2, network construction: constructing a convolution self-coding network module as a compression network, and inputting the time sequence characteristics and the sensor characteristics obtained in the step S1 into the compression network for characteristic learning and compression; constructing a Gaussian mixture model as an evaluation network, and fusing and inputting the learned compression characteristics, time sequence characteristics and reconstructed residual error characteristics into the evaluation network for learning;
s3, network training: designing a health degree index based on the SCADA sample probability distribution output by the compression network in the step S2 network according to the Gaussian mixture model, wherein the health degree index is used for evaluating the health condition of the wind power gearbox; inputting the data of the wind turbine generator set in the healthy state into the convolutional self-coding network constructed in the step S2, training a model to obtain a health evaluation model for learning the energy value distribution of the wind turbine generator set in the healthy state, and setting a threshold and storing the model;
s4, online evaluation: and inputting the SCADA data obtained through online monitoring into the trained health assessment model in the step S3, and outputting the health degree index of the wind power gear box.
2. The wind power gearbox health assessment method integrating timing information as set forth in claim 1, wherein: in the step S1, the method includes the following steps:
s11, performing outlier detection on the original SCADA data through an isolated forest algorithm, and removing data which do not accord with physical significance;
s12, performing resampling and normalization processing on selected characteristic variables and eliminating the coupling of temperature-related variables through multivariate time signals acquired by a wind turbine generator monitoring control and data acquisition system; the multivariable comprises a working condition variable and a temperature state variable, and the temperature difference processing is carried out on the temperature state variable to eliminate the influence of the working condition factors on the temperature variable;
s13, carrying out One-Hot coding on the time stamp information of the SCADA data according to month, day, time and minute to obtain binary time sequence characteristic information of 12, 31, 24 and 60 dimensions; respectively inputting the time sequence characteristics into a characteristic layer and a coding layer of the network in the network training process;
s14, selecting the health data of the unit under the normal operation state to construct the input of the network model, and the training model does not need extra fault data.
3. The wind power gearbox health assessment method integrating timing information as set forth in claim 1, wherein: in the step S2, the method includes the following steps:
s21, learning the correlation between the time sequence of each feature and the dimension of the sensor by the model input through a convolution self-coding network, and extracting the low-dimensional features of the network input; the convolution self-encoder adopts an asymmetric structure, and the encoder and the decoder comprise multilayer convolution and pooling operations; connecting and using a one-dimensional convolution layer, performing low-dimensional compression on the characteristic dimension by using a maximum pooling layer, and taking a ReLU layer as an activation function;
s22, constructing a Gaussian Mixture Model (GMM) as an evaluation network, and performing energy value evaluation on probability distribution information of an input signal to obtain the health degree of the wind power gearbox during operation; the estimation network can directly estimate the parameters of the GMM and evaluate the likelihood of the sample through unknown mixed component distribution, mixed mean and mixed covariance;
s23, the original network input can well distinguish partial sample abnormity under low-dimensional representation, in order to learn more space time sequence characteristics through the network, the compression network is connected with the evaluation network, the compression network can also learn the evaluation information of the evaluation network during characteristic extraction, and the low-dimensional and time sequence characteristics of a multivariate time sequence and the reconstruction error of a convolution self-encoder are fused into a group of new characteristic variables to be input into the evaluation network;
z c =h(x;θ e )
x′=g(z c ;θ d )
z r =f(x,x′)
z=[z c ,z r ]
constructing SCADA data samples X, z for step S1 c Representing the constructed reconstruction error theta e And theta d For convolutional autocoder parameters, x' is the SCADA sample reconstruction input, z r For the low-dimensional characterization learned by the deep compression network, z is the reconstruction error of the characteristic; h () denotes an encoding function, g () denotes a decoding function, and f () denotes a function that calculates the reconstruction error characteristics.
4. The wind power gearbox health assessment method integrating timing information as set forth in claim 1, wherein: in the step S3, the method includes the following steps:
s31, constructing a health degree index based on a Gaussian mixture model, and evaluating the running health degree of the gearbox;
given the feature z and the number feature K of the mixture components, the probability that a sample belongs to each of the Gaussian mixture distributions is
Wherein the content of the first and second substances,representing the prediction result, and is a vector with K dimensions; p is the output of the multilayer perceptron;
wherein the content of the first and second substances,for hidden layer representation, Z i Probability of belonging to the i-th component, andrespectively representing the mixing probability, the mean value and the variance of the kth component in the GMM; then, the Health Indicator (Health Indicator) of the gearbox may be defined as:
s32, weighting the health indexes of the signals on a time scale and then outputting the weighted health indexes of the signals due to the time sequence characteristics of the signals, so that the health indexes of the signals at a certain moment are influenced by the health indexes of the previous time sequences;
and S33, inputting the data of the set constructed in the step S1 in the healthy state into a convolution self-coding network, training a model to obtain an energy value distribution model under the condition that the wind turbine gearbox is in the normal operation state, and storing and setting a threshold value.
5. The wind power gearbox health assessment method integrating timing information as set forth in claim 2, wherein said operating condition variables comprise wind speed, power and rotor speed.
6. The wind turbine gearbox health assessment method with fusion of timing information according to claim 2, wherein the temperature state variables comprise ambient temperature, N temperatures at different locations of generator bearings, N temperatures at different locations of generator stator windings, hydraulic group oil temperature, gearbox bearing temperature on high speed shaft, nacelle temperature, N temperatures at different locations of high speed transformer, grid side inverter temperature, top nacelle controller temperature, hub controller temperature, VCP board temperature, split ring room temperature, head cone temperature, VCP choke coil temperature, N temperatures at different locations of IGBT-driver on rotor side inverter, VCP cooling water temperature, bus section temperature.
7. The utility model provides a fuse wind-powered electricity generation gear box health assessment system of sequential information which characterized in that includes:
the characteristic generating module is used for acquiring multivariate time signals acquired by a wind turbine monitoring control and data acquisition system, selecting characteristic variables for data preprocessing, and performing One-Hot coding on the timestamp information of the SCADA data to obtain time sequence characteristics and sensor characteristics;
the network construction module is used for constructing the convolution self-coding network module as a compression network and inputting the time sequence characteristics and the sensor characteristics into the compression network for characteristic learning and compression; constructing a Gaussian mixture model as an evaluation network, and fusing and inputting the learned compression characteristics, time sequence characteristics and reconstructed residual error characteristics into the evaluation network for learning;
the training module is used for designing a health degree index based on SCADA sample probability distribution output by a compression network according to a Gaussian mixture model and evaluating the health condition of the wind power gear box; inputting the data of the wind turbine generator in the healthy state into the convolutional self-coding network constructed in the step S2, training a model to obtain a health evaluation model for learning the energy value distribution of the wind turbine generator in the healthy state, and setting a threshold value and storing the model;
and the evaluation output module is used for inputting the SCADA data obtained by online monitoring into the trained health evaluation model and outputting the health degree index of the wind power gear box.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210423006.4A CN114841061A (en) | 2022-04-21 | 2022-04-21 | Wind power gear box health assessment method and system integrating timing sequence information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210423006.4A CN114841061A (en) | 2022-04-21 | 2022-04-21 | Wind power gear box health assessment method and system integrating timing sequence information |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114841061A true CN114841061A (en) | 2022-08-02 |
Family
ID=82565517
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210423006.4A Pending CN114841061A (en) | 2022-04-21 | 2022-04-21 | Wind power gear box health assessment method and system integrating timing sequence information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114841061A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117849620A (en) * | 2024-03-08 | 2024-04-09 | 东莞市星火齿轮有限公司 | Electric performance test method and system for brushless motor module and memory |
-
2022
- 2022-04-21 CN CN202210423006.4A patent/CN114841061A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117849620A (en) * | 2024-03-08 | 2024-04-09 | 东莞市星火齿轮有限公司 | Electric performance test method and system for brushless motor module and memory |
CN117849620B (en) * | 2024-03-08 | 2024-05-14 | 东莞市星火齿轮有限公司 | Electric performance test method and system for brushless motor module and memory |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107220469B (en) | Method and system for estimating state of fan | |
CN112834211A (en) | Fault early warning method for transmission system of wind turbine generator | |
CN111537219B (en) | Fan gearbox performance detection and health assessment method based on temperature parameters | |
CN110399986B (en) | Generating method of pump station unit fault diagnosis system | |
CN114215706B (en) | Early warning method and device for cracking faults of wind turbine generator blades | |
CN114372504A (en) | Wind turbine generator fault early warning method based on graph neural network | |
CN114841580A (en) | Generator fault detection method based on hybrid attention mechanism | |
Cui et al. | An anomaly detection approach using wavelet transform and artificial neural networks for condition monitoring of wind turbines' gearboxes | |
CN110969185A (en) | Equipment abnormal state detection method based on data reconstruction | |
CN114841061A (en) | Wind power gear box health assessment method and system integrating timing sequence information | |
Wang et al. | Wind turbine fault detection and identification through self-attention-based mechanism embedded with a multivariable query pattern | |
CN112308391A (en) | Real-time monitoring and anomaly detection method for equipment state based on neural network | |
CN116609055A (en) | Method for diagnosing wind power gear box fault by using graph convolution neural network | |
CN111814848A (en) | Self-adaptive early warning strategy design method for temperature fault of wind turbine generator | |
CN113468728B (en) | Variable pitch system fault prediction method based on neural network | |
CN114708675A (en) | DAE-based electric submersible pump fault diagnosis method, system, terminal and medium | |
CN116561691A (en) | Power plant auxiliary equipment abnormal condition detection method based on unsupervised learning mechanism | |
CN116498543A (en) | Pump abnormal condition online detection method based on multidimensional signal analysis | |
CN116383615A (en) | Multisource sensing information fusion and expansion method based on system state characterization | |
Song et al. | Anomaly detection of wind turbine generator based on temporal information | |
Souza et al. | Evaluation of data based normal behavior models for fault detection in wind turbines | |
Tutiv'en et al. | Wind turbine main bearing condition monitoring via convolutional autoencoder neural networks | |
Ayman et al. | Fault Detection in Wind Turbines using Deep Learning | |
Du et al. | Anomaly Root Cause Analysis for Wind Turbines Based on Denoising Autoencoder and Sparse Estimation | |
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 |