CN116306302A - Multi-working-condition residual life prediction method for key components of wind driven generator - Google Patents
Multi-working-condition residual life prediction method for key components of wind driven generator Download PDFInfo
- Publication number
- CN116306302A CN116306302A CN202310309847.7A CN202310309847A CN116306302A CN 116306302 A CN116306302 A CN 116306302A CN 202310309847 A CN202310309847 A CN 202310309847A CN 116306302 A CN116306302 A CN 116306302A
- Authority
- CN
- China
- Prior art keywords
- module
- domain
- loss
- service life
- residual
- 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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06N3/096—Transfer learning
-
- 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
- G06F2119/04—Ageing analysis or optimisation against ageing
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of wind power generation equipment prediction maintenance, in particular to a multi-working-condition residual life prediction method for key components of a wind power generator, and a source domain sample set D is constructed S And target domain sample set D T The method comprises the steps of carrying out a first treatment on the surface of the Constructing a feature extractor module; constructing a multi-characterization self-adaptive module; constructing a mobility attention mechanism module; constructing a residual service life predictor module; constructing a domain adaptation module; training a residual service life prediction model; and predicting the residual service life of the key component of the wind driven generator under the target domain working condition, inputting a target domain test sample set into the established network model, and outputting the residual service life prediction results of the key component at all moments. According to the multi-working-condition residual life prediction method for the key components of the wind driven generator, a movable attention mechanism module is provided, and the module can dynamically activate degradation characteristics with high mobility in model training, so that the generalization capability of the model is improved.
Description
Technical Field
The invention relates to the technical field of prediction maintenance of wind power generation equipment, in particular to a multi-station residual life prediction method for key components of a wind power generator.
Background
Wind energy is a renewable energy source with great potential and is widely focused on all countries around the world. The wind driven generator is a core component of a wind power generation system and is used for converting mechanical energy into electric energy. Most wind driven generators are installed in remote and severe environments such as offshore, highland, mountain areas and the like, and the operation working conditions are extremely harsh, so that the performance of parts of the wind driven generator is easy to degrade. At present, as the installed capacity of wind power generation is continuously increased, the loss caused by bearing faults of the wind power generator is also increased. Ensuring stable, safe and reliable operation of the wind turbine generator becomes a primary problem facing the wind power generation industry. The prediction of the residual service life, also called residual service life prediction, refers to the time that the equipment can ensure safe and reliable operation of the machine under a specified operation condition. The residual service life of key parts of the equipment is known in time, so that the method is an important method for reducing production loss, and the maintenance cost of the whole service life period can be saved. Therefore, the residual service life of key parts of the wind driven generator is predicted, safe and reliable operation of the unit is ensured, and shutdown caused by serious accidents is reduced, so that the economy of the operation of the unit is improved. With the rapid development of the global wind power generation industry, the residual service life prediction technology of key parts of the wind power generator is more and more important.
From the data disclosed at present, related researches are carried out to predict the residual service life of key parts of the wind driven generator through a machine learning or deep learning method, for example, patent numbers of a construction method of a residual service life prediction model of a bearing of the wind driven generator are as follows: CN 202010416839.9), a fan remaining life prediction method based on a multi-channel separable residual neural network (patent number: CN 202210031823.5), and the like. The prior art solutions still have the following drawbacks:
1. in the existing method, time domain analysis and frequency domain analysis are mostly adopted to extract characteristic parameters of signals, and a fitting means is utilized to conduct life regression analysis. But the correlation between the extracted features and the service life of the bearing is not strong, and the prediction performance is limited;
2. the existing method adopts deep learning to self-adaptively mine implicit degradation characteristics from vibration signals. However, the working conditions of the wind driven generator are changeable, and the generalization performance of the model can be affected due to the distribution change difference caused by the changeable working conditions.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a multi-station residual life prediction method for key parts of a wind driven generator, and aims to improve the prediction precision and generalization capability of a fan key part residual life prediction model under multi-station conditions.
In order to achieve the above purpose, the present invention provides the following technical solutions: a multi-working-condition residual life prediction method for key components of a wind driven generator comprises the following steps:
s1: constructing a Source Domain sample set D S And target domain sample set D T ：
And selecting data of key parts of the wind driven generator under two different working conditions from the historical database as a training set and a testing set. Training set data is composed of labeled source domain data D S And unlabeled target domain data D T Composition is prepared. The test set data consists of unlabeled target domain data;
s2: build feature extractor module M F ；
S3: construction of Multi-characterization adaptive Module M MR ；
S4: constructing a mobility attention mechanism module;
s5: building remaining life predictor Module M RP ；
S6: building domain adaptation module M DA ；
Calculating the conditional distribution difference loss of the weighted depth degradation feature AF extracted in S4 to obtain a conditional distribution difference loss L DA ；
S7: training a residual service life prediction model;
s8: and predicting the residual service life of the key component of the wind driven generator under the target domain working condition, inputting a target domain test sample set into the established network model, and outputting the residual service life prediction results of the key component at all moments.
Preferably, the step of constructing a feature extractor module in S2 specifically includes the following steps:
(2a) Building a convolutional neural network which is formed by stacking and connecting four sub-modules with the same structure and is used as a feature extractor module M F Each sub-module structure consists of a separation convolution layer, a batch normalization layer, a nonlinear activation function layer and a pooling layer;
(2b) Sample set D of source domain S And target domain sample set D T Input feature extractor module M F Depth degradation features F are extracted.
Preferably, the step of constructing a multi-characterization adaptive module in S3 specifically includes the following steps:
(3a) Constructing an InceptionA network connected in parallel by four different scales as a multi-characterization adaptive module M MR ；
(3b) Inputting depth degradation feature F into multi-characterization adaptive module M MR Obtaining four depth degradation features F with different scales s1 ，F s2 ，F s3 And F s4 。
Preferably, the step of constructing a mobility attention mechanism module in S4 specifically includes the following steps:
(4a) Constructing 4 domain classifiers consisting of a 4-layer fully connected network and a Softmax activation function layer;
(4b) S3, extracting scale i depth degradation characteristic F si Inputting the domain classifier corresponding to the scale to obtain domain classification probability P d Sum domain classification label l d ；
(4c) Depth degradation characteristic F of scale i extracted in S3 si Gradient inversion is carried out, and an entropy value H is utilized, so that the calculation formula is as follows:
H(F si )＝-∑P d ·log(P d )
(4d) Converting entropy value H into attention weight W si The calculation formula is as follows:
W si ＝1-H(F si )
(4e) Depth degradation feature F si And the depth degradation feature F si Corresponding attention weight W si +1 multiplication to obtain weighted depth degradation feature AF si ；
(4f) Domain classification tag/by negative log likelihood loss d Sum domain actual class label calculation migratable attention loss L A 。
Preferably, the step of constructing a remaining service life predictor module in S5 specifically includes the following steps:
(5a) Building 1 residual service life predictor composed of a layer 1 fully connected network;
(5b) Inputting the weighted depth degradation feature AF of the source domain extracted in the S4 into a residual service life predictor to obtain a predicted residual service life predictor Y of the source domain S ；
(5c) Residual life prediction value Y predicted in source domain by mean square error S And trueResidual life prediction value Y S R Comparing to obtain predicted loss L R 。
Preferably, the training of the residual service life prediction model in S7 specifically includes the following steps:
(7a) Setting a learning rate and model iteration times;
(7b) Sample set D of source domain S And target domain sample set D T The state monitoring signals of the sequence are sequentially input into a feature extractor module Gf to obtain depth source domain features and target domain features F, and then sequentially and respectively input into a feature extractor module M F Multi-characterization adaptive module M MR A migratable attention mechanism module, a remaining life predictor module M RP Sum domain adaptation module M DA Respectively calculating prediction loss L R Conditional distribution variation loss L DA And migratable attention loss L A ；
(7c) Predictive loss L obtained according to step (6 b) R Conditional distribution variation loss L DA And migratable attention loss L A The total loss function L is calculated as follows:
L＝L R +λ A L A +λ DA L DA
wherein lambda is A Is a migratable attention loss L A Penalty coefficient lambda DA Is the conditional distribution difference loss L DA Penalty coefficients;
(7d) Sequentially iterating the feature extractor module M by adopting Adam optimization algorithm F Multi-characterization adaptive module M MR A migratable attention mechanism module, a remaining life predictor module M RP And obtaining a trained residual service life prediction model until the maximum iteration times.
Compared with the prior art, the invention has the beneficial effects that:
1. the multi-working-condition residual life prediction method for the key parts of the wind driven generator is used for predicting the residual life of the key parts of the wind driven generator under different working conditions, and the multi-characterization movable attention network is provided, so that the distribution of degradation characteristics extracted from different characterization structures can be extracted and aligned by the model, and the high-order degradation information is enriched.
2. According to the multi-working-condition residual life prediction method for the key components of the wind driven generator, a movable attention mechanism module is provided, and the module can dynamically activate degradation characteristics with high mobility in model training, so that the generalization capability of the model is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort to a person skilled in the art.
FIG. 1 is a flow chart of an implementation of the present invention;
FIG. 2 is a MMR diagram of the multi-characterization adaptive module of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In the present invention, unless explicitly specified and limited otherwise, the terms "mounted," "connected," "secured," and the like are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally formed; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communicated with the inside of two elements or the interaction relationship of the two elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
Referring to fig. 1-2, the present invention provides a technical solution: a multi-working-condition residual life prediction method for key components of a wind driven generator comprises the following steps:
s1: constructing a Source Domain sample set D S And target domain sample set D T ：
And selecting data of key parts of the wind driven generator under two different working conditions from the historical database as a training set and a testing set. Training set data is composed of labeled source domain data D S And unlabeled target domain data D T Composition is prepared. The test set data consists of unlabeled target domain data;
s2: build feature extractor module M F ；
S3: construction of Multi-characterization adaptive Module M MR ；
S4: constructing a mobility attention mechanism module;
s5: building remaining life predictor Module M RP ；
S6: building domain adaptation module M DA ；
Calculating the conditional distribution difference loss of the weighted depth degradation feature AF extracted in S4 to obtain a conditional distribution difference loss L DA ；
S7: training a residual service life prediction model;
s8: and predicting the residual service life of the key component of the wind driven generator under the target domain working condition, inputting a target domain test sample set into the established network model, and outputting the residual service life prediction results of the key component at all moments.
The technical idea for realizing the purpose of the invention is that firstly, state monitoring data of key components of a wind driven generator under two different working conditions are collected, and a source domain sample set D is obtained S And target domain sample set D T The method comprises the steps of carrying out a first treatment on the surface of the Then, the source domain sample set D S And target domain sample set D T Input feature extractor module M F Extracting depth degradation characteristics F; secondly, the depth degradation characteristic F is input into a multi-characterization adaptive module M MR Obtaining four depth degradation features F with different scales s1 ，F s2 ，F s3 And F s4 . Each depth ofDegradation characteristics F si Will be input into the mobility attention mechanism module corresponding to the degradation feature to obtain the attention weight W of the depth degradation feature si And migratable attention loss L A . Depth degradation feature F si And the depth degradation feature F si Corresponding attention weight W si +1 multiplication to obtain weighted depth degradation feature AF si . Will weight depth degradation feature AF si Input remaining useful life predictor module M RP Sum domain adaptation module M DA Respectively obtaining a predicted residual service life predicted value Y and a conditional distribution difference loss L DA . Predicting a predicted remaining useful life value Y in a source domain S And a true residual life prediction value Y S R Comparing to obtain predicted loss L R . Will predict loss L R Migratable attention loss L A And conditional distribution difference loss L DA The addition results in a total loss. And finally, sequentially iterating and updating module parameters by adopting an Adam optimization algorithm to minimize the total loss, and finishing the training of the residual service life prediction model. And inputting the target domain test sample set into the established network model to obtain a final residual service life prediction result.
Step 1, acquiring a source domain sample set and a target domain sample set.
The embodiment adopts a laboratory bearing full-period degradation data set, and bearing state monitoring data are acquired through two acceleration sensors fixed on the outer ring of the bearing, wherein the sampling frequency is 25.6kHz. The present case study used data sets under two different operating conditions, as shown in table 1. The different working conditions are set as the source domain in turn, the other working condition is set as the target domain, and two migration tasks are set respectively, as shown in table 2.
TABLE 1
TABLE 2
Step 2, constructing a feature extractor module M F ：
Firstly, building a four-layer convolutional neural network which is formed by stacking and connecting four sub-modules with the same structure as a feature extractor module M F Each sub-module structure consists of a separate convolution layer, a batch normalization layer, a nonlinear activation function layer and a pooling layer, wherein the number of convolution kernels of each layer is [4,8, 16, 32 ]]The core size is set to 3×1 and the pooling layer size is set to 3×1.
Second, the source domain sample set D S And target domain sample set D T Input feature extractor module M F Extracting depth degradation characteristics F;
step 3, constructing a multi-characterization self-adaptive module M MR ：
Firstly, constructing an InceptionA network which is formed by connecting four different scales in parallel as a multi-characterization adaptive module M MR The structure is shown in fig. 2.
Then, the depth degradation feature F is input into a multi-characterization adaptive module M MR Obtaining four depth degradation features F with different scales s1 ，F s2 ，F s3 And F s4 ；
Step 4, constructing a mobility attention mechanism module:
first, 4 domain classifiers composed of a 4-layer fully connected network and a Softmax activation function layer are built, and nodes of the fully connected layer are set as follows: 288-80-20-2.
Then, the scale i depth degradation feature F extracted in step 3 si Inputting the domain classifier corresponding to the scale to obtain domain classification probability P d Sum domain classification label l d ；
Secondly, the depth degradation characteristic F of the scale i extracted in the step 3 is obtained si Gradient inversion is carried out, and an entropy value H is utilized, so that the calculation formula is as follows:
H(F si )＝-∑P d ·log(P d )
next, the entropy value H is converted into an attention weight W si The calculation formula is as follows:
W si ＝1-H(F si )
finally, depth degradation feature F si And the depth degradation feature F si Corresponding attention weight W si +1 multiplication to obtain weighted depth degradation feature AF si ；
Meanwhile, domain classification label l through negative log likelihood loss d Sum domain actual class label calculation migratable attention loss L A ；
Step 5, constructing a residual service life predictor module M RP ：
Firstly, 1 residual service life predictor composed of a 1-layer fully-connected network is built, and the nodes of the fully-connected layer are set as follows: 288-1.
Then, inputting the weighted depth degradation feature AF of the source domain extracted in the step (4) into a residual service life predictor to obtain a predicted residual service life predictor Y of the source domain S ；
Second, the residual life prediction value Y predicted in the source domain is calculated by mean square error S And a true residual life prediction value Y S R Comparing to obtain predicted loss L R ；
Step 6, constructing a domain adaptation module M DA Calculating the conditional distribution difference loss of the weighted depth degradation feature AF extracted in the step (4) to obtain a conditional distribution difference loss L DA 。
Step 7, training a residual service life prediction model:
firstly, setting a learning rate to be 0.0005 and a model iteration number to be 500;
then, the source domain sample set D S And target domain sample set D T Is sequentially input into the feature extractor module G f Obtaining depth source domain features and target domain features F, and sequentially and respectively inputting the depth source domain features and the target domain features F into a feature extractor module M F Multi-characterization adaptive module M MR A migratable attention mechanism module, a remaining life predictor module M RP Sum domain adaptation module M DA Respectively calculating predicted lossesLoss of L R Conditional distribution variation loss L DA And migratable attention loss L A ；
Next, the predicted loss L obtained according to step (6 b) R Conditional distribution variation loss L DA And migratable attention loss L A The total loss function L is calculated as follows:
L＝L R +λ A L A +λ DA L DA
wherein lambda is A Is a migratable attention loss L A Penalty coefficient, set to 0.01, λ in this embodiment DA Is the conditional distribution difference loss L DA Penalty factor, set to 10 in this embodiment;
finally, sequentially iterating the feature extractor module M by adopting Adam optimization algorithm F Multi-characterization adaptive module M MR A migratable attention mechanism module, a remaining life predictor module M RP Obtaining a trained residual service life prediction model until the maximum iteration times;
and 8, predicting the residual service life of the key components of the wind driven generator under the working condition of the target domain, inputting the test sample set of the target domain into the established network model, and outputting the residual service life prediction results of the key components at all times.
The effects of the present invention are further described below in conjunction with simulation experiments:
1. simulation experiment conditions:
the hardware platform of the simulation experiment of the invention is: the central processing unit is Intel (R) Core (TM) i5-7500 CPU, the main frequency is 3.40GHZ, and the memory is 16G.
The software platform of the simulation experiment of the invention is: WINDOWS 7 operating system and Python 3.7.
2. Simulation content and result analysis:
the simulation experiment of the invention adopts the method of the invention and 4 prior arts respectively, and comprises a residual service life prediction method based on non-contrast transfer learning and contrast transfer learning, and comprises a Condition Adaptive Network (CAN) using a condition maximum mean difference, a migratable convolutional neural network (TCNN) based on a multi-core maximum mean difference, a Condition Domain Adaptive Network (CDAN) [40] and a joint domain adaptive network (JAN) based on a joint maximum mean difference. In addition, the average absolute error MAE, root mean square error RMSE, and R2 index were used as evaluation indexes, and the comparison results are shown in table 3:
TABLE 3 Table 3
Analysis of Table 4 shows that the residual service life of the key component bearing of the wind driven generator is predicted by adopting the residual service life prediction model of the invention according to MAE, RMSE and R2 index results, and the optimal index performance is achieved. Compared with other methods, the model of the invention is greatly improved in index, which shows that the invention has good prediction performance on the residual service life of key parts of the wind driven generator under a multi-working-condition, and has higher accuracy and generalization capability.
It is noted that relational terms such as first and second, and the like are 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. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus 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 apparatus. Without further limitation, an element defined by the phrase "comprising one does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises an element.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (6)
1. A multi-working-condition residual life prediction method for key components of a wind driven generator is characterized by comprising the following steps of: the method comprises the following steps:
s1: constructing a Source Domain sample set D S And target domain sample set D T ：
And selecting data of key parts of the wind driven generator under two different working conditions from the historical database as a training set and a testing set. Training set data is composed of labeled source domain data D S And unlabeled target domain data D T Composition is prepared. The test set data consists of unlabeled target domain data;
s2: build feature extractor module M F ；
S3: construction of Multi-characterization adaptive Module M MR ；
S4: constructing a mobility attention mechanism module;
s5: building remaining life predictor Module M RP ；
S6: building domain adaptation module M DA ；
Calculating the conditional distribution difference loss of the weighted depth degradation feature AF extracted in S4 to obtain a conditional distribution difference loss L DA ；
S7: training a residual service life prediction model;
s8: and predicting the residual service life of the key component of the wind driven generator under the target domain working condition, inputting a target domain test sample set into the established network model, and outputting the residual service life prediction results of the key component at all moments.
2. The method for predicting the multi-station residual life of a key component of a wind driven generator according to claim 1, wherein the method comprises the following steps: the step S2 of constructing a feature extractor module specifically comprises the following steps:
(2a) Building a convolutional neural network which is formed by stacking and connecting four sub-modules with the same structure and is used as a feature extractor module M F Each sub-module structure consists of a separation convolution layer, a batch normalization layer, a nonlinear activation function layer and a pooling layer;
(2b) Sample set D of source domain S And target domain sample set D T Input feature extractor module M F Depth degradation features F are extracted.
3. The method for predicting the multi-station residual life of a key component of a wind driven generator according to claim 1, wherein the method comprises the following steps: the step S3 of constructing a multi-characterization self-adaptive module specifically comprises the following steps:
(3a) Constructing an InceptionA network connected in parallel by four different scales as a multi-characterization adaptive module M MR ；
(3b) Inputting depth degradation feature F into multi-characterization adaptive module M MR Obtaining four depth degradation features F with different scales s1 ，F s2 ，F s3 And F s4 。
4. The method for predicting the multi-station residual life of a key component of a wind driven generator according to claim 1, wherein the method comprises the following steps: the step S4 of constructing a mobility attention mechanism module specifically comprises the following steps:
(4a) Constructing 4 domain classifiers consisting of a 4-layer fully connected network and a Softmax activation function layer;
(4b) S3, extracting scale i depth degradation characteristic F si Inputting the domain classifier corresponding to the scale to obtain domain classification probability P d Sum domain classification label l d ；
(4c) Depth degradation characteristic F of scale i extracted in S3 si Gradient is carried outTurning over and using the entropy value H, the calculation formula is as follows:
H(F si )＝-∑P d ·log(P d )
(4d) Converting entropy value H into attention weight W si The calculation formula is as follows:
W si ＝1-H(F si )
(4e) Depth degradation feature F si And the depth degradation feature F si Corresponding attention weight W si +1 multiplication to obtain weighted depth degradation feature AF si ；
(4f) Domain classification tag/by negative log likelihood loss d Sum domain actual class label calculation migratable attention loss L A 。
5. The method for predicting the multi-station residual life of a key component of a wind driven generator according to claim 1, wherein the method comprises the following steps: and S5, constructing a residual service life predictor module, which specifically comprises the following steps:
(5a) Building 1 residual service life predictor composed of a layer 1 fully connected network;
(5b) Inputting the weighted depth degradation feature AF of the source domain extracted in the S4 into a residual service life predictor to obtain a predicted residual service life predictor Y of the source domain S ；
(5c) Residual life prediction value Y predicted in source domain by mean square error S And a true residual life prediction value Y S R Comparing to obtain predicted loss L R 。
6. The method for predicting the multi-station residual life of a key component of a wind driven generator according to claim 1, wherein the method comprises the following steps: and S7, training a residual service life prediction model, wherein the method specifically comprises the following steps of:
(7a) Setting a learning rate and model iteration times;
(7b) Sample set D of source domain S And target domain sample set D T Is sequentially input into the feature extractor module G f Obtaining depth sourcesThe domain features and the target domain features F are sequentially and respectively input into a feature extractor module M F Multi-characterization adaptive module M MR A migratable attention mechanism module, a remaining life predictor module M RP Sum domain adaptation module M DA Respectively calculating prediction loss L R Conditional distribution variation loss L DA And migratable attention loss L A ；
(7c) Predictive loss L obtained according to step (6 b) R Conditional distribution variation loss L DA And migratable attention loss L A The total loss function L is calculated as follows:
L＝L R +λ A L A +λ DA L DA
wherein lambda is A Is a migratable attention loss L A Penalty coefficient lambda DA Is the conditional distribution difference loss L DA Penalty coefficients;
(7d) Sequentially iterating the feature extractor module M by adopting Adam optimization algorithm F Multi-characterization adaptive module M MR A migratable attention mechanism module, a remaining life predictor module M RP And obtaining a trained residual service life prediction model until the maximum iteration times.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310309847.7A CN116306302A (en) | 2023-03-28 | 2023-03-28 | Multi-working-condition residual life prediction method for key components of wind driven generator |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310309847.7A CN116306302A (en) | 2023-03-28 | 2023-03-28 | Multi-working-condition residual life prediction method for key components of wind driven generator |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116306302A true CN116306302A (en) | 2023-06-23 |
Family
ID=86792441
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310309847.7A Pending CN116306302A (en) | 2023-03-28 | 2023-03-28 | Multi-working-condition residual life prediction method for key components of wind driven generator |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116306302A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117368724A (en) * | 2023-12-08 | 2024-01-09 | 天津国能津能滨海热电有限公司 | Motor life prediction method, device, medium and equipment |
-
2023
- 2023-03-28 CN CN202310309847.7A patent/CN116306302A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117368724A (en) * | 2023-12-08 | 2024-01-09 | 天津国能津能滨海热电有限公司 | Motor life prediction method, device, medium and equipment |
| CN117368724B (en) * | 2023-12-08 | 2024-03-19 | 天津国能津能滨海热电有限公司 | Motor life prediction method, device, medium and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109376801B (en) | Wind driven generator blade icing diagnosis method based on integrated deep neural network | |
| CN112070128B (en) | Transformer fault diagnosis method based on deep learning | |
| CN109784692B (en) | Rapid safety constraint economic dispatching method based on deep learning | |
| CN109000930A (en) | A kind of turbogenerator performance degradation assessment method based on stacking denoising self-encoding encoder | |
| CN110738349A (en) | Power grid fault first-aid repair duration prediction method based on multi-model fusion | |
| CN112785092B (en) | Switch residual life prediction method based on self-adaptive deep feature extraction | |
| CN109886403A (en) | A kind of industrial data generation method based on neural network model | |
| CN114218872B (en) | DBN-LSTM semi-supervised joint model-based residual service life prediction method | |
| CN116306302A (en) | Multi-working-condition residual life prediction method for key components of wind driven generator | |
| CN115561005A (en) | Chemical process fault diagnosis method based on EEMD decomposition and lightweight neural network | |
| Chen et al. | Research on wind power prediction method based on convolutional neural network and genetic algorithm | |
| CN112288147A (en) | Method for predicting insulation state of generator stator by BP-Adaboost strong predictor | |
| CN116842337A (en) | Transformer fault diagnosis method based on LightGBM (gallium nitride based) optimal characteristics and COA-CNN (chip on board) model | |
| Huang et al. | An adversarial learning approach for machine prognostic health management | |
| CN111141879A (en) | Deep learning air quality monitoring method, device and equipment | |
| CN114298080A (en) | Hydro-turbo generator set monitoring method based on throw data mining | |
| CN117520809A (en) | Transformer fault diagnosis method based on EEMD-KPCA-CNN-BiLSTM | |
| CN117113852A (en) | GIS life prediction method based on electric, chemical, structural and vibration parameters | |
| CN117388709A (en) | Energy storage battery fault diagnosis method based on whale optimization nuclear extreme learning machine | |
| CN117273188A (en) | Solar photovoltaic power generation power prediction method with incremental learning capability | |
| CN116401610A (en) | Inverter parameter fault diagnosis method based on depth residual error network and joint distribution | |
| CN112348700B (en) | Line capacity prediction method combining SOM clustering and IFOU equation | |
| CN115456287A (en) | Long-and-short-term memory network-based multi-element load prediction method for comprehensive energy system | |
| CN109165792B (en) | Photovoltaic short-term output power prediction method based on SOA-WNN | |
| CN113887571A (en) | Electronic equipment fault prediction method for improving SVR algorithm |
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 |