CN110298485A - Based on the pitch-controlled system failure prediction method for improving depth random forests algorithm - Google Patents
Based on the pitch-controlled system failure prediction method for improving depth random forests algorithm Download PDFInfo
- Publication number
- CN110298485A CN110298485A CN201910456862.8A CN201910456862A CN110298485A CN 110298485 A CN110298485 A CN 110298485A CN 201910456862 A CN201910456862 A CN 201910456862A CN 110298485 A CN110298485 A CN 110298485A
- Authority
- CN
- China
- Prior art keywords
- pitch
- controlled system
- prediction method
- random forest
- depth
- 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
- 238000007637 random forest analysis Methods 0.000 title claims abstract description 53
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 35
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000003066 decision tree Methods 0.000 claims abstract description 11
- 238000012549 training Methods 0.000 claims description 20
- 239000013598 vector Substances 0.000 claims description 20
- 238000004364 calculation method Methods 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 6
- 238000002790 cross-validation Methods 0.000 claims description 5
- 238000012706 support-vector machine Methods 0.000 claims description 4
- 238000007476 Maximum Likelihood Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000013459 approach Methods 0.000 claims description 2
- 238000007418 data mining Methods 0.000 claims description 2
- 238000011156 evaluation Methods 0.000 claims description 2
- 238000012216 screening Methods 0.000 claims 1
- 230000006872 improvement Effects 0.000 abstract description 5
- 230000008901 benefit Effects 0.000 abstract description 4
- 238000012423 maintenance Methods 0.000 abstract description 4
- 230000008439 repair process Effects 0.000 abstract description 3
- 230000005856 abnormality Effects 0.000 abstract description 2
- 230000004888 barrier function Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 12
- 238000012544 monitoring process Methods 0.000 description 9
- 235000019580 granularity Nutrition 0.000 description 6
- 241001269238 Data Species 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000005457 optimization Methods 0.000 description 3
- 238000012937 correction Methods 0.000 description 2
- 230000009977 dual effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000007257 malfunction Effects 0.000 description 2
- 101150110330 CRAT gene Proteins 0.000 description 1
- 102100036357 Carnitine O-acetyltransferase Human genes 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000025518 detection of mechanical stimulus involved in sensory perception of wind Effects 0.000 description 1
- 230000005684 electric field Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000010977 unit operation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—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
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02B—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO BUILDINGS, e.g. HOUSING, HOUSE APPLIANCES OR RELATED END-USER APPLICATIONS
- Y02B10/00—Integration of renewable energy sources in buildings
- Y02B10/30—Wind power
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Primary Health Care (AREA)
- General Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A kind of pitch-controlled system failure prediction method based on improvement depth random forests algorithm, wherein, depth forest algorithm includes multiple cascade modules, each cascade module includes parallel arrangement of completely random forest module and random forest module, and be all made of decision tree and carry out decision, above-mentioned failure prediction method includes the following steps: to build the fault prediction model based on the pitch-controlled system for improving depth random forests algorithm；Failure predication is carried out to wind power pitch-controlled system using fault prediction model；The EARLY RECOGNITION of Wind turbines abnormality is realized based on the pitch-controlled system failure prediction method for improving depth random forests algorithm, accident barrier driving maintenance reduces equipment failure rate and downtime, guarantee maximization of economic benefit actively to repair.
Description
Technical field
The invention belongs to wind power plants and technology monitoring field, relate generally to the detection of wind power pitch-controlled system state,
Specifically a kind of pitch-controlled system failure prediction method based on improvement depth random forests algorithm.
Background technique
The speed-changing oar-changing system of wind power generating set is the higher subsystem of failure rate in wind power generating set.If there is
Disorderly closedown is operated normally from overhauling to restoring, is takeed a long time.If failure has occurred in the wind speed preferable period
It shuts down, the generated energy of blower will be largely effected on, to cause huge economic loss to enterprise.If can occur in fan trouble
When preceding, by the Condition Monitoring Technology of Wind turbines, after being monitored and analyzed to the operating parameter of pitch-controlled system to failure into
Row prediction, it will greatly improve the availability of blower.Based on factors above, failure predication is carried out to wind power generating set, it is right
The economy and safety that improve wind power plant operation have the meaning of positive important.
Currently, the common data acquisition of wind power plant and monitoring control (Supervisory Control and Data
Acquisition, SCADA) system, the running equipment at scene can be monitored and be controlled, realize data acquisition, equipment control
System, the functions such as parameter regulation and various types of signal alarm.But this can not carry out early warning to failure, it can only be after the failure occurred
It alarms, and the fault message usually contains multiple failure causes, so fault category and reason can not be accurately located, leads
It causes the repair time longer, reduces generated energy.With the increase of the installed capacity of wind power generating set, wind-power electricity generation company is to power generation
The requirement of efficiency is higher and higher, therefore establishes the act that quickly and effectively fault early warning method is certainty.
Summary of the invention
The purpose of the present invention is design a kind of based on improving depth random forests algorithm in view of the deficiencies of the prior art
Pitch-controlled system failure prediction method.
A kind of pitch-controlled system failure prediction method based on improvement depth random forests algorithm, wherein depth forest algorithm
Including multiple cascade modules, each cascade module includes parallel arrangement of completely random forest module and random forest module, and
It is all made of decision tree and carries out decision, above-mentioned failure prediction method includes the following steps:
1) fault prediction model based on the pitch-controlled system for improving depth random forests algorithm, the side of building of the model are built
Method is as follows:
1-1) sample data is chosen from the history SCADA data that the operation of practical wind field generates；
Feature Selection 1-2) is carried out using the feature selection approach combined based on expertise and data mining algorithm, really
Determine the input variable and output variable of fault prediction model；
Processing 1-3) is weighted to each sample data；
Sliding sampling 1-4) is carried out to sample data by the sampling window that a length is k, obtains subsample vector；
1-5) each subsample by cascade module completely random forest and random forest training respectively generate a table
It levies vector H (x), the output for being superimposed two forests obtains feature vector ZiAs input, for training depth random forest；
1-6) in training process to prevent over-fitting, every layer depth random forest can all carry out cross validation, when verifying
Accuracy do not promoted compared to preceding layer, then training terminates, and obtains fault prediction model；
2) failure predication is carried out to wind power pitch-controlled system using fault prediction model obtained in step 1-5).
Further, the step 1-2) in carry out the method for Feature Selection are as follows: first with expertise to all
Feature carries out primary dcreening operation, forms candidate feature set；Then using support vector machines combination sweep backward strategy to candidate characteristic set
Conjunction is further screened.
Further, the step 1-3) in weighting when weight determination method are as follows: weight wiBy Logic Regression Models
It generates, Logic Regression Models are as follows:
Wherein g (x)=w0+w1x1+...+wpxp,
The likelihood function of Logic Regression Models are as follows:
With maximum likelihood estimate, parameter w is solved0, w1..., wp, this group of parameter is exactly required weight,
Logarithm is taken to obtain likelihood equation:
To parameter wkLocal derviation is sought, following equations are obtained,
Parameter w can be solved by solving equation0, w1..., wpAs weight, wherein p indicates that sample dimension, y are the sight of its sample
Measured value.
Further, the decision tree is using the regression tree in cart.
Further, the output of the random forest and completely random forest is equal are as follows:
In formula, H (x) is final result, hiFor i-th of decision tree classification as a result, Y is true classification, I is metric function, N
For the quantity of decision tree.
Further, the step 1-6) when being trained to depth random forest, integrated using following index
It judges, error is smaller, and coefficient is higher, shows that the performance of depth random forest is better, when each performance indicator is all satisfied setting condition
The calculation method of Shi Caineng deconditioning, each judgment criteria is as follows:
(1) mean absolute error MAE calculation formula:
(2) root-mean-square error RMSE calculation formula:
(3) coefficient of determination R2Calculation formula:
(4) coefficient of determination R after correctinga 2Calculation formula:
Wherein, n is sample size, and p is characterized quantity, yiFor actual value,For predicted value,For being averaged for actual value
Value.
Further, when the step 2) carries out failure predication to wind power pitch-controlled system, by the defeated of fault prediction model
It is compared out with setting early warning value, sends a signal to monitoring center if being more than early warning value.
Compared with prior art, wind power pitch-controlled system failure predication provided by the invention has the advantage that realization wind-powered electricity generation
The EARLY RECOGNITION of unit abnormality, accident barrier driving maintenance reduce equipment failure rate and downtime, guarantee actively to repair
Maximization of economic benefit；Failure prediction method based on data does not need or only needs the priori knowledge of a small amount of objective system,
It is based primarily upon actual operating data, excavating incipient fault information by machine learning algorithm more has practicability；It is not only wind
Electric field monitoring personnel provides real-time operating states of the units information, and the exception of unit operation can be found before failure occurs
State and alarm simultaneously notify related maintenance personal, equipment damage, compressor emergency shutdown caused by avoiding because of failure, reduce loss.More
Granularity scan phase distinguishes the significance level of each feature, by increasing weight to feature so that the characterization ability of feature
It is stronger.It is weighted by the output to cascade forest, overcomes feature vector redundancy present in original depth forest model
Problem, while every layer of Enhanced feature vector is reduced to 4 dimensions by 8 original dimensions, while reducing algorithm space complexity
Improve operating rate.
Data acquisition unit extracts pitch-controlled system operation data from SCADA in real time, inputs trained based on improved
It is analyzed in the fault prediction model that depth random forest is established, compares whether residual error curve exceeds given threshold, if being more than
Monitoring center is then sent a signal to, early warning is carried out, realizes the intellectual monitoring of pitch-controlled system, monitoring velocity is high, and accuracy rate is high.
Detailed description of the invention
Fig. 1 is depth random forests algorithm flow chart；
Fig. 2 is pitch-controlled system monitoring figure；
Fig. 3 is improved more granularity Scan Architecture figures；
Fig. 4 is improved cascade forest structure；
Specific embodiment
The present invention is described in more detail with reference to the accompanying drawing:
Based on the SCADA data that the operation of practical wind field generates, before training set and selection failure that acquisition normal data is used as
30 minutes data as test set as experiment sample, and data prediction.The data prediction is carried out to data
Cleaning, that is, give up following data, the number under wrong data, unreasonable data and unusual service condition when including generation communication failure
According to.Data that treated are for making training set and test set.
The feature extracting method is divided into two stages: the first stage combines existing expertise to select candidate feature
Collection, gained candidate feature not only has more priori knowledge, but also facilitates the understanding of engineering operation maintenance personnel；Second stage utilizes
Support vector machines combination sweep backward strategy, which further screens feature, to be obtained important feature and determines input and output.
By the wind speed in sample, wind speed round, generator bearing a temperature, generator bearing b temperature, temperature, machine outside cabin
Cabin temperature, generator speed, 1 temperature of variable pitch axle box are inputted as regression model, and propeller pitch angle and generator active power return mould
Type output.
For a linearly inseparable problem, (xi,yi)(xi∈Rp,yi∈ { -1,1 }, i=1,2 ... it n) indicates by n
The training set of sample and p feature composition, and yiIt is training sample xiCategory, support vector cassification problem equivalent is in one
A optimization problem, objective function are as follows:
Wherein, J is objective function, and w is characterized weight, and C is penalty coefficient, ξ iiFor relaxation factor, i=1,2 ... n,
The classification results of sample x pass through decision function f (x)=yi(wTxi+ b) it determines, wherein w=[w1,w2,...,wp] it is feature
Weight vectors.
Introduce Lagrangian
Wherein, 0≤αi≤ C, i=1,2 ... n.
L is minimized to w and b difference derivation, can be obtained according to KKT:
Formula (b) and (c) are substituted into formula (a), obtain the dual problem of Lagrangian:
In order to avoid the inner product operation of higher-dimension, inner product calculating is carried out using kernel function:
K(xi,xj)=Φ (xi)Φ(xj)
After replacing inner product using kernel function, former dual problem becomes:
Corresponding classification problem also becomes:
By removing feature, objective function changes into Δ J (i), considers ith feature pairInfluence, benefit
With second order Taylor series expansion:
Single order local derviationWhen, objective function is optimal, and single order item can be ignored, and above formula becomes:
Δ J (i)=(Δ wi)2
According to above formula, when feature j is removed, the approximate variation of optimization problem and the size of corresponding J are consistent, with Jj
Indicate the value of J after feature j is removed, then Jj≈J+wj 2.The principle that support vector machines carries out feature selecting is deleted with smaller
w2The feature of value will bring least variation to optimization problem.The purpose of the algorithm is to find a character subset, can make J
It minimizes, selects fault signature crucial in data.
The training of depth forest model.Former more granularity Scan Architectures are improved to weight more granularities scannings by the present invention, i.e., in original
On the basis of elder generation, processing is weighted for feature vector, each feature vector is multiplied by corresponding weight.And more granularity scannings
Structure includes a random forest and a completely random forest.The data x that a complete intrinsic dimensionality is P is first inputted, and
Multiplied by corresponding weight w, weight is returned by Logistic and is generated.
Logic Regression Models are as follows:
Wherein g (x)=w0+w1x1+…+wpxp。
Likelihood function are as follows:
With maximum likelihood estimate, parameter w is solved0, w1..., wp, this group of parameter is exactly required weight,
Logarithm is taken to obtain likelihood equation:
To parameter wkLocal derviation is sought, following equations are obtained,
Parameter w can be solved by solving equation0, w1..., wpAs weight.
Wherein p indicates that sample dimension, y are its sample observations.
Sliding sampling (step-length L) is carried out by the sampling window that a length is K, obtains S=(P-K)/L+1 K dimension
Feature subsample vector, then each subsample is generated the table of C dimension by completely random forest and random forest training respectively
It levies vector H (x), final each forest obtains the variable that length is S*C, and the output for being superimposed two forests obtains the spy of 2*S*C dimension
Levy vector ZiAs input for training cascade random forest.Here length K, step-length L, selected feature vector digit P, gloomy
Woods exports the dimension C of H (x), the quantity N of decision tree is hyper parameter, in the preceding self-setting of training or uses default value.
Wherein, each forest is made of many a decision trees, here using the regression tree in CART, regression tree choosing
Taking Gain_ σ is the index for evaluating Split Attribute.The attribute and its attribute value with minimum Gain_ σ are chosen, as optimal division
Attribute and optimal Split Attribute value.Gain_ σ value is smaller, illustrate two/after subsample " otherness " it is smaller, illustrate to select
It is better as the effect of Split Attribute (value) to select the attribute (value).For sample set S, the calculation formula of population variance is as follows:
Wherein, μ indicates the mean value of prediction result, ykIndicate the prediction result of k-th of sample.
According to the ith attribute value of attribute A, data set S is divided into two parts S1And S2, then divide after Gain_ σ
It calculates as follows:
Gain_σA,i(S)=σ (S1)+σ(S2)
For attribute A, the Gain_ σ that data set is divided into after two parts by any attribute value is calculated separately, is chosen wherein
Minimum value, optimal two offshoot program obtained as attribute A:
For sample set X, optimal two offshoot program of all properties is calculated, minimum value therein is chosen, as sample set X's
Optimal two offshoot program:
Resulting attribute A and its i-th attribute value, the as optimal Split Attribute of sample set S and optimal classification attribute value are used
To classify to data.
After having executed CRAT regression tree, the classification results of all trees in forest are integrated, each of which forest
Exporting result is
H (x) is final result, hiFor i-th of decision tree classification as a result, Y is true classification, I is metric function, and N is certainly
The quantity of plan tree.
The output result of the forest of each sample is spliced, the feature vector Z of 2*S*C dimension is obtainediIt is gloomy as cascading
The input of woods.
Z1=[H1(x1),H1(x2)…H1(xs),H2(x1),H2(x2)…H2(xs)]T
By feature vector ZiThe first layer of cascade structure is inputted, each layer of cascade structure includes two random forests and two
A completely random forest, each forest calculating process are same as above.
Each forest can export H (x) in cascade structure first layer, in the present invention, cascade structure be improved, by two
The resulting normal probability of a completely random forest and probability of malfunction are averaged, and generate a two-dimensional Enhanced feature vector；It will
The resulting normal probability of two random forests and probability of malfunction are averaged, and generate a two-dimensional Enhanced feature vector, in this way
One four-dimensional Enhanced feature vector of every layer of generation, it is condensed together to be formed with input feature value:
As next layer of input, and so on, until training result is completed.Entirely the number of plies of more granularities cascade forests is
Automatic adjusument, over-fitting, every layer of forest can all carry out cross validation in order to prevent.In the construction phase of cascade forest, only
It when constructing current layer, is not promoted by the verifying accuracy rate of cross validation compared to preceding layer, then the structure of cascade forest
This stopping is brought up, total also just completes, namely training is completed.Depth forest algorithm default uses 3 folding cross validations.It is right
The output result H (x) of each forest of the last layer n-th layer seeks mean value, finally export 4 C dimension data H (x), by this 4
The data of a C dimension are averaged, and obtain the data of C dimension, then taken inside the data of this C dimension it is maximum, as prediction
Value.Output is
The wind power pitch-controlled system intelligent fault forecast method based on improvement depth random forest, using following index
Carry out Comprehensive Evaluation:
(1) mean absolute error MAE calculation formula:
(2) root-mean-square error RMSE calculation formula:
(3) coefficient of determination R2Calculation formula:
(4) coefficient of determination R after correctinga 2Calculation formula:
Wherein, n is sample size, and p is characterized quantity, yiFor actual value,For predicted value,For the average value of actual value.
Embodiment 1:
The related data for having chosen 3 ° of position sensors of variable pitch is verified.By being obtained after data prediction
27000 datas, wherein normal data 18000, fault data 9000.Randomly select 12000 normal datas and 6000
As training set for training fault prediction model, remaining 6000 normal datas and 3000 fault datas are made for fault data
For test failure prediction model.
Wherein, the setting of hyper parameter is shown in Table 1. in depth random forests algorithm
Table 1
Its test result is shown in Table 2, from the coefficient of determination 4 after mean absolute error, root-mean-square error, the coefficient of determination and correction
A index evaluates selected algorithm, and depth forest algorithm is more preferable compared to conventional machines learning algorithm effect, while after improvement
Depth forest algorithm mean absolute error and root-mean-square error be below other three kinds of algorithms, after the coefficient of determination and correction certainly
Determine coefficient and be above other algorithms, it was demonstrated that validity of the improved depth forest algorithm in regression forecasting problem.
Table 2
Data acquisition unit extracts pitch-controlled system operation data from SCADA in real time, inputs trained based on improved
It is analyzed in the fault prediction model that depth random forest is established, the output of fault prediction model and setting early warning value is carried out
Compare, send a signal to monitoring center if being more than early warning value, realize the intellectual monitoring of pitch-controlled system, monitoring velocity is high, quasi-
True rate is high.
Claims (7)
1. a kind of based on the pitch-controlled system failure prediction method for improving depth random forests algorithm, wherein depth forest algorithm packet
Multiple cascade modules are included, each cascade module includes parallel arrangement of completely random forest module and random forest module, and equal
Decision is carried out using decision tree, which is characterized in that above-mentioned failure prediction method includes the following steps:
1) fault prediction model based on the pitch-controlled system for improving depth random forests algorithm is built, the building method of the model is such as
Under:
1-1) sample data is chosen from the history SCADA data that the operation of practical wind field generates；
Feature Selection 1-2) is carried out using the feature selection approach combined based on expertise and data mining algorithm, determines event
Hinder the input variable and output variable of prediction model；
Processing 1-3) is weighted to each sample data；
Sliding sampling 1-4) is carried out to sample data by the sampling window that a length is k, obtains subsample vector；
1-5) each subsample from cascade module completely random forest and random forest training respectively generate one characterize to
It measures H (x), the output for being superimposed two forests obtains feature vector ZiAs input, for training depth random forest；
1-6) in training process to prevent over-fitting, every layer depth random forest can all carry out cross validation, when verifying just
True rate is not promoted compared to preceding layer, then training terminates, and obtains fault prediction model；
2) failure predication is carried out to wind power pitch-controlled system using fault prediction model obtained in step 1-5).
2. it is according to claim 1 based on the pitch-controlled system failure prediction method for improving depth random forests algorithm, it is special
Sign is, the step 1-2) in carry out the method for Feature Selection are as follows: all features are carried out just first with expertise
Sieve forms candidate feature set；Then candidate feature set is carried out into one using support vector machines combination sweep backward strategy
Step screening.
3. it is according to claim 1 based on the pitch-controlled system failure prediction method for improving depth random forests algorithm, it is special
Sign is, the step 1-3) in weighting when weight determination method are as follows: weight wiIt is generated by Logic Regression Models, logic is returned
Return model are as follows:
Wherein g (x)=w0+w1x1+...+wpxp,
The likelihood function of Logic Regression Models are as follows:
With maximum likelihood estimate, parameter w is solved0, w1..., wp, this group of parameter is exactly required weight,
Logarithm is taken to obtain likelihood equation:
To parameter wkLocal derviation is sought, following equations are obtained,
Parameter w can be solved by solving equation0, w1..., wpAs weight, wherein p indicates that sample dimension, y are its sample observations.
4. it is according to claim 3 based on the pitch-controlled system failure prediction method for improving depth random forests algorithm, it is special
Sign is that the decision tree is using the regression tree in cart.
5. it is according to claim 4 based on the pitch-controlled system failure prediction method for improving depth random forests algorithm, it is special
Sign is that the output of the random forest and completely random forest is equal are as follows:
In formula, H (x) is final result, hiFor i-th of decision tree classification as a result, Y is true classification, I is metric function, and N is certainly
The quantity of plan tree.
6. it is according to claim 5 based on the pitch-controlled system failure prediction method for improving depth random forests algorithm, it is special
Sign is, the step 1-6) when being trained to depth random forest, Comprehensive Evaluation is carried out using following index, error is got over
Small, coefficient is higher, shows that the performance of depth random forest is better, could stop instructing when each performance indicator is all satisfied setting condition
Practice, the calculation method of each judgment criteria is as follows:
(1) mean absolute error MAE calculation formula:
(2) root-mean-square error RMSE calculation formula:
(3) coefficient of determination R2Calculation formula:
(4) coefficient of determination R after correctinga 2Calculation formula:
Wherein, n is sample size, and p is characterized quantity, yiFor actual value,For predicted value,For the average value of actual value.
7. it is according to claim 4 based on the pitch-controlled system failure prediction method for improving depth random forests algorithm, it is special
Sign is, when the step 2) carries out failure predication to wind power pitch-controlled system, the output of fault prediction model and setting is pre-
Alert value is compared, and sends a signal to monitoring center if being more than early warning value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910456862.8A CN110298485A (en) | 2019-05-29 | 2019-05-29 | Based on the pitch-controlled system failure prediction method for improving depth random forests algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910456862.8A CN110298485A (en) | 2019-05-29 | 2019-05-29 | Based on the pitch-controlled system failure prediction method for improving depth random forests algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110298485A true CN110298485A (en) | 2019-10-01 |
Family
ID=68027278
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910456862.8A Pending CN110298485A (en) | 2019-05-29 | 2019-05-29 | Based on the pitch-controlled system failure prediction method for improving depth random forests algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110298485A (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111259947A (en) * | 2020-01-13 | 2020-06-09 | 国网浙江省电力有限公司信息通信分公司 | Power system fault early warning method and system based on multi-mode learning |
| CN111461461A (en) * | 2020-04-29 | 2020-07-28 | 中国水利水电科学研究院 | Hydraulic engineering anomaly detection method and system |
| CN111539553A (en) * | 2020-03-31 | 2020-08-14 | 华北电力大学 | Wind turbine generator fault early warning method based on SVR algorithm and skewness |
| CN111651933A (en) * | 2020-05-22 | 2020-09-11 | 宁波诺丁汉新材料研究院有限公司 | Industrial boiler fault early warning method and system based on statistical inference |
| CN111661289A (en) * | 2020-04-23 | 2020-09-15 | 武汉船用机械有限责任公司 | Method and device for identifying faults of controllable pitch propeller |
| CN111950166A (en) * | 2020-08-26 | 2020-11-17 | 广州博依特智能信息科技有限公司 | Cost optimization method for household paper making machine based on data mining |
| CN112347069A (en) * | 2020-08-17 | 2021-02-09 | 广东工业大学 | Manufacturing enterprise prediction type maintenance service method based on large manufacturing data |
| CN112765878A (en) * | 2021-01-11 | 2021-05-07 | 龙源（北京）风电工程技术有限公司 | Intelligent diagnosis method and device for abnormity of generator bearing of wind turbine generator |
| CN112943224A (en) * | 2019-12-11 | 2021-06-11 | 中国石油化工股份有限公司 | Method for calculating dynamic liquid level of heavy oil well |
| CN112966434A (en) * | 2021-02-26 | 2021-06-15 | 西安理工大学 | Random forest sudden fault early warning method based on sliding window |
| CN114077922A (en) * | 2021-10-15 | 2022-02-22 | 华能射阳新能源发电有限公司 | Fault prediction method and device for wind turbine generator and electronic equipment |
| CN114184695A (en) * | 2021-11-09 | 2022-03-15 | 国网内蒙古东部电力有限公司电力科学研究院 | Parameter optimization-based method and system for predicting gas concentration in random forest oil |
| CN114781762A (en) * | 2022-06-21 | 2022-07-22 | 四川观想科技股份有限公司 | Equipment fault prediction method based on life consumption |
| US20220334550A1 (en) * | 2021-04-12 | 2022-10-20 | Chongqing University | Method for identifying critical error of worm gear machine and method for iteratively compensating critical error of worm gear machine |
| CN115268350A (en) * | 2022-09-27 | 2022-11-01 | 江苏永鼎股份有限公司 | Fault early warning method and system for voltage stabilizing transformer |
| CN116882966A (en) * | 2023-06-27 | 2023-10-13 | 广州慧云网络科技有限公司 | Fault judging method and device for inspection result of operation and maintenance equipment |
| CN117436569A (en) * | 2023-09-18 | 2024-01-23 | 华能核能技术研究院有限公司 | Nuclear power equipment fault prediction and intelligent calibration method and system based on random forest |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180060728A1 (en) * | 2016-08-31 | 2018-03-01 | Microsoft Technology Licensing, Llc | Deep Embedding Forest: Forest-based Serving with Deep Embedding Features |
| CN108343599A (en) * | 2018-01-10 | 2018-07-31 | 中国水利水电科学研究院 | A kind of water pump assembly intelligent failure diagnosis method cascading forest based on more granularities |
| CN108682007A (en) * | 2018-04-28 | 2018-10-19 | 华中师范大学 | Jpeg image resampling automatic testing method based on depth random forest |
| CN109102032A (en) * | 2018-09-03 | 2018-12-28 | 中国水利水电科学研究院 | A kind of pumping plant unit diagnostic method based on depth forest and oneself coding |
| CN109297689A (en) * | 2018-09-11 | 2019-02-01 | 中国水利水电科学研究院 | A kind of large-scale hydraulic intelligent diagnosing method introducing weight factor |
-
2019
- 2019-05-29 CN CN201910456862.8A patent/CN110298485A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180060728A1 (en) * | 2016-08-31 | 2018-03-01 | Microsoft Technology Licensing, Llc | Deep Embedding Forest: Forest-based Serving with Deep Embedding Features |
| CN108343599A (en) * | 2018-01-10 | 2018-07-31 | 中国水利水电科学研究院 | A kind of water pump assembly intelligent failure diagnosis method cascading forest based on more granularities |
| CN108682007A (en) * | 2018-04-28 | 2018-10-19 | 华中师范大学 | Jpeg image resampling automatic testing method based on depth random forest |
| CN109102032A (en) * | 2018-09-03 | 2018-12-28 | 中国水利水电科学研究院 | A kind of pumping plant unit diagnostic method based on depth forest and oneself coding |
| CN109297689A (en) * | 2018-09-11 | 2019-02-01 | 中国水利水电科学研究院 | A kind of large-scale hydraulic intelligent diagnosing method introducing weight factor |
Non-Patent Citations (4)
| Title |
|---|
| GUANGZHENG HU ET AL: ""A deep Boltzmann machine and multi-grained scanning forest ensemble"", 《COMPUTERS IN INDUSTRY》 * |
| LEV UTKIN ET AL: ""A Deep Forest Improvement by Using Weighted"", 《PROCEEDINGS OF THE XXTH CONFERENCE OF OPEN INNOVATIONS ASSOCIATION FRUCT》 * |
| XIAOLIAN LIU ET AL: ""Deep forest based intelligent fault diagnosis of hydraulic turbine"", 《JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY》 * |
| 薛参观 等: ""基于改进深度森林算法的软件缺陷预测"", 《计算机科学》 * |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112943224A (en) * | 2019-12-11 | 2021-06-11 | 中国石油化工股份有限公司 | Method for calculating dynamic liquid level of heavy oil well |
| CN112943224B (en) * | 2019-12-11 | 2023-02-07 | 中国石油化工股份有限公司 | Method for calculating dynamic liquid level of heavy oil well |
| CN111259947A (en) * | 2020-01-13 | 2020-06-09 | 国网浙江省电力有限公司信息通信分公司 | Power system fault early warning method and system based on multi-mode learning |
| CN111539553A (en) * | 2020-03-31 | 2020-08-14 | 华北电力大学 | Wind turbine generator fault early warning method based on SVR algorithm and skewness |
| CN111539553B (en) * | 2020-03-31 | 2023-10-24 | 华北电力大学 | Wind turbine generator fault early warning method based on SVR algorithm and off-peak degree |
| CN111661289B (en) * | 2020-04-23 | 2022-04-15 | 武汉船用机械有限责任公司 | Method and device for identifying faults of controllable pitch propeller |
| CN111661289A (en) * | 2020-04-23 | 2020-09-15 | 武汉船用机械有限责任公司 | Method and device for identifying faults of controllable pitch propeller |
| CN111461461A (en) * | 2020-04-29 | 2020-07-28 | 中国水利水电科学研究院 | Hydraulic engineering anomaly detection method and system |
| CN111461461B (en) * | 2020-04-29 | 2022-06-07 | 中国水利水电科学研究院 | Hydraulic engineering abnormity detection method and system |
| CN111651933A (en) * | 2020-05-22 | 2020-09-11 | 宁波诺丁汉新材料研究院有限公司 | Industrial boiler fault early warning method and system based on statistical inference |
| CN111651933B (en) * | 2020-05-22 | 2023-09-26 | 宁波诺丁汉新材料研究院有限公司 | Industrial boiler fault early warning method and system based on statistical inference |
| CN112347069B (en) * | 2020-08-17 | 2024-02-20 | 广东工业大学 | Manufacturing enterprise prediction type maintenance service method based on manufacturing big data |
| CN112347069A (en) * | 2020-08-17 | 2021-02-09 | 广东工业大学 | Manufacturing enterprise prediction type maintenance service method based on large manufacturing data |
| CN111950166A (en) * | 2020-08-26 | 2020-11-17 | 广州博依特智能信息科技有限公司 | Cost optimization method for household paper making machine based on data mining |
| CN111950166B (en) * | 2020-08-26 | 2021-04-02 | 广州博依特智能信息科技有限公司 | Cost optimization method for household paper making machine based on data mining |
| CN112765878A (en) * | 2021-01-11 | 2021-05-07 | 龙源（北京）风电工程技术有限公司 | Intelligent diagnosis method and device for abnormity of generator bearing of wind turbine generator |
| CN112966434A (en) * | 2021-02-26 | 2021-06-15 | 西安理工大学 | Random forest sudden fault early warning method based on sliding window |
| CN112966434B (en) * | 2021-02-26 | 2023-06-23 | 四化信息科技(深圳)有限公司 | Random forest sudden fault early warning method based on sliding window |
| US20220334550A1 (en) * | 2021-04-12 | 2022-10-20 | Chongqing University | Method for identifying critical error of worm gear machine and method for iteratively compensating critical error of worm gear machine |
| CN114077922A (en) * | 2021-10-15 | 2022-02-22 | 华能射阳新能源发电有限公司 | Fault prediction method and device for wind turbine generator and electronic equipment |
| CN114184695A (en) * | 2021-11-09 | 2022-03-15 | 国网内蒙古东部电力有限公司电力科学研究院 | Parameter optimization-based method and system for predicting gas concentration in random forest oil |
| CN114781762B (en) * | 2022-06-21 | 2022-09-23 | 四川观想科技股份有限公司 | Equipment fault prediction method based on life consumption |
| CN114781762A (en) * | 2022-06-21 | 2022-07-22 | 四川观想科技股份有限公司 | Equipment fault prediction method based on life consumption |
| CN115268350A (en) * | 2022-09-27 | 2022-11-01 | 江苏永鼎股份有限公司 | Fault early warning method and system for voltage stabilizing transformer |
| CN116882966A (en) * | 2023-06-27 | 2023-10-13 | 广州慧云网络科技有限公司 | Fault judging method and device for inspection result of operation and maintenance equipment |
| CN116882966B (en) * | 2023-06-27 | 2024-04-19 | 广东慧云科技股份有限公司 | Fault judging method and device for inspection result of operation and maintenance equipment |
| CN117436569A (en) * | 2023-09-18 | 2024-01-23 | 华能核能技术研究院有限公司 | Nuclear power equipment fault prediction and intelligent calibration method and system based on random forest |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110298485A (en) | Based on the pitch-controlled system failure prediction method for improving depth random forests algorithm | |
| CN111237134B (en) | Offshore double-fed wind driven generator fault diagnosis method based on GRA-LSTM-stacking model | |
| CN104807644B (en) | A kind of fault early warning method and system of paddle change system of wind turbines | |
| CN109209790B (en) | Wind power blade icing conjecture method based on data modeling | |
| CN109543210A (en) | A kind of Wind turbines failure prediction system based on machine learning algorithm platform | |
| CN112633317A (en) | CNN-LSTM fan fault prediction method and system based on attention mechanism | |
| CN108363382A (en) | A kind of complex equipment fault diagnosis method and system | |
| Yang et al. | Fault detection of wind turbine generator bearing using attention-based neural networks and voting-based strategy | |
| CN110737976B (en) | Mechanical equipment health assessment method based on multidimensional information fusion | |
| CN111598179B (en) | Power monitoring system user abnormal behavior analysis method, storage medium and equipment | |
| CN114856941A (en) | Offshore wind power plant and unit fault diagnosis operation and maintenance system and diagnosis operation and maintenance method thereof | |
| Afrasiabi et al. | Wind turbine fault diagnosis with generative-temporal convolutional neural network | |
| Yang et al. | Fault early warning of wind turbine gearbox based on multi‐input support vector regression and improved ant lion optimization | |
| Shi et al. | Study of wind turbine fault diagnosis and early warning based on SCADA data | |
| CN114215706A (en) | Wind turbine generator blade cracking fault early warning method and device | |
| CN109325553B (en) | Wind power gear box fault detection method, system, equipment and medium | |
| Chen et al. | Prediction of icing fault of wind turbine blades based on deep learning | |
| CN116624341A (en) | Wind turbine generator power abnormality diagnosis method, system and equipment | |
| CN110705693A (en) | Unmanned aerial vehicle abnormal behavior recognition module and recognition method thereof | |
| CN114742165A (en) | Aero-engine gas circuit performance abnormity detection system based on depth self-encoder | |
| CN113222067A (en) | Intelligent island detection method based on SVM-Adaboost algorithm | |
| Jiang et al. | Intelligent icing detection model of wind turbine blades based on scada data | |
| CN117200060A (en) | Intelligent inspection method and system based on new energy power plant | |
| Toubakh et al. | Advanced pattern recognition approach for fault diagnosis of wind turbines | |
| CN114753980B (en) | Method and system for monitoring icing of fan blade |
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: 20191001 |
|
| RJ01 | Rejection of invention patent application after publication |