CN107608937A - A kind of machine learning fan condition monitoring method and device based on cloud computing platform - Google Patents
A kind of machine learning fan condition monitoring method and device based on cloud computing platform Download PDFInfo
- Publication number
- CN107608937A CN107608937A CN201710810976.9A CN201710810976A CN107608937A CN 107608937 A CN107608937 A CN 107608937A CN 201710810976 A CN201710810976 A CN 201710810976A CN 107608937 A CN107608937 A CN 107608937A
- Authority
- CN
- China
- Prior art keywords
- mtd
- extremelearningmachine
- layer
- msub
- mrow
- 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.)
- Granted
Links
Landscapes
- Feedback Control In General (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a kind of machine learning fan condition monitoring method and device based on cloud computing platform.The step of this method is:Layering ExtremeLearningMachine model structure is set to obtain the initial output matrix of first layer hidden layerH;Using compression sensing method to the initial output matrix of first layer hidden layer obtainedHIt is compressed processing;Data after compression are uploaded to cloud computing platform by Internet of Things;The compressed data for being uploaded to cloud computing platform is recovered;Complete the status monitoring model training of layering ExtremeLearningMachine method;Completed according to the blower fan data gathered in real time to blower fan subsystems real-time state monitoring.The present invention utilizes the data compression characteristic of compressed sensing, reduces the quantity that data are uploaded to cloud computing platform;And due to only uploading the layering initial output matrix of ExtremeLearningMachine first layer hidden layer, ensure forecast model structure and its parameter safety.There is higher monitoring accuracy compared to traditional ExtremeLearningMachine method with layering ExtremeLearningMachine method.
Description
Technical field
The present invention relates to remote state monitoring system, more particularly, to a kind of machine learning blower fan based on cloud computing platform
State monitoring method and device.
Background technology
In this year, in order to reduce greenhouse gas emission and pollution to environment, China has built substantial amounts of large scale wind hair
Power plant.However, the area that wind resource enriches often is in high mountain, desert or sea;Often weather conditions are disliked in these areas
Bad, unsuitable operating personnel are on duty for a long time, so install remote state monitoring system for wind-driven generator monitors blower fan fortune in real time
Row situation is necessary.One typical large-scale MW class wind turbine generally requires to monitor simultaneously more than 100
State variable, according to different variable species, signal sampling frequencies are between 100KHz-1Hz.This means that each second can produce
Raw substantial amounts of Monitoring Data needs to be processed, and the amount of calculation of machine-learning process is very huge.Traditional status monitoring system
System often sets a data processing centre to complete substantial amounts of data processing, and data processing centre is often equipped with large-scale service
Device, data machine learning, analysis and calculating are carried out by large server.At present, installed in China, some super-huge wind parks
Hundred typhoon power generators, it means that wind power plant needs to carry out state to typhoon power generators up to a hundred in real time simultaneously
Monitoring, this just proposes huge challenge to the computing capability of large server, and traditional large server computation capability
With calculating speed or limited.
In past ten years, cloud computing technology achieves very big development.It is compared to traditional server to calculate, cloud
Calculating has the advantages of can realizing ultra-large calculating, use cost is extremely cheap, and extended capability is strong.Therefore, based on cloud meter
The processing of typhoon power generator Monitoring Datas up to a hundred can be quickly completed by calculating the wind-driven generator condition monitoring system of platform, and
And operating personnel are transferred and read blower fan fortune at any time in any place using hand-held intelligent equipment (such as tablet personal computer or smart mobile phone)
Row status information.Due to that need not purchase large server, the wind-driven generator condition monitoring system based on cloud computing platform makes
With cost well below traditional server data handling system.
However, needing Monitoring Data being uploaded to high in the clouds by internet using cloud computing platform, data calculate and storage
All implement beyond the clouds, this just brings the risk of data message leakage, and the use cost of cloud computing platform depends on uploading
The quantity of data.Therefore, traditional machine learning method can not meet the needs of practical engineering application.
The content of the invention
For the deficiency of background technology, it is an object of the invention to provide a kind of machine learning wind based on cloud computing platform
Machine state monitoring method and device, the present invention have data compression, encryption and machine learning function simultaneously.Number can not only be reduced
According to upload amount, use cost is further reduced, while also assures that the safety of blower fan data, there is very high commercial Application valency
Value.
In order to achieve the above object, the technical solution adopted by the present invention is:、
First, a kind of machine learning fan condition monitoring method based on cloud computing platform, it is the step of this method:
Step 1) sets layering ExtremeLearningMachine model structure, extreme using layering to the signal after sensor collection conditioning
Learning machine method obtains the initial output matrix H of first layer hidden layer;
Step 2) is compressed processing using compression sensing method to the initial output matrix H of first layer hidden layer obtained, from
And reduce the dimension of matrix;
Step 3) is by the initial output matrix H of first layer hidden layer after compression by based on the system wireless of raspberry pi 3.0
Transport module is uploaded to cloud computing platform by Internet of Things;
Step 4) is recovered to the compressed data for being uploaded to cloud computing platform;
For step 5) by using cloud computing technology, completion is layered the status monitoring model training of ExtremeLearningMachine method, and
The structural parameters of the follow-up neural net layer in addition to the first layer of layering ExtremeLearningMachine after the completion of training are downloaded to
Line condition monitoring system;
Step 6) is completed to blower fan subsystems real-time status using the optimum state monitoring model after the training established
Monitoring.
The initial output matrix H of first layer hidden layer is obtained using layering ExtremeLearningMachine method in the step 1), works as layering
After determination is set in ExtremeLearningMachine model structure, with the wind speed after sensor collection conditioning, generated output power, Yi Jiqi
Warm signal calculates the initial output matrix H of first layer hidden layer as the input signal of layering ExtremeLearningMachine method meter;
There is the first layer hidden layer of layering ExtremeLearningMachine the hidden layer neural network of L hidden neuron to export:
In formula (1), g (ωi, bi, xj) it is output with corresponding i-th of the hidden neuron of input x, ωiIt is input power
Vector, biIt is neutral net deviant vector, g () is activation primitive, and matrix H claims the first layer hidden layer for being layered ExtremeLearningMachine
Output matrix.
The step 2) is compressed place using compression sensing method to the initial output matrix H of first layer hidden layer obtained
Reason, the mathematic(al) representation such as formula (2) of compression sensing method:
Y=Φ x=Φ Ψ θ(2)
Y=A θ (3)
In formula:Y represents the data after compression;Φ represents calculation matrix;A represents to perceive matrix;θ represents x in Ψ transform domains
Rarefaction representation after result;X is data to be compressed, and variable x is that the first layer of layering ExtremeLearningMachine is hidden in the present invention
The initial output matrix H of layer.
The step 4) is recovered to the compressed data for being uploaded to cloud computing platform, uploads the data of cloud computing platform and makes
With minimum norm l1Algorithm carries out data reconstruction, so as to which the first layer hidden layer of the layering ExtremeLearningMachine after being restored is initially defeated
Go out matrix H, so as to carry out the training of the status monitoring model in next step based on layering ExtremeLearningMachine;Minimum norm l1Algorithm is used
Equation below (4) is stated:
min||x||1s.t.||Φx-y||2≤ε (4)
The essence of formula (4) is exactly in the case where calculation matrix determines, passes through minimum norm l1Best practice find
Correct sparse solutionSo as to reconstruct primary signal, ε is error amount.
The step 5) by using cloud computing technology, instruct by the status monitoring model for completing layering ExtremeLearningMachine method
Practice, the status monitoring model based on layering ExtremeLearningMachine method is with wind speed, generated output output, and temperature is input quantity, model
Output quantity is the blower fan subsystems running status of prediction, i.e. temperature, voltage, electric current, is vibrated;Model predication value shows with blower fan
Field actual measured value compares, and when both obvious deviation occur, shows that failure occurs in blower fan;Proceeding by model prediction
Before, forecast model need be trained to so that obtain optimal forecast model structure, this method comprises the following steps that:
First step:The structure of layering ExtremeLearningMachine model and its initial each layer weight vector and offset vector ginseng are set
Number, and select a healthy blower fan data to be used as and take training sample;
Second step:Last layer of hidden layer output matrix H is calculated using sampleN;
Third step:The output weights β of N-1 layer hidden layers before calculating respectivelyN-1;
Four steps:Finally calculate last layer of hidden layer output weights
λ is one on the occasion of constant coefficient, HNBe be layered ExtremeLearningMachine n-th layer hidden neuron hidden layer neural network it is defeated
Go out;T is the target output for being layered ExtremeLearningMachine.
2nd, a kind of device of the machine learning fan condition monitoring method based on cloud computing platform:
Including sensor and signal conditioning circuit, STM32F429 signal-processing boards, the microcomputers of raspberry pi 3.0
And cloud computing platform;The field data gathered in real time is transferred at STM32F429 signals by sensor and signal conditioning circuit
Plate is managed, the signal-processing board completes the analog-to-digital conversion of data, and the first layer hidden layer for calculating layering ExtremeLearningMachine is initially defeated
Go out matrix H, reuse compression sensing method and matrix H data are compressed, the data after compression sensing method by
STM32F429 signal-processing boards are transferred to the microcomputers of raspberry pi 3.0, then micro- by raspberry pi 3.0
Wireless transport module on type computer is uploaded to cloud computing platform by Internet of Things.
The invention has the advantages that:
The present invention utilizes the data compression characteristic of compressed sensing, reduces the quantity that data are uploaded to cloud computing platform;And
And due to only uploading the layering initial output matrix of ExtremeLearningMachine first layer hidden layer, so ExtremeLearningMachine inputs weights and first
The offset vector of layer hidden layer can not be obtained by other people, ensure that forecast model structure and its parameter safety.Learned with layering is extreme
Habit machine method has higher monitoring accuracy compared to traditional ExtremeLearningMachine method.
Brief description of the drawings
Fig. 1 is the flow chart of the machine learning fan condition monitoring method based on cloud computing platform of the present invention.
The multilayer neural network wind-driven generator status monitoring based on layering ExtremeLearningMachine that Fig. 2 is the present invention opens up benefit knot
Composition.
Fig. 3 is the construction block diagram of the machine learning fan condition monitoring device based on cloud computing platform of the present invention.
Embodiment
The invention will be further described with reference to the accompanying drawings and examples.
As shown in figure 1, a kind of machine learning fan condition monitoring method based on cloud computing platform, the step of this method
It is:
Step 1) sets layering ExtremeLearningMachine model structure, extreme using layering to the signal after sensor collection conditioning
Learning machine method obtains the initial output matrix H of first layer hidden layer;
Step 2) is compressed processing using compression sensing method to the initial output matrix H of first layer hidden layer obtained, from
And reduce the dimension of matrix;
Step 3) is by the initial output matrix H of first layer hidden layer after compression by based on the system wireless of raspberry pi 3.0
Transport module is uploaded to cloud computing platform by Internet of Things;
Step 4) is recovered to the compressed data for being uploaded to cloud computing platform;
For step 5) by using cloud computing technology, completion is layered the status monitoring model training of ExtremeLearningMachine method, and
The structural parameters of the follow-up neural net layer in addition to the first layer of layering ExtremeLearningMachine after the completion of training are downloaded to
Line condition monitoring system;
Step 6) is completed to blower fan subsystems real-time status using the optimum state monitoring model after the training established
Monitoring.
The initial output matrix H of first layer hidden layer is obtained using layering ExtremeLearningMachine method in the step 1), works as layering
After determination is set in ExtremeLearningMachine model structure, with the wind speed after sensor collection conditioning, generated output power, Yi Jiqi
Warm signal calculates the initial output matrix H of first layer hidden layer as the input signal of layering ExtremeLearningMachine method meter;
As shown in Fig. 2 the first layer hidden layer of layering ExtremeLearningMachine has the hidden layer neural network of L hidden neuron defeated
Go out for:
In formula (1), g (ωi, bi, xj) it is output with corresponding i-th of the hidden neuron of input x, ωiIt is input power
Vector, biIt is neutral net deviant vector, g () is activation primitive, and matrix H claims the first layer hidden layer for being layered ExtremeLearningMachine
Output matrix.
The first layer hidden layer neural network for being layered ExtremeLearningMachine exports H as next layer of input quantity, and extends to N
Layer, statement such as formula (5)
HN=g (HN-1·βN-1) (5)
The output weights β of preceding N-1 layers hidden layerN-1It can be tried to achieve by formula (6) and (7)
Finally, n-th layer output weight vectorIt can be calculated
λ is one on the occasion of constant coefficient, H in formula (8)NIt is the hidden layer for the n-th layer hidden neuron for being layered ExtremeLearningMachine
Neural network exports;T is the target output for being layered ExtremeLearningMachine.
In the layering ExtremeLearningMachine training process of the present invention, temperature, wind speed, generated output power are as layering pole
Hold the input variable of learning machine;Need output variable of the monitored object as layering ExtremeLearningMachine.By training process,
Optimal n-th layer output weight vector can be obtainedSo as to obtain optimal layering ExtremeLearningMachine status monitoring model.
The step 2) is compressed place using compression sensing method to the initial output matrix H of first layer hidden layer obtained
Reason, the mathematic(al) representation such as formula (2) of compression sensing method:
Y=Φ x=Φ Ψ θ (2)
Y=A θ (3)
In formula:Y represents the data after compression;Φ represents calculation matrix;A represents to perceive matrix;θ represents x in Ψ transform domains
Rarefaction representation after result;X is data to be compressed, and variable x is that the first layer of layering ExtremeLearningMachine is hidden in the present invention
The initial output matrix H of layer.
Due to compression sensing method have compression amount of calculation it is small the advantages of, so the step of data compression can based on
Implement in the STM32F429 ARM systems of ST Microelectronics.When the first layer hidden layer of layering ExtremeLearningMachine initially exports
After what matrix H calculated obtains, it is transferred in STM32F429 ARM systems, the system completes the analog-to-digital conversion of data, uses compression
Cognitive method completes data compression.Data digital signal after compression is transferred to raspberry by STM32F429 ARM systems
The systems of pi 3.0, cloud computing is then uploaded to by Internet of Things by the wireless transport module in the systems of raspberry pi 3.0
Platform.The systems of raspberry pi 3.0 are a kind of microcomputer systems.Compared to traditional computer, raspberry pi
3.0 system prices are cheaply a lot, are only equivalent to the 5-10% of traditional computer Platform Price.
In traditional data compression method, the amount of calculation of compressed data is far longer than the amount of calculation of data decompression, then
Compression sensing method then has opposite characteristic, and its data compression amount of calculation is far smaller than data decompression reconstruction calculations amount.This
Feature is highly suitable for field of cloud calculation, is calculated since it is desired that computationally intensive process is placed on Cloud Server.Use
The initial output matrix H of first layer hidden layer of compression sensing method compress-layering ExtremeLearningMachine, which is not only reduced, uploads cloud platform
Data volume, and avoid and directly upload layering ExtremeLearningMachine first layer input weight vector and offset vector, it ensure that pole
Hold the data safety of learning machine model structure parameter.
The step 4) is recovered to the compressed data for being uploaded to cloud computing platform, uploads the data of cloud computing platform and makes
With minimum norm l1Algorithm carries out data reconstruction, so as to which the first layer hidden layer of the layering ExtremeLearningMachine after being restored is initially defeated
Go out matrix H, so as to carry out the training of the status monitoring model in next step based on layering ExtremeLearningMachine;Minimum norm l1Algorithm is used
Equation below (4) is stated:
min||x||1s.t.||Φx-y||2≤ε (4)
The essence of formula (4) is exactly in the case where calculation matrix determines, passes through minimum norm l1Best practice find
Correct sparse solutionSo as to reconstruct primary signal, ε is error amount.Compared to the process of compressed data, restructuring procedure needs
Substantial amounts of feedback iteration calculating is carried out, therefore amount of calculation is far longer than data compression process.
ExtremeLearningMachine is a kind of very effective neural network training method, widely applies to neural networks with single hidden layer
Structure.ExtremeLearningMachine only needs random generation input weights and hidden layer threshold value, and according to mole generalized inverse matrix theory, parsing is asked
The Minimal Norm Least Square Solutions of equation group are obtained, output weights can is uniquely determined.Most of all, ExtremeLearningMachine
Optimal solution can just be obtained by being not required to want the traditional neural network training calculating that needs to iterate like that, therefore ExtremeLearningMachine overcomes
The inherent defect of traditional neural network training method has very fast training speed, and has a more preferable Generalization Capability
Energy.But the ExtremeLearningMachine method based on neural networks with single hidden layer has many limitations in practical engineering application.Work as application
Under conditions of complicated and strong nonlinearity, traditional method based on neural networks with single hidden layer structure is difficult to reach preferable essence
Degree.Therefore layering of the sampling based on multilayer hidden layer neutral net can extremely obtain higher precision, be suitably applied in blower fan shape
State monitors field.
The step 5) by using cloud computing technology, instruct by the status monitoring model for completing layering ExtremeLearningMachine method
Practice, the status monitoring model based on layering ExtremeLearningMachine method is with wind speed, generated output output, and temperature is input quantity, model
Output quantity is the blower fan subsystems running status of prediction, i.e. temperature, voltage, electric current, is vibrated;Model predication value shows with blower fan
Field actual measured value compares, and when both obvious deviation occur, shows that failure occurs in blower fan;Proceeding by model prediction
Before, forecast model need be trained to so that obtain optimal forecast model structure, this method comprises the following steps that:
First step:The structure of layering ExtremeLearningMachine model and its initial each layer weight vector and offset vector ginseng are set
Number, and select a healthy blower fan data to be used as and take training sample;
Second step:Last layer of hidden layer output matrix H is calculated using sampleN;
Third step:The output weights β of N-1 layer hidden layers before calculating respectivelyN-1;
Four steps:Finally calculate last layer of hidden layer output weights
λ is one on the occasion of constant coefficient, HNBe be layered ExtremeLearningMachine n-th layer hidden neuron hidden layer neural network it is defeated
Go out;T is the target output for being layered ExtremeLearningMachine.
As shown in figure 3, it is the construction block diagram of the machine learning fan condition monitoring device based on cloud computing platform of the present invention.
Including sensor and signal conditioning circuit, STM32F429 signal-processing boards, the microcomputers of raspberry pi 3.0 and cloud computing
Platform;The field data gathered in real time is transferred to STM32F429 signal-processing boards by sensor and signal conditioning circuit, the letter
Number process plate completes the analog-to-digital conversion of data, and calculates the initial output matrix H of first layer hidden layer of layering ExtremeLearningMachine, then
Matrix H data are compressed using compression sensing method, the data after compression sensing method are by STM32F429 signals
Reason plate is transferred to the microcomputers of raspberry pi 3.0, then passes through the wireless biography on the microcomputers of raspberry pi 3.0
Defeated module is uploaded to cloud computing platform by Internet of Things.
Above-mentioned embodiment is used for illustrating the present invention, rather than limits the invention, the present invention's
In spirit and scope of the claims, to any modifications and changes of the invention made, protection model of the invention is both fallen within
Enclose.
Claims (6)
1. a kind of machine learning fan condition monitoring method based on cloud computing platform, it is characterised in that be the step of this method:
Step 1) sets layering ExtremeLearningMachine model structure, and the extreme study of layering is used to the signal after sensor collection conditioning
Machine method obtains the initial output matrix H of first layer hidden layer;
Step 2) is compressed processing using compression sensing method to the initial output matrix H of first layer hidden layer obtained, so as to subtract
The dimension of few matrix;
Step 3) is by the initial output matrix H of first layer hidden layer after compression by being transmitted based on the system wireless of raspberry pi 3.0
Module is uploaded to cloud computing platform by Internet of Things;
Step 4) is recovered to the compressed data for being uploaded to cloud computing platform;
Step 5) completes the status monitoring model training of layering ExtremeLearningMachine method, and will instruct by using cloud computing technology
The structural parameters of the follow-up neural net layer in addition to the first layer of layering ExtremeLearningMachine after the completion of white silk are downloaded in wire
State monitoring system;
Step 6) is completed to blower fan subsystems real-time state monitoring using the optimum state monitoring model after the training established.
2. a kind of machine learning fan condition monitoring method based on cloud computing platform according to claim 1, its feature
It is:The initial output matrix H of first layer hidden layer is obtained using layering ExtremeLearningMachine method in the step 1), when layering is extreme
After determination is set in learning machine model structure, with the wind speed after sensor collection conditioning, generated output power, and temperature letter
Number as layering ExtremeLearningMachine method meter input signal calculate the initial output matrix H of first layer hidden layer;
There is the first layer hidden layer of layering ExtremeLearningMachine the hidden layer neural network of L hidden neuron to export:
<mrow>
<mi>H</mi>
<mo>=</mo>
<msub>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&omega;</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>b</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>x</mi>
<mn>1</mn>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
<mtd>
<mn>...</mn>
</mtd>
<mtd>
<mrow>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&omega;</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>b</mi>
<mi>L</mi>
</msub>
<mo>,</mo>
<msub>
<mi>x</mi>
<mn>1</mn>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mrow></mrow>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mn>...</mn>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mo>.</mo>
</mtd>
<mtd>
<mrow></mrow>
</mtd>
<mtd>
<mo>.</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&omega;</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>b</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>x</mi>
<mi>N</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
<mtd>
<mn>...</mn>
</mtd>
<mtd>
<mrow>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&omega;</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>b</mi>
<mi>L</mi>
</msub>
<mo>,</mo>
<msub>
<mi>x</mi>
<mi>N</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mrow>
<mi>N</mi>
<mo>&times;</mo>
<mi>L</mi>
</mrow>
</msub>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
In formula (1), g (ωi, bi, xj) it is output with corresponding i-th of the hidden neuron of input x, ωiIt is input weight vector,
biIt is neutral net deviant vector, g () is activation primitive, and matrix H claims the first layer hidden layer output for being layered ExtremeLearningMachine
Matrix.
3. a kind of machine learning fan condition monitoring method based on cloud computing platform according to claim 1, its feature
It is:The step 2) is compressed processing using compression sensing method to the initial output matrix H of first layer hidden layer obtained, pressure
The mathematic(al) representation of contracting cognitive method such as formula (2):
Y=Φ x=Φ Ψ θ (2)
Y=A θ (3)
In formula:Y represents the data after compression;Φ represents calculation matrix;A represents to perceive matrix;θ represents x in the dilute of Ψ transform domains
Dredge the result after representing;X is data to be compressed, and variable x is at the beginning of being layered the first layer hidden layer of ExtremeLearningMachine in the present invention
Beginning output matrix H.
4. a kind of machine learning fan condition monitoring method based on cloud computing platform according to claim 1, its feature
It is:The step 4) is recovered to the compressed data for being uploaded to cloud computing platform, and the data for uploading cloud computing platform use
Minimum norm l1Algorithm carries out data reconstruction, so as to which the first layer hidden layer of the layering ExtremeLearningMachine after being restored initially exports
Matrix H, so as to carry out the training of the status monitoring model in next step based on layering ExtremeLearningMachine;Minimum norm l1Algorithm is with such as
Lower formula (4) statement:
min||x||1s.t.||Φx-y||2≤ε (4)
The essence of formula (4) is exactly in the case where calculation matrix determines, passes through minimum norm l1Best practice find correctly
Sparse solutionSo as to reconstruct primary signal, ε is error amount.
5. a kind of machine learning fan condition monitoring method based on cloud computing platform according to claim 1, its feature
It is:The step 5) is completed to be layered the status monitoring model training of ExtremeLearningMachine method, base by using cloud computing technology
In layering ExtremeLearningMachine method status monitoring model with wind speed, generated output output, temperature is input quantity, model output
For the blower fan subsystems running status of prediction, i.e. temperature, voltage, electric current, vibration;Model predication value and blower fan scene are actual
Measured value compares, and when both obvious deviation occur, shows that failure occurs in blower fan;Before model prediction is proceeded by,
Forecast model need be trained to so that obtain optimal forecast model structure, this method comprises the following steps that:
First step:The structure and its initial each layer weight vector and offset vector parameter of layering ExtremeLearningMachine model are set,
And select a healthy blower fan data to be used as and take training sample;
Second step:Last layer of hidden layer output matrix H is calculated using sampleN;
Third step:The output weights β of N-1 layer hidden layers before calculating respectivelyN-1;
Four steps:Finally calculate last layer of hidden layer output weights
λ is one on the occasion of constant coefficient, HNIt is the hidden layer neural network output for the n-th layer hidden neuron for being layered ExtremeLearningMachine;T
It is the target output for being layered ExtremeLearningMachine.
6. it is used for a kind of device of machine learning fan condition monitoring method based on cloud computing platform described in claim 1,
It is characterized in that:It is miniature including sensor and signal conditioning circuit, STM32F429 signal-processing boards, raspberry pi 3.0
Computer and cloud computing platform;The field data gathered in real time is transferred to STM32F429 letters by sensor and signal conditioning circuit
Number process plate, the signal-processing board complete the analog-to-digital conversion of data, and at the beginning of calculating the first layer hidden layer of layering ExtremeLearningMachine
Beginning output matrix H, reuse compression sensing method and matrix H data be compressed, the data after compression sensing method by
STM32F429 signal-processing boards are transferred to the microcomputers of raspberry pi 3.0, then micro- by raspberry pi 3.0
Wireless transport module on type computer is uploaded to cloud computing platform by Internet of Things.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710810976.9A CN107608937B (en) | 2017-09-11 | 2017-09-11 | Machine learning fan state monitoring method and device based on cloud computing platform |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710810976.9A CN107608937B (en) | 2017-09-11 | 2017-09-11 | Machine learning fan state monitoring method and device based on cloud computing platform |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107608937A true CN107608937A (en) | 2018-01-19 |
CN107608937B CN107608937B (en) | 2020-08-18 |
Family
ID=61063317
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710810976.9A Active CN107608937B (en) | 2017-09-11 | 2017-09-11 | Machine learning fan state monitoring method and device based on cloud computing platform |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107608937B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109755683A (en) * | 2018-12-04 | 2019-05-14 | 厦门大学 | A kind of battery pack internal temperature method of real-time based on compressive sensing theory |
CN109800869A (en) * | 2018-12-29 | 2019-05-24 | 深圳云天励飞技术有限公司 | Data compression method and relevant apparatus |
CN110472684A (en) * | 2019-08-14 | 2019-11-19 | 树根互联技术有限公司 | A kind of icing monitoring method of fan blade, its device and readable storage medium storing program for executing |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102706560A (en) * | 2012-05-25 | 2012-10-03 | 华锐风电科技(江苏)有限公司 | State monitoring method and device of wind turbine generator set |
CN103195727A (en) * | 2013-03-05 | 2013-07-10 | 天津理工大学 | Device for on-line monitoring and evaluating state of air blower |
US20150095004A1 (en) * | 2013-09-27 | 2015-04-02 | Korea Electric Power Corporation | Apparatus for simulating wind power farm |
CN105243259A (en) * | 2015-09-02 | 2016-01-13 | 上海大学 | Extreme learning machine based rapid prediction method for fluctuating wind speed |
-
2017
- 2017-09-11 CN CN201710810976.9A patent/CN107608937B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102706560A (en) * | 2012-05-25 | 2012-10-03 | 华锐风电科技(江苏)有限公司 | State monitoring method and device of wind turbine generator set |
CN103195727A (en) * | 2013-03-05 | 2013-07-10 | 天津理工大学 | Device for on-line monitoring and evaluating state of air blower |
US20150095004A1 (en) * | 2013-09-27 | 2015-04-02 | Korea Electric Power Corporation | Apparatus for simulating wind power farm |
CN105243259A (en) * | 2015-09-02 | 2016-01-13 | 上海大学 | Extreme learning machine based rapid prediction method for fluctuating wind speed |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109755683A (en) * | 2018-12-04 | 2019-05-14 | 厦门大学 | A kind of battery pack internal temperature method of real-time based on compressive sensing theory |
CN109800869A (en) * | 2018-12-29 | 2019-05-24 | 深圳云天励飞技术有限公司 | Data compression method and relevant apparatus |
CN109800869B (en) * | 2018-12-29 | 2021-03-05 | 深圳云天励飞技术有限公司 | Data compression method and related device |
CN110472684A (en) * | 2019-08-14 | 2019-11-19 | 树根互联技术有限公司 | A kind of icing monitoring method of fan blade, its device and readable storage medium storing program for executing |
CN110472684B (en) * | 2019-08-14 | 2022-02-08 | 树根互联股份有限公司 | Method and device for monitoring icing of fan blade and readable storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN107608937B (en) | 2020-08-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Xiaoyun et al. | Short-term prediction of wind power based on deep long short-term memory | |
Yan et al. | Forecasting the high penetration of wind power on multiple scales using multi-to-multi mapping | |
CN106529719B (en) | Wind power prediction method based on particle swarm optimization algorithm wind speed fusion | |
CN107194625B (en) | Wind power plant wind curtailment electric quantity evaluation method based on neural network | |
CN111753893A (en) | Wind turbine generator power cluster prediction method based on clustering and deep learning | |
CN101793907A (en) | Short-term wind speed forecasting method of wind farm | |
CN107730044A (en) | A kind of hybrid forecasting method of renewable energy power generation and load | |
CN114493050B (en) | Multi-dimensional fusion new energy power parallel prediction method and device | |
CN107608937A (en) | A kind of machine learning fan condition monitoring method and device based on cloud computing platform | |
CN112991090B (en) | Photovoltaic power prediction method based on transducer model | |
CN107092989A (en) | The Forecasting Methodology and equipment of short-term wind-electricity power | |
CN115345076B (en) | Wind speed correction processing method and device | |
Zhao et al. | Deep learning based model-free robust load restoration to enhance bulk system resilience with wind power penetration | |
CN116307291B (en) | Distributed photovoltaic power generation prediction method and prediction terminal based on wavelet decomposition | |
CN109146192B (en) | Wind power prediction method considering operation condition of wind turbine generator | |
Chen et al. | Research on wind power prediction method based on convolutional neural network and genetic algorithm | |
Zhang et al. | Short-term PV output power forecasting based on CEEMDAN-AE-GRU | |
CN115293424A (en) | New energy maximum power generation capacity calculation method, terminal and storage medium | |
CN115659844B (en) | Simulation method and device based on wind power plant dynamic model and computer equipment | |
CN106295175A (en) | Station meteorological data missing value fill method based on svd algorithm | |
CN113627670B (en) | Ultra-short-term photovoltaic power prediction system based on deep Fourier network | |
Wan et al. | Turbine location wind speed forecast using convolutional neural network | |
CN113887843B (en) | Group prediction method and system for wind power plant output power | |
Zheng et al. | Short Term Wind Power Prediction Based on Wavelet Transform and BP Neural Network | |
CN116231645B (en) | Offshore wind farm power generation amount calculation method, calculation system and calculation terminal |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |