CN110348513A - A kind of Wind turbines failure prediction method based on deep learning - Google Patents
A kind of Wind turbines failure prediction method based on deep learning Download PDFInfo
- Publication number
- CN110348513A CN110348513A CN201910618219.0A CN201910618219A CN110348513A CN 110348513 A CN110348513 A CN 110348513A CN 201910618219 A CN201910618219 A CN 201910618219A CN 110348513 A CN110348513 A CN 110348513A
- Authority
- CN
- China
- Prior art keywords
- failure
- fault
- wind turbines
- data
- prediction method
- 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
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000013135 deep learning Methods 0.000 title claims abstract description 22
- 230000010354 integration Effects 0.000 claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims description 29
- 238000013527 convolutional neural network Methods 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 26
- 230000000306 recurrent effect Effects 0.000 claims description 23
- 239000010410 layer Substances 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 8
- 230000007774 longterm Effects 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 238000002360 preparation method Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 239000002356 single layer Substances 0.000 claims description 3
- 230000002123 temporal effect Effects 0.000 claims description 3
- 238000010998 test method Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims 1
- 230000000694 effects Effects 0.000 description 3
- 230000015654 memory Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013479 data entry Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000003862 health status Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
Classifications
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Human Resources & Organizations (AREA)
- Life Sciences & Earth Sciences (AREA)
- Marketing (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Entrepreneurship & Innovation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Wind Motors (AREA)
Abstract
The invention discloses a kind of Wind turbines failure prediction method based on deep learning, the failure prediction method are directed to the SCADA data of running of wind generating set state, and the failure prediction method is based on fault prediction model and carries out failure predication；Failure prediction method is the following steps are included: be reassembled as operation data figure for the SCADA data of running of wind generating set state；The fault signature in operation data figure is extracted using CNN；Using LSTM as fault prediction model main body, timing failure feature is converted by single frames fault signature；The integration to timing fault signature is completed by full articulamentum, obtains failure predication result.The present invention is directed to magnanimity SCADA operation data, Wind turbines failure long period is completed using CNN-LSTM frame to predict, it can handle the high sample frequency long period timing SCADA data of multidimensional in Wind turbines, can effectively realize the Accurate Prediction of long period Wind turbines failure.
Description
Technical field
The invention belongs to wind-driven generator failure predication technical fields, are related to a kind of Wind turbines event based on deep learning
Hinder prediction technique more particularly to a kind of Wind turbines failure predication based on deep learning for Wind turbines SCADA data
Method.
Background technique
As environmental pollution is increasingly prominent with problem of energy crisis, wind energy is more next as a kind of clean renewable energy
Attention more by countries in the world has obtained quick development in recent years.Wind Power Generation Industry is ever-increasing simultaneously in installed capacity,
Also some new problems are faced.Due to the volume and working method of wind-engaging resource distribution, natural environment and Wind turbines itself
Etc. factors limitation, Wind turbines more be distributed in the more changeable area of uninhabited and environment, be subject to extreme environment because
Element interference, causes strong influence to the normal operation of Wind turbines.According to statistics, the operation and maintenance cost of Wind turbines and
Because disorderly closedown and caused by loss account for about the 10%-20% of entire wind power plant cost of electricity-generating.Therefore, accurate to realize to wind-powered electricity generation
The operation troubles of unit is predicted have for flexibly disposing and improving running of wind generating set reliability in advance and reduce loss
Significance.
Wind turbines failure prediction method is generally divided into the prediction technique based on model and the prediction technique based on data.Base
The physical model of fan part is established by analyzing the physical characteristic of fan part in the failure prediction method of model, according to model
The situation of change of parameter realizes failure predication, such as by establishing its faulty physical mechanism of gear case of blower model analysis, to mention
The fault signature under different faults state is taken to carry out gearbox fault prediction.Failure prediction method based on data mainly passes through pair
The acquisition of Wind turbines timing operation data, i.e. data and monitoring control (Supervisory Control And Data
Acquisition, SCADA) big data analysis of data realizes failure predication, have that data volume is big, parameter type is abundant, component
The characteristics of source is wide and high real-time.Use the long-term time series data of SCADA as the main body of running of wind generating set status predication
Can influence of the different components to Wind turbines failure more effectively in Cooperative Analysis Wind turbines, and be able to reflect failure shape
The timing variations of state, and then improve precision of prediction.For example, the base bearing health status using linear regression is predicted, feedforward is used
The methods of generating set status predication of neural network, the method relative to tradition based on Condition Monitoring Data are accurate in prediction
There is biggish promotion in rate.
Although the above method achieves preferable effect, also there are still biggish for current Wind turbines failure prediction method
Insufficient and room for promotion.Although failure prediction method based on model has the advantages that prediction effect is good and interpretation is strong, but
Modelling is highly dependent on the validity of skilled practitioner's modelling in certain circumstances, when Wind turbines equipment or outside
It when environment converts, requires skilled practitioner and intervenes and model is adjusted and is optimized again, wide usage and dynamic are suitable
It is poor with property.Based on the prediction technique of SCADA data because the feature by analyzing time series data itself has to predict failure then
Stronger wide usage, and accurately predict to provide better guarantee for long period.But due to current blower prediction technique shortage pair
Means are effectively treated in magnanimity time series data, at the place of the feature learning ability of fault data, the high real-time status data of long period
All there is limitation in reason ability, forecasting efficiency and precision etc., thus current Wind turbines failure prediction method there are still
The space for being further improved and being promoted.
Summary of the invention
To solve deficiency in the prior art, the present invention provides a kind of Wind turbines failure predication side based on deep learning
Method.On the one hand, by the way that the sequence SCADA data in long period is formed " operation data figure ", and convolutional neural networks are used
(Convolutional Neural Networks, CNN) carries out feature extraction to it, improves data throughput, self-study to reach
Practise the purpose of data characteristics and dimensionality reduction；On the other hand, consider that the feature of Wind turbines time series data, the present invention use circulation nerve
Network is as failure predication main body, since neuron connection structure has Memorability and parameter sharing characteristic in its unique layer,
It can be learnt with nonlinear characteristic of the higher efficiency to time series task.The length of excellent performance is used in the present invention
Short-term memory Recognition with Recurrent Neural Network network (Long Short-Term Memory, LSTM), complicated memory module and inside
Link structure has stronger memory capability, is also more suited for being spaced and postponing in processing Wind turbines failure predication time series
Relatively long task.
In order to achieve the above objectives, the present invention adopts the following technical scheme:
A kind of Wind turbines failure prediction method based on deep learning, the failure prediction method are transported for Wind turbines
The SCADA data of row state, the failure prediction method are based on fault prediction model and carry out failure predication；
The fault prediction model includes characteristic extracting module, failure predication module and full articulamentum；
The characteristic extracting module is constructed by multilayer convolutional neural networks structure, for extracting the event of multiframe operation data figure
Hinder feature；
The failure predication module is constructed by multilayer circulation neural network structure, for the fault signature of single frames to be integrated into
Continuous timing failure feature；
The full articulamentum, for integrating timing failure feature and exporting failure predication result；
The failure prediction method the following steps are included:
S1: the SCADA data of running of wind generating set state is reassembled as operation data figure；
S2: the fault signature in operation data figure is extracted using convolutional neural networks；
S3: using Recognition with Recurrent Neural Network as fault prediction model main body, timing failure spy is converted by single frames fault signature
Sign；
S4: the integration to timing fault signature is completed by full articulamentum, obtains failure predication result.
The present invention further comprises following preferred embodiment:
The training method of above-mentioned fault prediction model, comprising the following steps:
1) prepare training data；
2) the Wind turbines fault condition in specified time is predicted；
3) fault prediction model loss is calculated；
4) fault prediction model weight is updated；
5) judge whether penalty values or evaluation index failure predication accuracy rate are stable in threshold range or training of judgement
Whether number reaches the maximum number of iterations upper limit, if meeting condition, thens follow the steps 6), and otherwise frequency of training adds to return together and hold
Row step 2), the frequency of training initial value is zero；
6) fault prediction model file is exported.
The test method of above-mentioned fault prediction model, comprising the following steps:
1) setup test data, the test data set is without cutting；
2) load fault prediction model file calls trained fault prediction model file, and test data is inputted event
Hinder prediction model；
3) Wind turbines real-time running data is received, predicts the Wind turbines fault condition in specified time.
Above-mentioned preparation training data, specifically: training data is reassembled as corresponding single-pass by specified time interval
Road operation data figure and the input as fault prediction model, while the Wind turbines after period specified time that needs are predicted
Fail result is exported as corresponding model.
Wind turbines fault condition in above-mentioned prediction specified time, comprising the following steps:
Receive multiple operation data figures of certain time interval；
Feature extraction is carried out to the operation data figure by deep layer convolutional neural networks, by the Data distribution information of low layer
It is converted into high-rise fault signature；
Recognition with Recurrent Neural Network carries out temporal aspect conversion to fault signature, and the fault signature of each time slice is integrated into
Unified timing failure feature；
Timing fault signature is integrated by full articulamentum (FC), obtains failure predication result.
Above-mentioned calculating Wind turbines fault prediction model loss specifically:
Cross entropy (Cross Entropy) is used to calculate the mistake between output valve and target value as the loss function of model
Difference.
Model optimizer used in above-mentioned update Wind turbines fault prediction model weight is Adam algorithm.
Above-mentioned multilayer convolutional neural networks are difficult using the training that the strategy that same layer weight is shared reduces fault prediction model
Degree, and improve the speed of service.
In above-mentioned multilayer circulation neural network structure, input layer Recognition with Recurrent Neural Network has the mind for being equal to input frame number
Through first number, to complete to carry out forward-backward correlation processing to the fault signature of each frame.
Above-mentioned full articulamentum is the full articulamentum of single layer, and the Recognition with Recurrent Neural Network is that shot and long term remembers Recognition with Recurrent Neural Network
Network.
The SCADA data of running of wind generating set state is reassembled as operation data figure described in step S1, comprising the following steps:
S101: the SCADA data of running of wind generating set state described in normalized maps data into [0,1] section；
S102: the data after normalized are reassembled as operation data figure with time sequencing arrangement.
The fault signature in operation data figure is extracted using convolutional neural networks described in step S2, comprising the following steps:
S201: the single channel operation data figure of multiple X (m) × T (N) dimension of certain time interval is received as convolutional Neural
The input of network；
S202: feature extraction is carried out to operation data figure by convolutional neural networks, every single channel operation data figure turns
Turn to the fault signature of the dimension of X (m) × 1.
Using Recognition with Recurrent Neural Network as fault prediction model main body described in step S3, timing is converted by single frames fault signature
Fault signature, comprising the following steps:
S301: using the fault signature that step S202 is exported as the input of Recognition with Recurrent Neural Network；
S302: the fault signature of each time slice is integrated into unified timing failure feature.
The integration to timing fault signature is completed by full articulamentum described in step S4, obtains failure predication as a result, specific
Are as follows:
Whether Wind turbines are broken down and is considered as two classification problems, in full articulamentum, letter is activated by Softmax
Number, obtain whether the prediction result of failure.
Advantageous effects of the invention:
By the way that running of wind generating set status data to be reassembled as to " operation data figure ", feature is completed using convolutional neural networks
It extracts, and is realized by remembering Recognition with Recurrent Neural Network Innovation Integration with shot and long term to magnanimity timing running of wind generating set state diagram
It is effectively treated, and then improves the time span of prediction and the precision of long period failure predication.
Detailed description of the invention
Fig. 1 is fault prediction model structure chart of the invention；
Fig. 2 is the overall flow figure of the method for the present invention；
Fig. 3 is that SCADA data is reassembled as " operation data figure " schematic diagram in the embodiment of the present invention；
Fig. 4 is fault prediction model training and test flow chart of the invention.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings.Following embodiment is only used for clearly illustrating the present invention
Technical solution, and not intended to limit the protection scope of the present invention.
A kind of Wind turbines failure prediction method based on deep learning of the invention, the failure prediction method are directed to wind
The SCADA data of electric operating states of the units, the failure prediction method are based on fault prediction model and carry out failure predication；
As shown in Figure 1, the fault prediction model includes characteristic extracting module, failure predication module and full articulamentum；
The characteristic extracting module is constructed by multilayer convolutional neural networks structure (i.e. depth convolutional neural networks), for mentioning
Take the fault signature of multiframe operation data figure；
In embodiment, convolutional neural networks part needs to carry out feature extraction to whole k frame operating status figures, in order to mention
High efficiency, the shared strategy of multiple groups convolutional neural networks structure same layer weight, which can be used, reduces the training difficulty of model, and improves
The speed of service；
The failure predication module is constructed by multilayer circulation neural network structure, for the fault signature of single frames to be integrated into
Continuous timing failure feature；
In embodiment, the Recognition with Recurrent Neural Network is that shot and long term remembers Recognition with Recurrent Neural Network network；Recognition with Recurrent Neural Network needs
Fault signature to each frame to carry out forward-backward correlation processing respectively, thus input layer Recognition with Recurrent Neural Network has and is equal to input
The neuron number of frame number k.
The full articulamentum, for integrating timing failure feature and exporting failure predication result；
In embodiment, the full articulamentum is the full articulamentum of single layer.
As shown in figure 4, the training method of the fault prediction model, comprising the following steps:
1) prepare training data；
In embodiment, step 1) the preparation training data, specifically: by training data by specified time interval (one
As be 2 seconds) be reassembled as corresponding single channel operation data figure and the input as fault prediction model, while will need to predict
Wind turbines fail result after period specified time is exported as corresponding model.
2) the Wind turbines fault condition in specified time is predicted；
Wind turbines fault condition in embodiment, in step 2) the prediction specified time, comprising the following steps:
Receive multiple operation data figures of certain time interval；
Feature extraction is carried out to the operation data figure by deep layer convolutional neural networks, by the Data distribution information of low layer
It is converted into high-rise fault signature；
Shot and long term remembers Recognition with Recurrent Neural Network and carries out temporal aspect conversion to fault signature, by the failure of each time slice
Feature integration is unified timing failure feature；
Timing fault signature is integrated by full articulamentum, obtains failure predication result.
3) fault prediction model loss is calculated, specifically:
Cross entropy is used to calculate the error between output valve and target value as the loss function of model.
4) fault prediction model weight is updated；
In embodiment, model optimizer used in step 4) the update Wind turbines fault prediction model weight is Adam
Algorithm, wherein Adam optimizer uses β1=0.9, β2=0.999, ε=10-8As initial parameter.
5) judge whether penalty values or evaluation index failure predication accuracy rate are stable in threshold range or training of judgement
Whether number reaches maximum number of iterations upper limit N, if meeting condition, thens follow the steps 6), and otherwise frequency of training adds returns together
Step 2) is executed, the frequency of training initial value is zero；
6) fault prediction model file is exported.
The test method of the fault prediction model, comprising the following steps:
1) setup test data, in test process, in order to which the overall performance to model is assessed, test data set is not required to
Carry out additional cutting；
2) load fault prediction model file calls trained fault prediction model file, and test data is inputted event
Hinder prediction model；
3) Wind turbines real-time running data is received, predicts the Wind turbines fault condition in specified time.
As shown in Fig. 2, the failure prediction method the following steps are included:
S1: the SCADA data of running of wind generating set state is reassembled as operation data figure；
As shown in figure 3, the SCADA data of running of wind generating set state is reassembled as running described in step S1 in embodiment
Datagram, comprising the following steps:
S101: to guarantee that input data initial weight having the same needs to normalize place in order to carry out model training
The SCADA data for managing the running of wind generating set state maps data into [0,1] section；
S102: the data after normalized are reassembled as operation data figure with time sequencing arrangement, wherein T1 indicates the
One collected SCADA operation data of unit time institute, X1 indicate the specific value in data entry in first character section.
S2: the fault signature in operation data figure is extracted using convolutional neural networks, comprising the following steps:
S201: the single channel operation data figure of multiple X (m) × T (N) dimension of certain time interval is received as convolutional Neural
The input of network；
S202: feature extraction is carried out to operation data figure by convolutional neural networks, every single channel operation data figure turns
Turn to the fault signature of the dimension of X (m) × 1.
S3: using Recognition with Recurrent Neural Network as fault prediction model main body, timing failure spy is converted by single frames fault signature
Sign, comprising the following steps:
S301: using the fault signature that step S202 is exported as the input of Recognition with Recurrent Neural Network；
S302: the fault signature of each time slice is integrated into unified timing failure feature.
S4: completing integration to timing fault signature by full articulamentum, obtain failure predication as a result, specifically:
Whether Wind turbines are broken down and is considered as two classification problems, in full articulamentum, letter is activated by Softmax
Number, obtain whether the prediction result of failure.
This law is bright to have carried out Innovation Integration Design for CNN and two kinds of structures of LSTM, firstly, CNN is used as " operation data figure "
Characteristic extracting module, using CNN network to the good characterization ability of higher-dimension complex characteristic by the operation data in certain time
Figure conversion fault signature, to improve the purpose of data throughout, self study data characteristics；Secondly, using LSTM as failure predication
The failure predication module of main body will receive the fault signature input of multiple time serieses, and by the complicated memory module of design with
Internal links structure completes the processing to timing fault signature；The method of the present invention is directed to magnanimity SCADA operation data, uses CNN-
LSTM frame completes the prediction of Wind turbines failure long period, when can handle the high sample frequency long period of multidimensional in Wind turbines
Sequence SCADA data can effectively realize the Accurate Prediction of long period Wind turbines failure.
Inventor combines Figure of description to be described in detail and describe implementation example of the invention, still
It should be appreciated by those skilled in the art that implementing example above is only the preferred embodiments of the invention, explanation is only in detail
Help reader more fully understands spirit of that invention, and it is not intended to limit the protection scope of the present invention, on the contrary, any be based on this hair
Any improvement or modification made by bright spirit should all be fallen within the scope and spirit of the invention.
Claims (14)
1. a kind of Wind turbines failure prediction method based on deep learning, the failure prediction method is directed to running of wind generating set
The SCADA data of state, it is characterised in that:
The failure prediction method is based on fault prediction model and carries out failure predication；
The fault prediction model includes characteristic extracting module, failure predication module and full articulamentum；
The characteristic extracting module is constructed by multilayer convolutional neural networks structure, and the failure for extracting multiframe operation data figure is special
Sign；
The failure predication module is constructed by multilayer circulation neural network structure, for the fault signature of single frames to be integrated into continuously
Timing failure feature；
The full articulamentum, for integrating timing failure feature and exporting failure predication result；
The failure prediction method the following steps are included:
S1: the SCADA data of running of wind generating set state is reassembled as operation data figure；
S2: the fault signature in operation data figure is extracted using convolutional neural networks；
S3: using Recognition with Recurrent Neural Network as fault prediction model main body, timing failure feature is converted by single frames fault signature；
S4: the integration to timing fault signature is completed by full articulamentum, obtains failure predication result.
2. a kind of Wind turbines failure prediction method based on deep learning according to claim 1, it is characterised in that:
The training method of the fault prediction model, comprising the following steps:
1) prepare training data；
2) the Wind turbines fault condition in specified time is predicted；
3) fault prediction model loss is calculated；
4) fault prediction model weight is updated；
5) judge whether penalty values or evaluation index failure predication accuracy rate are stable in threshold range or training of judgement number
Whether reach the maximum number of iterations upper limit, if meeting condition, then follow the steps 6), otherwise frequency of training, which adds to return together, executes step
It is rapid 2), the frequency of training initial value is zero；
6) fault prediction model file is exported.
3. a kind of Wind turbines failure prediction method based on deep learning according to claim 2, it is characterised in that:
Step 1) the preparation training data, specifically: training data is reassembled as corresponding single-pass by specified time interval
Road operation data figure and the input as fault prediction model, while the Wind turbines after period specified time that needs are predicted
Fail result is exported as corresponding model.
4. a kind of Wind turbines failure prediction method based on deep learning according to claim 2, it is characterised in that:
Wind turbines fault condition in step 2) the prediction specified time, comprising the following steps:
Receive multiple operation data figures of certain time interval；
Feature extraction is carried out to the operation data figure by deep layer convolutional neural networks, the Data distribution information of low layer is converted
For high-rise fault signature；
Recognition with Recurrent Neural Network carries out temporal aspect conversion to fault signature, and the fault signature of each time slice is integrated into unification
Timing failure feature；
Timing fault signature is integrated by full articulamentum, obtains failure predication result.
5. a kind of Wind turbines failure prediction method based on deep learning according to claim 2, it is characterised in that:
Step 3) the calculating Wind turbines fault prediction model loss specifically:
Cross entropy is used to calculate the error between output valve and target value as the loss function of model.
6. a kind of Wind turbines failure prediction method based on deep learning according to claim 2, it is characterised in that:
Model optimizer used in step 4) the update Wind turbines fault prediction model weight is Adam algorithm.
7. a kind of Wind turbines failure prediction method based on deep learning according to claim 1, it is characterised in that:
The test method of the fault prediction model, comprising the following steps:
1) setup test data, the test data set is without cutting；
2) load fault prediction model file calls trained fault prediction model file, and test data input fault is pre-
Survey model；
3) Wind turbines real-time running data is received, predicts the Wind turbines fault condition in specified time.
8. -7 any a kind of Wind turbines failure prediction method based on deep learning according to claim 1, feature
It is:
The multilayer convolutional neural networks reduce the training difficulty of fault prediction model using the strategy that same layer weight is shared, and mention
The high speed of service.
9. -7 any a kind of Wind turbines failure prediction method based on deep learning according to claim 1, feature
It is:
In the multilayer circulation neural network structure, input layer Recognition with Recurrent Neural Network has the neuron for being equal to input frame number
Number, to complete to carry out forward-backward correlation processing to the fault signature of each frame.
10. -7 any a kind of Wind turbines failure prediction method based on deep learning according to claim 1, feature
It is:
The full articulamentum is the full articulamentum of single layer, and the Recognition with Recurrent Neural Network is that shot and long term remembers Recognition with Recurrent Neural Network network.
11. a kind of Wind turbines failure prediction method based on deep learning according to claim 1, it is characterised in that:
The SCADA data of running of wind generating set state is reassembled as operation data figure described in step S1, comprising the following steps:
S101: the SCADA data of running of wind generating set state described in normalized maps data into [0,1] section；
S102: the data after normalized are reassembled as operation data figure with time sequencing arrangement.
12. a kind of Wind turbines failure prediction method based on deep learning according to claim 1, it is characterised in that:
The fault signature in operation data figure is extracted using convolutional neural networks described in step S2, comprising the following steps:
S201: the single channel operation data figure of multiple X (m) × T (N) dimension of certain time interval is received as convolutional neural networks
Input；
S202: feature extraction is carried out to operation data figure by convolutional neural networks, every single channel operation data figure is converted into X
(m) × 1 the fault signature tieed up.
13. a kind of Wind turbines failure prediction method based on deep learning according to claim 12, it is characterised in that:
Using Recognition with Recurrent Neural Network as fault prediction model main body described in step S3, timing failure is converted by single frames fault signature
Feature, comprising the following steps:
S301: using the fault signature that step S202 is exported as the input of Recognition with Recurrent Neural Network；
S302: the fault signature of each time slice is integrated into unified timing failure feature.
14. a kind of Wind turbines failure prediction method based on deep learning according to claim 1, it is characterised in that:
Integration to timing fault signature is completed by full articulamentum described in step S4, obtain failure predication as a result, specifically:
Whether Wind turbines are broken down and is considered as two classification problems, in full articulamentum, by Softmax activation primitive,
Obtain whether the prediction result of failure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910618219.0A CN110348513A (en) | 2019-07-10 | 2019-07-10 | A kind of Wind turbines failure prediction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910618219.0A CN110348513A (en) | 2019-07-10 | 2019-07-10 | A kind of Wind turbines failure prediction method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110348513A true CN110348513A (en) | 2019-10-18 |
Family
ID=68177008
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910618219.0A Pending CN110348513A (en) | 2019-07-10 | 2019-07-10 | A kind of Wind turbines failure prediction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110348513A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110880062A (en) * | 2019-10-31 | 2020-03-13 | 广西电网有限责任公司电力科学研究院 | Method for determining state maintenance time of power distribution equipment |
| CN111141412A (en) * | 2019-12-25 | 2020-05-12 | 深圳供电局有限公司 | Cable temperature and anti-theft dual-monitoring method and system and readable storage medium |
| CN111565024A (en) * | 2020-06-03 | 2020-08-21 | 华北电力大学 | Solar cell panel fault detection and operation efficiency prediction device and method |
| CN111582396A (en) * | 2020-05-13 | 2020-08-25 | 江南大学 | Fault diagnosis method based on improved convolutional neural network |
| CN111709190A (en) * | 2020-06-24 | 2020-09-25 | 国电联合动力技术有限公司 | Wind turbine generator operation data image identification method and device |
| CN112131787A (en) * | 2020-09-18 | 2020-12-25 | 江西兰叶科技有限公司 | Unsupervised self-evolving motor design method and system |
| CN112286749A (en) * | 2020-11-23 | 2021-01-29 | 四川长虹电器股份有限公司 | Method for detecting edge cloud fault |
| CN112834211A (en) * | 2020-12-31 | 2021-05-25 | 江苏国科智能电气有限公司 | Fault early warning method for transmission system of wind turbine generator |
| CN112906915A (en) * | 2021-01-22 | 2021-06-04 | 江苏安狮智能技术有限公司 | Rail transit system fault diagnosis method based on deep learning |
| CN114936532A (en) * | 2022-07-20 | 2022-08-23 | 浙江大学 | Offshore wind turbine generator set fault diagnosis method |
| CN116484751A (en) * | 2023-06-21 | 2023-07-25 | 北京尚文汇通能源科技有限公司 | Fatigue life assessment method and device for wind turbine generator components |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106603293A (en) * | 2016-12-20 | 2017-04-26 | 南京邮电大学 | Network fault diagnosis method based on deep learning in virtual network environment |
| CN107132478A (en) * | 2017-04-13 | 2017-09-05 | 南京师范大学 | A kind of wind-driven generator method for diagnosing faults measured based on polarization maintaining optical fibre with neural network classification |
| CN109001631A (en) * | 2018-04-27 | 2018-12-14 | 张家港莫特普数据科技有限公司 | Battery performance online evaluation system and method |
| CN109033450A (en) * | 2018-08-22 | 2018-12-18 | 太原理工大学 | Lift facility failure prediction method based on deep learning |
| CN109034368A (en) * | 2018-06-22 | 2018-12-18 | 北京航空航天大学 | A kind of complex device Multiple Fault Diagnosis Method based on DNN |
| CN109117890A (en) * | 2018-08-24 | 2019-01-01 | 腾讯科技（深圳）有限公司 | A kind of image classification method, device and storage medium |
| CN109190540A (en) * | 2018-06-06 | 2019-01-11 | 腾讯科技（深圳）有限公司 | Biopsy regions prediction technique, image-recognizing method, device and storage medium |
| CN109555566A (en) * | 2018-12-20 | 2019-04-02 | 西安交通大学 | A kind of turbine rotor method for diagnosing faults based on LSTM |
| CN109814523A (en) * | 2018-12-04 | 2019-05-28 | 合肥工业大学 | Method for diagnosing faults based on CNN-LSTM deep learning method and more attribute time series datas |
-
2019
- 2019-07-10 CN CN201910618219.0A patent/CN110348513A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106603293A (en) * | 2016-12-20 | 2017-04-26 | 南京邮电大学 | Network fault diagnosis method based on deep learning in virtual network environment |
| CN107132478A (en) * | 2017-04-13 | 2017-09-05 | 南京师范大学 | A kind of wind-driven generator method for diagnosing faults measured based on polarization maintaining optical fibre with neural network classification |
| CN109001631A (en) * | 2018-04-27 | 2018-12-14 | 张家港莫特普数据科技有限公司 | Battery performance online evaluation system and method |
| CN109190540A (en) * | 2018-06-06 | 2019-01-11 | 腾讯科技（深圳）有限公司 | Biopsy regions prediction technique, image-recognizing method, device and storage medium |
| CN109034368A (en) * | 2018-06-22 | 2018-12-18 | 北京航空航天大学 | A kind of complex device Multiple Fault Diagnosis Method based on DNN |
| CN109033450A (en) * | 2018-08-22 | 2018-12-18 | 太原理工大学 | Lift facility failure prediction method based on deep learning |
| CN109117890A (en) * | 2018-08-24 | 2019-01-01 | 腾讯科技（深圳）有限公司 | A kind of image classification method, device and storage medium |
| CN109814523A (en) * | 2018-12-04 | 2019-05-28 | 合肥工业大学 | Method for diagnosing faults based on CNN-LSTM deep learning method and more attribute time series datas |
| CN109555566A (en) * | 2018-12-20 | 2019-04-02 | 西安交通大学 | A kind of turbine rotor method for diagnosing faults based on LSTM |
Non-Patent Citations (1)
| Title |
|---|
| 江国乾: "基于排序模式分析与深度学习的风电设备故障诊断方法研究", 《中国博士学位论文全文数据库 (工程科技Ⅱ辑)》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110880062B (en) * | 2019-10-31 | 2022-07-08 | 广西电网有限责任公司电力科学研究院 | Method for determining state maintenance time of power distribution equipment |
| CN110880062A (en) * | 2019-10-31 | 2020-03-13 | 广西电网有限责任公司电力科学研究院 | Method for determining state maintenance time of power distribution equipment |
| CN111141412A (en) * | 2019-12-25 | 2020-05-12 | 深圳供电局有限公司 | Cable temperature and anti-theft dual-monitoring method and system and readable storage medium |
| CN111582396A (en) * | 2020-05-13 | 2020-08-25 | 江南大学 | Fault diagnosis method based on improved convolutional neural network |
| CN111582396B (en) * | 2020-05-13 | 2023-05-02 | 江南大学 | Fault diagnosis method based on improved convolutional neural network |
| CN111565024A (en) * | 2020-06-03 | 2020-08-21 | 华北电力大学 | Solar cell panel fault detection and operation efficiency prediction device and method |
| CN111709190A (en) * | 2020-06-24 | 2020-09-25 | 国电联合动力技术有限公司 | Wind turbine generator operation data image identification method and device |
| CN112131787B (en) * | 2020-09-18 | 2022-05-27 | 江西兰叶科技有限公司 | Unsupervised self-evolving motor design method and system |
| CN112131787A (en) * | 2020-09-18 | 2020-12-25 | 江西兰叶科技有限公司 | Unsupervised self-evolving motor design method and system |
| CN112286749A (en) * | 2020-11-23 | 2021-01-29 | 四川长虹电器股份有限公司 | Method for detecting edge cloud fault |
| CN112834211A (en) * | 2020-12-31 | 2021-05-25 | 江苏国科智能电气有限公司 | Fault early warning method for transmission system of wind turbine generator |
| CN112906915A (en) * | 2021-01-22 | 2021-06-04 | 江苏安狮智能技术有限公司 | Rail transit system fault diagnosis method based on deep learning |
| CN112906915B (en) * | 2021-01-22 | 2024-03-22 | 江苏安狮智能技术有限公司 | Rail transit system fault diagnosis method based on deep learning |
| CN114936532A (en) * | 2022-07-20 | 2022-08-23 | 浙江大学 | Offshore wind turbine generator set fault diagnosis method |
| CN116484751A (en) * | 2023-06-21 | 2023-07-25 | 北京尚文汇通能源科技有限公司 | Fatigue life assessment method and device for wind turbine generator components |
| CN116484751B (en) * | 2023-06-21 | 2023-09-05 | 北京尚文汇通能源科技有限公司 | Fatigue life assessment method and device for wind turbine generator components |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110348513A (en) | A kind of Wind turbines failure prediction method based on deep learning | |
| CN110852655B (en) | Household electricity load decomposition system with solar power supply system and household electricity load decomposition method | |
| CN110298488A (en) | A kind of multi-energy data analysis method and system based on data mining | |
| CN113255210B (en) | Method and system for diagnosing federal fault of wind turbine generator | |
| Raman et al. | Energy monitoring in solar-powered buildings using internet of things | |
| CN117713221B (en) | Micro-inversion photovoltaic grid-connected optimization system | |
| Amayri et al. | Condition based maintenance of wind turbine systems considering different turbine types | |
| Fonte et al. | Wind speed prediction using artificial neural networks. | |
| CN114861539A (en) | Modeling and implementation method for 5G urban comprehensive energy Internet digital twin system | |
| CN112183877A (en) | Photovoltaic power station fault intelligent diagnosis method based on transfer learning | |
| CN117035154A (en) | Multi-element load prediction method and device for comprehensive energy system | |
| Wang et al. | A hybrid deep learning method based on CEEMDAN and attention mechanism for network traffic prediction | |
| Lin et al. | On-line monitoring and fault diagnosis of PV array based on BP neural network optimized by genetic algorithm | |
| CN116485139A (en) | Short-term photovoltaic power generation amount prediction method based on multi-feature fusion | |
| CN114320773B (en) | Wind turbine generator system fault early warning method based on power curve analysis and neural network | |
| Aikhuele et al. | Wind turbine systems operational state and reliability evaluation: An artificial neural network approach | |
| Sivaranjani et al. | Renewable Energy Generation and Grid Synchronization using IOT | |
| CN111639742A (en) | System and method for diagnosing state fault of desulfurization and denitrification circulating pump | |
| Lahiouel et al. | Photovoltaic fault diagnosis algorithm using fuzzy logic controller based on calculating distortion ratio of values | |
| Miao et al. | Energy Availability Analysis of Offshore Wind Farms Considering the Correlation between Wind Speed Cloud Model and Parameters | |
| Jiurong et al. | A data-driven fault diagnosis method for photovoltaic modules | |
| Zhao et al. | Stochastic Degradation Model and Preventive Maintenance Decision of Wind Turbines | |
| Shi-Bo et al. | Classification of bearing fault based on multi-class recurrent neural network | |
| Cococcioni et al. | Short-time forecasting of renewable production energy in solar photovoltaic installations | |
| Yang et al. | Photovoltaic power prediction based on optimized BP neural network |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191018 |
|
| RJ01 | Rejection of invention patent application after publication |