CN109522627A - Fan blade icing prediction technique based on SCADA data - Google Patents
Fan blade icing prediction technique based on SCADA data Download PDFInfo
- Publication number
- CN109522627A CN109522627A CN201811296712.7A CN201811296712A CN109522627A CN 109522627 A CN109522627 A CN 109522627A CN 201811296712 A CN201811296712 A CN 201811296712A CN 109522627 A CN109522627 A CN 109522627A
- Authority
- CN
- China
- Prior art keywords
- fan blade
- icing
- data
- prediction model
- icing prediction
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- 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
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/70—Wind energy
- Y02E10/72—Wind turbines with rotation axis in wind direction
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Structures Of Non-Positive Displacement Pumps (AREA)
Abstract
The machine blade icing prediction technique based on SCADA data that the invention proposes a kind of, it is lower for solving predictablity rate existing in the prior art, and cannot achieve when fan blade icing conditions are unobvious the limitation technical problem of prediction.Realize step are as follows: obtain normalization SCADA data;Obtaining influences the characteristic data set that fan blade freezes;It constructs and trains fan blade icing prediction model;Blower icing prediction model after training is optimized;Judge whether fan blade freezes.The present invention fully considers the influence that the data of all acquisitions freeze to fan blade, under the premise of promoting fan blade icing condition prediction accuracy, realizes the real-time predictive diagnosis to freeze to fan blade.
Description
Technical field
The invention belongs to wind motor icing detection fields, are related to a kind of prediction technique that fan blade freezes, specifically
It is related to a kind of fan blade icing prediction technique based on SCADA data.
Background technique
Blower obtains wind energy by blade, and converts wind energy into mechanical energy, so that driven generator generates electricity, Jin Ershi
Conversion of the existing wind energy to electric energy.Therefore, blade is the core component of blower, and the efficiency for obtaining wind energy is to guarantee the power generation of blower height
The key factor of amount.
SCADA (Supervisory Control And Data Acquisition) system in blower, i.e. data are adopted
Collection and supervisor control are computer-based DCS and power automation monitoring system, can set to the operation at scene
It is standby to be monitored and controlled, to realize the functions such as data acquisition, equipment control, measurement, parameter regulation.
In different geographical, for example lower temperature of the varying environment that blower is faced, higher humidity, biggish dust storm
Deng can all have some impact on to the normal work of blower unit, and it may cause fan blade in certain low temperature and high relative humidity environment
It freezes, due to the uneven distribution of ice cube on blade, so that the distribution loaded of blade institute changes, to influence opening for blade
Efficiency of movement reduces the generated energy of blower unit, can also change the resonant frequency of fan blade, and then change its dynamic response row
To will lead to leaf destruction when serious and bring damage to blower, so blade ice formation issues caused by low temperature and high relative humidity environment
It is that wind power plant safeguards the main problem faced.It is not only blower unit generation caused by being frozen due to blower unit
Measure the production efficiency problem of decline, it is also possible in the case where the decline of ice sheet adhesion, ice cube occur and fall off, cause operation thing
Therefore.In addition, the generation of serious machine halt trouble freezes often caused by blade as time integral constantly deteriorates, if can freeze
Early-time analysis identifies the presence of borneol, can greatly reduce the generation of catastrophe failure.So to fan blade icing prediction technique
It is very necessary for carrying out further investigation.
From the point of view of presently disclosed data, the method for fan blade icing detection is main are as follows: by installing additional, blower is airborne to be set
Standby includes that sensor, collector, processor and memory etc. detect whether icing to fan blade.Technical way is past
Identification inspection is carried out to blade icing condition toward the generated noise different from normal condition is frozen by analysis fan blade
It surveys, can not accurately identify noise caused by other situations such as blade loading change, sandstone weather and icing condition
Noise difference, so that the accurate detection to fan blade icing condition cannot be reached.At the same time, pass through the means such as identification noise
It can not be identified on the first appearance in borneol, can only could often find, can not accomplish when icing has already appeared and ice cube is larger
It finds and handles in time.
Research in terms of excavating at present for fan blade icing prediction data is still more deficient, research master at this stage
Concentrate on how research frost influences anemometer tower and anemobiagraph, in addition, how to be quantified as specifically sending out by this ice condition
Electric quantity loss, the quantitative relationship established between icing intensity and generated energy loss is also the main goal in research of other scholars, right
In the focus of blower unit ice formation issues be not Accurate Prediction blower turbines vane ice-formation condition and icing degree, to be
Manual or automatic deicing is laid a good foundation.For example, the paper that the Makkonen L of Technical Research Centre of Finland is delivered at it
“Models for the growth of rime,glaze,icicles and wet snow on structures[J]”
It is developed in (Philosophical transactions of the royal society, 2012 (358): 2913-2939)
A kind of simulation blade is under the conditions of wind tunnel experiment, by the Turbice model for being dried to the propagation process of ice under wet environment.It should
Heating amount needed for modeling ice detachment, at the same provide droplet trajectory, liquid collection efficiency, energy and mass balance,
Information such as the shape of accumulated ice and thickness, but the modeling condition is permanent external environment, two dimensional model mainly for
The frost of overhead power line emulates, and threedimensional model is then only applied to the emulation of aircraft blade accumulated ice.This kind of method is not considered
Meteorological condition (such as wind speed) or mechanical factor (as vibrated or being bent the influence to wind power generation unit blade), so that accumulated ice process be joined
Numberization and prolonged freezing process can be simulated, also cannot accurately predict fan blade icing degree.
Summary of the invention
It is an object of the invention to overcome the problems of the above-mentioned prior art, provide a kind of based on SCADA data
Fan blade icing prediction technique, it is intended to realize to fan blade icing is predicted and is diagnosed, thus make stress type maintenance
Mode becomes the maintenance mode of active predicting type, lower for solving predictablity rate existing in the prior art, and in blower
Blade icing conditions cannot achieve the limitation technical problem of prediction when unobvious.
To achieve the above object, the technical solution that the present invention takes includes the following steps:
(1) normalization SCADA data is obtained:
To the SCADA number including fan condition supplemental characteristic and operational parameter data of the passing acquisition of blower SCADA system
According to being normalized, normalization SCADA data is obtained;
(2) obtaining influences the characteristic data set that fan blade freezes:
The threshold value for extracting key parameter data weighting is set as δ, and the weight that will be filtered out from normalization SCADA data
Duty parameter data and operational parameter data greater than δ, as the key parameter data characteristics for influencing fan blade icing, composition
Characteristic data set;
(3) it constructs and trains fan blade icing prediction model:
Fan blade icing prediction model A is constructed, and A is trained by characteristic data set, the wind after being trained
Machine blade icing prediction model B;
(4) the fan blade icing prediction model B after training is optimized:
The negative gradient direction of the weight θ of the fan blade icing prediction model B of (4a) after training is iterated more θ
Newly, loss function corresponding to each weight after updating is calculated separately, and using the corresponding weight of least disadvantage function as most
Excellent weight, the fan blade icing prediction model with best initial weights are the fan blade icing prediction model C of right-value optimization;
(4b) is iterated update to the threshold value beta of the blower icing prediction model C of right-value optimization, calculates separately every after updating
Icing prediction accuracy corresponding to one threshold value, and using the corresponding threshold value of highest icing prediction accuracy as optimal threshold,
Fan blade icing prediction model with best initial weights and optimal threshold is the fan blade icing prediction model after optimizing
D;
(5) judge whether fan blade freezes:
Blower SCADA system acquires the corresponding blower SCADA data of key parameter data characteristics in real time, and is entered into
In fan blade icing prediction model D after optimization, then compare the output valve and threshold size of D, when D output valve be greater than etc.
When the threshold value of D, fan blade is determined to freeze, when the output valve of D is less than the threshold value of D, determines that fan blade is not freeze.
Compared with the prior art, the invention has the following advantages:
1. the present invention uses data mining and modeling method, by passing collected including wind from blower SCADA system
The SCADA data of machine duty parameter data and operational parameter data fully considers all data being collected into fan blade knot
Influence caused by ice problem establishes the fan blade knot that depth optimization was carried out by big data training, weight and threshold value
Ice prediction model, later can be by acquiring the corresponding blower SCADA number of key parameter data characteristics for blower SCADA system in real time
According to input model, detection judgement, accuracy with higher are carried out to fan blade icing degree.
2. data collection system of the invention can sufficiently acquire and store the passing and real time data about fan blade,
Model prediction accuracy, removal noise data and reduction operand are improved by extracting key parameter feature, by big data mode
Real-time is combined with failure predication and diagnosis, establishes a kind of maintenance mode of active predicting, is meeting icing condition inspection
Under the premise of surveying accuracy, cost needed for detection required time, deicing is greatly reduced and the Failure risk that may cause that freezes
Degree.Since the present invention explicitly represents fan blade icing degree in the form of a model output value i.e. quantizating index
Come, so can not only be determined as that blade freezes, and works as wind when fan blade icing model output value is more than or equal to icing threshold value
It makes prediction when the ascending close icing threshold value of machine blade icing model output value and is likely to occur blade icing;And it can work as
Model output value is determined as fan blade when being equal to icing threshold value, and there are mixture of ice and water, are higher by icing threshold value with model output value
Size degree characterization fan blade icing severity, so that it is unknown in fan blade icing conditions to solve the prior art
It cannot achieve the limitation technical problem of prediction when aobvious.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention.
Specific embodiment
In the following with reference to the drawings and specific embodiments, present invention is further described in detail.
A kind of fan blade icing prediction technique based on SCADA data of referring to Fig.1, includes the following steps:
Step 1) obtains normalization SCADA data:
To the SCADA number including fan condition supplemental characteristic and operational parameter data of the passing acquisition of blower SCADA system
According to being normalized, normalization SCADA data is obtained;
Acquire western certain wind field separate unit blower in November, 2015 and two months December totally four ten altogether by blower SCADA system
Ten thousand datas and moment icing condition information is corresponded to as obtained by artificial observation, including 26 different parameters, it is specific as follows shown in:
Duty parameter data include: wind speed, generator speed, net side active power, to wind angle, 25 seconds mean wind direction angles,
Environment temperature and cabin temperature;
Operational parameter data includes: cabin yaw position, cabin yawing velocity, the 1st blade angle and speed, the 2nd blade
Angle and speed, the 3rd blade angle and speed, the 1st pitch motor temperature, the 2nd pitch motor temperature, the 3rd pitch motor temperature,
Planar horizontal X-direction acceleration locating for three blades and vertical Y directional acceleration, 1Ng5 charger temperature and DC current,
2Ng5 charger temperature and DC current, 3Ng5 charger temperature and DC current.
Step 2), which obtains, influences the characteristic data set that fan blade freezes:
The threshold value for extracting key parameter data weighting is set as δ, and the weight that will be filtered out from normalization SCADA data
Duty parameter data and operational parameter data greater than δ, as the key parameter data characteristics for influencing fan blade icing, composition
Characteristic data set;
In order to reduce data dimension, influence of the extraneous data to blade icing model is removed, while reducing calculation amount raising
Calculating speed, spy realize that key parameter data characteristics is extracted using Relief algorithm, realize step are as follows:
Step 2a) define the passing collected duty parameter data of SCADA system and operational parameter data vector is A, mistake
The blower history data set formed toward collected duty parameter data and operational parameter data is E, and sample frequency in sampling is m,
The threshold value for extracting feature weight is δ, and feature weight vector W (A) is 0;
Step 2b) a sample R is randomly selected from history data set E, and sample is determined according to blower historical failure information
Whether this R is in icing condition;
Step 2c) Euclidean distance is in the sample of same icing condition most with sample R from extracting in history data set E
Close sample is used as and guesses neighbour H, while the sample that different icing conditions are in from sample R is extracted from history data set E
The farthest sample of middle Euclidean distance is as the neighbour M that guesses wrong;
Step 2d) according to neighbour H is guessed and the neighbour M that guesses wrong, calculating feature weight vector is W (A):
W (A)=W (A)-diff (R, H)+diff (R, M);
Step 2e) judge whether sample drawn number reaches sample frequency in sampling m, if so, from feature weight vector W (A)
Otherwise middle choose repeats higher than the feature vector for extracting feature weight threshold value δ as key parameter data characteristics T is extracted
Step (2b)~(2d).
Step 3) constructs and training fan blade icing prediction model:
Fan blade icing prediction model A is constructed, and A is trained by characteristic data set, the wind after being trained
Machine blade icing prediction model B;
Being detected to fan blade icing, which is after all one, is divided into icing for fan condition and does not freeze two kinds not
With the classification problem of state, in order to preferably distinguish icing condition, therefore chooses logistic regression model and complete classification times
Business in order to solve the problems, such as much fewer than the normal data amount not frozen of data volume to freeze, while being also more concerned about icing condition
Influence, need to assign the data of icing condition bigger weight, to logistic regression model using particle swarm algorithm optimize
It can be well solved the above problem, realize step are as follows:
Step 3a) number of particles is set as pN, search space dimension is dim, aceleration pulse c1And c2, the number of iterations is
IMax, maximum circle in the air speed be vMax;
Step 3b) initialization inertial factor be w, initialize each particle position be vi, initialize the speed of each particle
Degree is pi;
Step 3c) fitness value of current each particle is calculated according to fitness function, and current each particle is set
Fitness is individual extreme value Jbest;
Step (3d) is by the individual extreme value J of each particlebestIn maximum individual extreme value as global extremum gbesti;
Step 3e) according to inertial factorSpeed and position to each particle carry out more
Newly, and according to fitness function the fitness value of each particle after renewal speed and position is calculated;
Step 3f) more each particle individual extreme value JbestWith the size of the fitness value after renewal speed and position,
If fitness value is greater than individual extreme value Jbest, then individual extreme value J is replaced with fitness valuebest, otherwise, individual extreme value JbestNo
Become;
Step 3g) more each particle global extremumWith the size of speed after update and the fitness value behind position,
If fitness value is greater than in global extremumThen global extremum is replaced with fitness valueOtherwise, global extremum
It is constant;
Step 3h) judge whether to reach greatest iteration update times IMax, if so, output global extremumAs
Otherwise the weighting parameter of Logistic regression model repeats step (3c)~(3g).
Step 4) optimizes the fan blade icing prediction model B after training:
Step 4a) the negative gradient direction of weight θ of fan blade icing prediction model B after training is iterated θ
Update, calculate separately loss function corresponding to each weight after update, and using the corresponding weight of least disadvantage function as
Best initial weights, the fan blade icing prediction model with best initial weights are the fan blade icing prediction model of right-value optimization
C;
In order to find out the hyperplane that icing can be perfectly divided into both sides as far as possible in data space with the data that do not freeze,
The weight of fan blade icing model is optimized, realizes step are as follows:
Step 4a1) set gradient descent algorithm optimization fan blade icing prediction model initial weight as
The weight of Logistic regression model, and set maximum number of iterations;
Step 4a2) along initial weight by negative gradient direction according to iterative formula update weight θ, iterative formula are as follows:
θ :=θ+ρ datamatrixT(Y-P)
Wherein, datamatrix is the corresponding matrix of data set;Y=(y1,y2,…,yn)TIt is the sample of history data set
Tag along sort vector;P=(p1,p2,…,pn)T, whereinxijFor history data set
I-th row jth column element of homography;θ=(α, β1,β2,…,βk) be Logistic regression model weight vector, ρ be step
It is long;
Step 4a3) output best initial weights.
Step 4b) update is iterated to the threshold value beta of the fan blade icing prediction model C of right-value optimization, it calculates separately
Icing prediction accuracy corresponding to each threshold value after update, and using the corresponding threshold value of highest icing prediction accuracy as most
Excellent threshold value, the blower icing prediction model with best initial weights and optimal threshold are the fan blade icing prediction mould after optimizing
Type D;
Threshold value locating for blade icing condition accurately is defined, so that mixture of ice and water state is just equal to threshold value, to threshold value
The realization step of optimization are as follows:
Step 4b1) threshold value beta and greatest iteration step-length i are set;
Step 4b2) according to formula β :=β+0.05i calculates the threshold value beta under current iteration step-length, and using the threshold value as leaf
The icing threshold value of piece icing prediction model.
Step 5) judges whether fan blade freezes:
Blower SCADA system acquires the corresponding blower SCADA data of key parameter data characteristics in real time, and is entered into
In fan blade icing prediction model D after optimization, then compare the output valve and threshold size of D, when D output valve be greater than etc.
When the threshold value of D, fan blade is determined to freeze, when the output valve of D is less than the threshold value of D, determines that fan blade is not freeze.
Fan blade icing degree is explicitly showed in the form of a model output value i.e. quantizating index, so
It can not only be determined as that blade freezes when fan blade icing model output value is more than or equal to icing threshold value, when fan blade knot
It makes prediction when the ascending close icing threshold value of ice model output valve and is likely to occur blade icing;And it can be exported when model
Value is determined as fan blade when being equal to icing threshold value, and there are mixture of ice and water, and the size journey of icing threshold value is higher by with model output value
The icing severity of degree characterization fan blade.
Below in conjunction with emulation experiment, technical effect of the invention is described further.
1, experiment condition
Data simulation experiment of the invention is the hardware environment in Intel (R) Core (TN) [email protected]
Under, it is carried out under the software environment of Matlab2014b.
2. experiment content and interpretation of result
It acquires current working supplemental characteristic, the operational parameter data of wind-driven generator in real time by SCADA system, shares such as
Lower 26 kinds of parameters:
Duty parameter data include: wind speed, generator speed, net side active power, to wind angle, 25 seconds mean wind direction angles,
Environment temperature and cabin temperature;
Operational parameter data includes: cabin yaw position, cabin yawing velocity, the 1st blade angle and speed, the 2nd blade
Angle and speed, the 3rd blade angle and speed, the 1st pitch motor temperature, the 2nd pitch motor temperature, the 3rd pitch motor temperature,
Planar horizontal X-direction acceleration locating for three blades and vertical Y directional acceleration, 1Ng5 charger temperature and DC current,
2Ng5 charger temperature and DC current, 3Ng5 charger temperature and DC current.
Blower SCADA data is normalized and pre-processed, normalization SCADA data is obtained.
According to the historical data of blower and the record that freezes, using Relief algorithm to collected duty parameter data and fortune
Row supplemental characteristic is screened, and sets sampling number as 1000 times, weight threshold 3000 obtains key parameter data characteristics such as
Under:
Wind speed, generator speed, net survey active power, to wind angle, 25 seconds mean wind direction angles, cabin yaw position, the 3rd leaf
Piece angle, the 2nd pitch motor temperature, planar horizontal X-direction acceleration and vertical Y directional acceleration, environment locating for three blades
Temperature.
According to the historical data of blower and the record that freezes, using the key parameter data characteristics of extraction, by being based on particle
The Logistic regression model of group's algorithm establishes blade icing prediction model, sets number of particles pNIt is 40, the dimension of search space
Spend dim=20, aceleration pulse c1=c2=2, the number of iterations IMax=20, maximum is circled in the air speed vMax=0.9, constraint factor α setting
It is 1, particle position and speed are initialized to the random number between [0,1].
Using gradient descent algorithm, using the weight coefficient in the Logistic regression model based on particle swarm algorithm as just
Beginning weight optimizes the weight of blade icing prediction model, and setting maximum number of iterations is set as 1000, step-length ρ setting
It is 0.001, obtains the best initial weights of blade icing prediction model.
Using grid-search algorithms, on the basis of having obtained best initial weights, to the former threshold value of blade icing prediction model
0.5 optimizes, and setting greatest iteration step-length is 20, obtains the optimal threshold 0.5918 of blade icing prediction model.
The SCADA data acquired in real time is inputted in blade icing prediction model, and utilizes the key parameter feature extracted,
The diagnosis and prediction for carrying out blade icing, obtain the real-time icing condition of current vane, and prediction and diagnosis effect are commented with accuracy
Sentence, accuracy can reach 93.2, and existing technical method accuracy is generally 85 or so.
Accuracy formula is as follows:
Wherein, α, β are different icing condition weights, NnormalFor normal data quantity, NfaultFor icing data bulk, FN
With being explained as follows shown in table for FP:
Claims (6)
1. a kind of fan blade icing prediction technique based on SCADA data, which comprises the steps of:
(1) normalization SCADA data is obtained:
To the passing acquisition of blower SCADA system include fan condition supplemental characteristic and operational parameter data SCADA data into
Row normalization obtains normalization SCADA data;
(2) obtaining influences the characteristic data set that fan blade freezes:
The threshold value for extracting key parameter data weighting is set as δ, and the weight filtered out from normalization SCADA data is greater than
The duty parameter data and operational parameter data of δ, as the key parameter data characteristics for influencing fan blade icing, composition characteristic
Data set;
(3) it constructs and trains fan blade icing prediction model:
Fan blade icing prediction model A is constructed, and A is trained by characteristic data set, the blower leaf after being trained
Piece icing prediction model B;
(4) the fan blade icing prediction model B after training is optimized:
The negative gradient direction of the weight θ of the blower icing prediction model B of (4a) after training is iterated update to θ, counts respectively
Loss function corresponding to each weight after updating is calculated, and using the corresponding weight of least disadvantage function as best initial weights, tool
The fan blade icing prediction model for having best initial weights is the fan blade icing prediction model C of right-value optimization;
(4b) is iterated update to the threshold value beta of the fan blade icing prediction model C of right-value optimization, calculates separately every after updating
Icing prediction accuracy corresponding to one threshold value, and using the corresponding threshold value of highest icing prediction accuracy as optimal threshold,
Fan blade icing prediction model with best initial weights and optimal threshold is the fan blade icing prediction model after optimizing
D;
(5) judge whether fan blade freezes:
Blower SCADA system acquires the corresponding blower SCADA data of key parameter data characteristics in real time, and is entered into optimization
In fan blade icing prediction model D afterwards, then compare the output valve and threshold size of D, when the output valve of D is more than or equal to D
Threshold value when, determine fan blade be freeze, when the output valve of D be less than D threshold value when, determine fan blade be do not freeze.
2. the fan blade icing prediction technique according to claim 1 based on SCADA data, which is characterized in that step
(1) duty parameter data and operational parameter data described in, in which:
Duty parameter data include: wind speed, generator speed, net side active power, to wind angle, 25 seconds mean wind direction angles, environment
Temperature and cabin temperature;
Operational parameter data includes: cabin yaw position, cabin yawing velocity, the 1st blade angle and speed, the 2nd blade angle
With speed, the 3rd blade angle and speed, the 1st pitch motor temperature, the 2nd pitch motor temperature, the 3rd pitch motor temperature, three
Planar horizontal X-direction acceleration and vertical Y directional acceleration locating for blade, 1Ng5 charger temperature and DC current,
2Ng5 charger temperature and DC current, 3Ng5 charger temperature and DC current.
3. the fan blade icing prediction technique according to claim 1 based on SCADA data, which is characterized in that step
(2) the extraction key parameter data characteristics described in realizes step using Relief algorithm are as follows:
(2a) defines the passing collected duty parameter data of SCADA system and operational parameter data vector is A, passing to collect
Duty parameter data and operational parameter data composition blower history data set be E, sample frequency in sampling be m, extract feature
The threshold value of weight is δ, and feature weight vector W (A) is 0;
(2b) randomly selects a sample R from history data set E, and whether determines sample R according to blower historical failure information
In icing condition;
(2c) is in the sample that Euclidean distance is nearest in the sample of same icing condition with sample R from extracting in history data set E
Neighbour H is guessed in this conduct, while being in Euclidean in the sample of different icing conditions from sample R from extracting in history data set E
Apart from farthest sample as the neighbour M that guesses wrong;
(2d) according to neighbour H is guessed and the neighbour M that guesses wrong, calculating feature weight vector is W (A):
W (A)=W (A)-diff (R, H)+diff (R, M);
(2e) judges whether sample drawn number reaches sample frequency in sampling m, if so, choosing from feature weight vector W (A) high
Otherwise step (2b) is repeated as key parameter data characteristics T is extracted in the feature vector for extracting feature weight threshold value δ
~(2d).
4. the fan blade icing prediction technique according to claim 1 based on SCADA data, which is characterized in that step
(3) the building fan blade icing prediction model A described in, using the logistic regression model optimized through particle swarm algorithm,
Realize step are as follows:
(3a) sets number of particles as pN, search space dimension is dim, aceleration pulse c1And c2, the number of iterations IMax, maximum
Speed of circling in the air is vMax;
It is w that (3b), which initializes inertial factor, and the position for initializing each particle is vi, the speed for initializing each particle is pi;
(3c) calculates the fitness value of current each particle according to fitness function, and the fitness value of current each particle is made
For individual extreme value Jbest;
(3d) is by the individual extreme value J of each particlebestIn maximum individual extreme value as global extremum
(3e) is according to inertial factorThe speed and position of each particle are updated, and root
The fitness value of each particle after calculating renewal speed and position according to fitness function;
The individual extreme value J of (3f) more each particlebestWith the size of the fitness value after renewal speed and position, if adapt to
Angle value is greater than individual extreme value Jbest, then individual extreme value J is replaced with fitness valuebest, otherwise, individual extreme value JbestIt is constant;
The global extremum of (3g) more each particleWith the size of speed after update and the fitness value behind position, if suitable
Angle value is answered to be greater than in global extremumThen global extremum is replaced with fitness valueOtherwise, global extremumIt is constant;
(3h) judges whether to reach greatest iteration update times IMax, if so, output global extremumIt is returned as Logistic
Return the weighting parameter of model, otherwise, repeats step (3c)~(3g).
5. the fan blade icing prediction technique according to claim 1 based on SCADA data, which is characterized in that step
Update is iterated to θ along the negative gradient direction of the weight θ of fan blade icing prediction model B described in (4a), realizes step
Suddenly are as follows:
The initial weight that (4a1) sets the fan blade icing prediction model of gradient descent algorithm optimization returns mould as Logistic
The weight of type, and set maximum number of iterations;
(4a2) is updated weight θ by negative gradient direction along initial weight, obtains best initial weights and exports, iterative formula are as follows:
θ :=θ+ρ datamatrixT(Y-P)
Wherein, datamatrix is the corresponding matrix of data set;Y=(y1,y2,…,yn)TIt is the sample classification mark of history data set
Sign vector;P=(p1,p2,…,pn)T, whereinxijSquare is corresponded to for history data set
I-th row jth column element of battle array;θ=(α, β1,β2,…,βk) be Logistic regression model weight vector, ρ is step-length.
6. the fan blade icing prediction technique according to claim 1 based on SCADA data, which is characterized in that step
Update is iterated to the threshold value beta of the blower icing prediction model C of right-value optimization described in (4b), realizes step are as follows:
Threshold value beta and greatest iteration step-length i is arranged in (4b1);
(4b2) is according to formula β :=β+0.05i calculates the threshold value beta under current iteration step-length, and β is frozen as blade and predicts mould
The icing threshold value of type.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811296712.7A CN109522627B (en) | 2018-11-01 | 2018-11-01 | Fan blade icing prediction method based on SCADA (Supervisory control and data acquisition) data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811296712.7A CN109522627B (en) | 2018-11-01 | 2018-11-01 | Fan blade icing prediction method based on SCADA (Supervisory control and data acquisition) data |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109522627A true CN109522627A (en) | 2019-03-26 |
CN109522627B CN109522627B (en) | 2022-12-02 |
Family
ID=65774362
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811296712.7A Active CN109522627B (en) | 2018-11-01 | 2018-11-01 | Fan blade icing prediction method based on SCADA (Supervisory control and data acquisition) data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109522627B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110147811A (en) * | 2019-04-02 | 2019-08-20 | 宜通世纪物联网研究院(广州)有限公司 | Fan blade prediction method and system based on time window hybrid model |
CN110162888A (en) * | 2019-05-24 | 2019-08-23 | 北京天泽智云科技有限公司 | A method of the fan blade based on semi-supervised learning, which freezes, to be detected |
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 |
CN111968084A (en) * | 2020-08-08 | 2020-11-20 | 西北工业大学 | Method for quickly and accurately identifying defects of aero-engine blade based on artificial intelligence |
CN112270814A (en) * | 2020-12-21 | 2021-01-26 | 长沙树根互联技术有限公司 | Dynamic alarm method, device, electronic equipment and readable storage medium |
CN112836424A (en) * | 2021-01-08 | 2021-05-25 | 上海电机学院 | Early icing fault prediction method for fan blade |
CN113847216A (en) * | 2021-10-14 | 2021-12-28 | 远景智能国际私人投资有限公司 | Method, device and equipment for predicting state of fan blade and storage medium |
CN114166460A (en) * | 2022-02-11 | 2022-03-11 | 中国空气动力研究与发展中心低速空气动力研究所 | Aircraft air inlet passage test device and system and hot gas anti-icing test stability judgment method |
CN114180023A (en) * | 2021-12-08 | 2022-03-15 | 中国船舶重工集团公司第七一六研究所 | Multi-energy ship control management method and device based on load prediction algorithm |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105139274A (en) * | 2015-08-16 | 2015-12-09 | 东北石油大学 | Power transmission line icing prediction method based on quantum particle swarm and wavelet nerve network |
CN106991047A (en) * | 2017-03-27 | 2017-07-28 | 中国电力科学研究院 | A kind of method and system for being predicted to object-oriented software defect |
CN107516150A (en) * | 2017-08-25 | 2017-12-26 | 广东工业大学 | A kind of Forecasting Methodology of short-term wind-electricity power, apparatus and system |
-
2018
- 2018-11-01 CN CN201811296712.7A patent/CN109522627B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105139274A (en) * | 2015-08-16 | 2015-12-09 | 东北石油大学 | Power transmission line icing prediction method based on quantum particle swarm and wavelet nerve network |
CN106991047A (en) * | 2017-03-27 | 2017-07-28 | 中国电力科学研究院 | A kind of method and system for being predicted to object-oriented software defect |
CN107516150A (en) * | 2017-08-25 | 2017-12-26 | 广东工业大学 | A kind of Forecasting Methodology of short-term wind-electricity power, apparatus and system |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110147811A (en) * | 2019-04-02 | 2019-08-20 | 宜通世纪物联网研究院(广州)有限公司 | Fan blade prediction method and system based on time window hybrid model |
CN110162888B (en) * | 2019-05-24 | 2022-12-09 | 北京天泽智云科技有限公司 | Fan blade icing detection method based on semi-supervised learning |
CN110162888A (en) * | 2019-05-24 | 2019-08-23 | 北京天泽智云科技有限公司 | A method of the fan blade based on semi-supervised learning, which freezes, to be detected |
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 |
CN111968084A (en) * | 2020-08-08 | 2020-11-20 | 西北工业大学 | Method for quickly and accurately identifying defects of aero-engine blade based on artificial intelligence |
CN112270814A (en) * | 2020-12-21 | 2021-01-26 | 长沙树根互联技术有限公司 | Dynamic alarm method, device, electronic equipment and readable storage medium |
CN112836424A (en) * | 2021-01-08 | 2021-05-25 | 上海电机学院 | Early icing fault prediction method for fan blade |
CN113847216A (en) * | 2021-10-14 | 2021-12-28 | 远景智能国际私人投资有限公司 | Method, device and equipment for predicting state of fan blade and storage medium |
CN113847216B (en) * | 2021-10-14 | 2023-09-26 | 远景智能国际私人投资有限公司 | Fan blade state prediction method, device, equipment and storage medium |
CN114180023A (en) * | 2021-12-08 | 2022-03-15 | 中国船舶重工集团公司第七一六研究所 | Multi-energy ship control management method and device based on load prediction algorithm |
CN114180023B (en) * | 2021-12-08 | 2023-09-01 | 中国船舶集团有限公司第七一六研究所 | Multi-energy ship control management method and device based on load prediction algorithm |
CN114166460B (en) * | 2022-02-11 | 2022-04-19 | 中国空气动力研究与发展中心低速空气动力研究所 | Aircraft air inlet passage test device and system and hot gas anti-icing test stability judgment method |
CN114166460A (en) * | 2022-02-11 | 2022-03-11 | 中国空气动力研究与发展中心低速空气动力研究所 | Aircraft air inlet passage test device and system and hot gas anti-icing test stability judgment method |
Also Published As
Publication number | Publication date |
---|---|
CN109522627B (en) | 2022-12-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109522627A (en) | Fan blade icing prediction technique based on SCADA data | |
CN109958588B (en) | Icing prediction method, icing prediction device, storage medium, model generation method and model generation device | |
KR102076426B1 (en) | System for managing detailed weather prediction information in real time and method to use for impact forecasting of heat-wave and tropical night using the system | |
CN112818590A (en) | Method for evaluating power generation capacity of wind power plant and establishing micro-site selection model | |
CN105508152B (en) | Blade icing model construction method and icing state monitoring method and device | |
US20160215759A1 (en) | Methods and systems for wind plant power optimization | |
EP2128441A2 (en) | Optimizing turbine layout in wind turbine farm | |
US20060173623A1 (en) | System and method for enhanced measure-correlate-predict for a wind farm location | |
CN107194097A (en) | Analysis method based on wind power plant pneumatic analog and wind speed and direction data | |
CN109751206A (en) | fan blade icing failure prediction method, device and storage medium | |
CN106951997A (en) | The method and apparatus for predicting the failure of blower fan | |
CN108107911A (en) | A kind of autonomous optimizing path planning method of solar powered aircraft | |
US20190277257A1 (en) | Improving annual energy production of wind turbine sites | |
KR20190100238A (en) | Wind Farm Complex Load Management Control Strategy | |
CN109376801A (en) | Blade of wind-driven generator icing diagnostic method based on integrated deep neural network | |
CN103026266A (en) | Estimation of wind conditions at a wind turbine | |
CN111709184B (en) | Accumulated ice combined decision-making method based on pavement temperature prediction and simulation | |
CN115859837B (en) | Digital twin modeling-based fan blade dynamic impact detection method and system | |
CN113994087B (en) | Method and system for controlling the amount of a wind turbine by selecting a controller via machine learning | |
CN103278326A (en) | Method for diagnosing faults of wind generating set gear case | |
CN111878308A (en) | Wind turbine cluster prediction group control method | |
CN113886928A (en) | Architectural mechanics analog system | |
CN109494716A (en) | Wind power output power confidence interval prediction technique based on Bootstrap | |
Van Pelt et al. | Design, construction, and calibration of a portable boundary layer wind tunnel for field use | |
CN105138729A (en) | Filling method based on PSO-GRNN (Particle Swarm Optimization-Generalized Regression Neural Network) for defect wind speed values of wind turbines in wind power plant |
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 |