CN114060213A - Quick unit startup control method and system based on icing level estimation - Google Patents

Quick unit startup control method and system based on icing level estimation Download PDF

Info

Publication number
CN114060213A
CN114060213A CN202111344768.7A CN202111344768A CN114060213A CN 114060213 A CN114060213 A CN 114060213A CN 202111344768 A CN202111344768 A CN 202111344768A CN 114060213 A CN114060213 A CN 114060213A
Authority
CN
China
Prior art keywords
unit
state
icing
blade
ice
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
Application number
CN202111344768.7A
Other languages
Chinese (zh)
Other versions
CN114060213B (en
Inventor
夏祥孟
董帅
黄国燕
王镔
张贝
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
MingYang Smart Energy Group Co Ltd
Original Assignee
MingYang Smart Energy Group Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by MingYang Smart Energy Group Co Ltd filed Critical MingYang Smart Energy Group Co Ltd
Priority to CN202111344768.7A priority Critical patent/CN114060213B/en
Publication of CN114060213A publication Critical patent/CN114060213A/en
Application granted granted Critical
Publication of CN114060213B publication Critical patent/CN114060213B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F03—MACHINES OR ENGINES FOR LIQUIDS; WIND, SPRING, OR WEIGHT MOTORS; PRODUCING MECHANICAL POWER OR A REACTIVE PROPULSIVE THRUST, NOT OTHERWISE PROVIDED FOR
    • F03D—WIND MOTORS
    • F03D7/00—Controlling wind motors 
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F03—MACHINES OR ENGINES FOR LIQUIDS; WIND, SPRING, OR WEIGHT MOTORS; PRODUCING MECHANICAL POWER OR A REACTIVE PROPULSIVE THRUST, NOT OTHERWISE PROVIDED FOR
    • F03D—WIND MOTORS
    • F03D7/00—Controlling wind motors 
    • F03D7/02—Controlling wind motors  the wind motors having rotation axis substantially parallel to the air flow entering the rotor
    • F03D7/026—Controlling wind motors  the wind motors having rotation axis substantially parallel to the air flow entering the rotor for starting-up
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F03—MACHINES OR ENGINES FOR LIQUIDS; WIND, SPRING, OR WEIGHT MOTORS; PRODUCING MECHANICAL POWER OR A REACTIVE PROPULSIVE THRUST, NOT OTHERWISE PROVIDED FOR
    • F03D—WIND MOTORS
    • F03D80/00—Details, components or accessories not provided for in groups F03D1/00 - F03D17/00
    • F03D80/40—Ice detection; De-icing means
    • 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
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A30/00—Adapting or protecting infrastructure or their operation
    • 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

Abstract

The invention discloses a quick unit startup control method and a quick unit startup control system based on icing grade estimation, wherein the method comprises the following steps of: 1) acquiring a state video of a set blade; 2) constructing a unit blade key frame extraction network; 3) constructing an unsupervised unit blade ice melting data set according to the extracted unit blade state video frames; 4) training an ice melting state network and determining a startup threshold curve according to an unsupervised unit blade ice melting data set; 5) predicting the ice melting state of the blades of the unit according to the ice melting state network and the startup threshold value to obtain a KL divergence value; 6) according to the KL divergence value, predicting the normal starting time of the unit, and further controlling the unit to start quickly; the method can actively acquire the state video of the blades of the unit, can recognize and extract the state video frames of the blades of the unit by utilizing the running state of the unit, and avoids manual marking work; meanwhile, the ice melting state of the blades of the unit and the normal starting time of the unit are predicted, so that the unit is started in time, and the generating capacity of the unit is improved.

Description

Quick unit startup control method and system based on icing level estimation
Technical Field
The invention relates to the technical field of wind turbine generator starting analysis, in particular to a generator set quick starting control method and system based on icing level estimation.
Background
When a large wind turbine generator encounters storm or sleet weather, blade icing linearity can occur. When the fan blade freezes, need accurate judgement icing state to in time shut down and avoid causing the blade damage. After the coated ice melts, the ice melting state needs to be judged, and the machine is started as soon as possible after the coated ice melts, so that the generating capacity is improved. The technical scheme mainly judges the icing state without judging subsequent starting conditions. The main methods for judging the icing state include: 1) based on visual sensors and video image analysis techniques; 2) blade temperature sensor based technology; 3) methods based on manual observation.
At present, most wind fields lack a scheme for analyzing automatic startup conditions of a unit after blades melt ice. And an indirect prejudgment mode is usually adopted for a small part of wind power plants, whether the actual operating power of a fan is matched with the theoretical power or not is compared through wind speed and direction monitoring, and whether the ice is melted or not is prejudged in combination with the current outdoor temperature and the time period. The prejudgment mode depends on whether the wind power curves are matched or not, and the factors causing the wind power mismatch are many. Therefore, the pre-judgment of whether the unit melts the ice is carried out by a plurality of false alarms, and the unit is not started in time, so that part of the generated energy is lost.
Disclosure of Invention
The invention aims to solve the defects in the prior art and provides a unit quick startup control method and a unit quick startup control system based on icing grade estimation. The method can realize real-time analysis of the state video of the unit blade, can monitor the ice-melted state of the unit blade in time, and can start the unit blade in time, thereby improving the generated energy.
In order to achieve the purpose, the technical scheme provided by the invention is as follows: a quick unit startup control method based on icing level estimation comprises the following steps:
1) shooting a state video of the blades of the unit in real time or at regular time by using a network cloud deck camera;
2) establishing a unit blade key frame extraction network, analyzing a newly acquired unit blade state video in real time, and extracting unit blade state video frames of the unit blade in normal states of icing, ice melting and non-icing respectively;
3) constructing an unsupervised unit blade ice melting data set according to the extracted unit blade state video frame, wherein the unsupervised unit blade ice melting data set comprises an expert data set and a training set;
4) training an ice melting state network and determining a startup threshold curve according to the constructed unsupervised unit blade ice melting data set;
5) predicting the ice melting state of the blades of the unit according to the ice melting state network and the startup threshold value to obtain a KL divergence value;
6) and predicting the normal starting time of the unit according to the KL divergence value, and further controlling the unit to start quickly.
Further, in step 2), the following operations are specifically performed:
setting a network backbone of the unit blade key frame extraction network as a Resnet or alexnet deep convolutional network, and setting the output of the unit blade key frame extraction network as single regression output or binary classification output; when the output result of the unit blade key frame extraction network is the angle of the unit blade in the unit blade state video frame, setting an angle threshold value, and when the angle of the unit blade is smaller than the angle threshold value, extracting the unit blade state video frames of the unit blade in normal states of icing, ice melting and non-icing respectively; when the output result of the unit blade key frame extraction network is to identify whether blades exist in all unit blade state video frames, the unit blade state video frames with the blades need to be screened out first, then the position of the unit blade needs to be analyzed by using a computer vision technology, and then the unit blade state video frames of the unit blade in normal states of icing, ice melting and non-icing are extracted.
Further, the computer vision technique is an inter-frame difference method, a contour extraction method, or an optical flow method.
Further, in step 3), the following operations are specifically performed:
constructing an expert data set:
P＝{Ti,Vi,Bi,Si,Mi,Fi|i＝1,2,…}
Ti＝[t1,t2,…]
Vi＝[v1,v2,…]
Figure BDA0003353569330000031
Mi＝[m1,m2,…]
wherein, FiWhen 0 indicates that the blade is not frozen, FiWhen 1 indicates that the unit blade is frozen, TiIs a sequence of times t, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps of (1) collecting all unit blade state video frames of unit blades in normal states of icing, ice melting and non-icing respectively, wherein i is a constant;
constructing a training set:
sequence V for calculating wind speed ViMean value of wind speed of
Figure BDA0003353569330000032
And according to the mean value of wind speed
Figure BDA0003353569330000033
Ordering the expert data sets; getWind speed mean of any two sets of data
Figure BDA0003353569330000034
And
Figure BDA0003353569330000035
where i is 1,2, …, j is 1,2, … and j ≠ i, if the wind speed average of two sets of data
Figure BDA0003353569330000036
And
Figure BDA0003353569330000037
is not greater than epsilon, i.e.
Figure BDA0003353569330000038
And both sets of data are in a non-icing state or one set of data is in a non-icing state, namely Fi+FjWhen the number is less than 2, the two groups of data are constructed into a training sample
Figure BDA0003353569330000039
Training set Q ═ QkWhere k is a constant, ε is a constant, MjIs the set of all the video frames of the state of the blades of the unit in the normal states of icing, ice melting and non-icing respectively SjIs a sequence of generator/impeller speed omega, torque tau and pitch angle beta.
Further, in step 4), the following operations are specifically performed:
the ice-melting state network is of a twin network structure, two samples are randomly extracted from an expert data set P, a unit blade state video frame is selected from each sample and input into a backbone network of the ice-melting state network, and the characteristics of the unit blade state video frame are extracted; and then splicing the characteristics of the two unit blade state video frames, inputting the spliced characteristics and the average wind speed into a full-connection layer of the ice-melting state network, wherein the output result of the ice-melting state network is the measurement of KL divergence of two samples, which is used for measuring the difference of the unit running states, and KL (S) is used for simultaneously measuring the difference of the unit running statesi,Sj) As supervisionInformation to train the ice-melt state network;
meanwhile, a scatter diagram of the training set Q is drawn, with the horizontal axis as
Figure BDA0003353569330000041
The longitudinal axis is KL (S)i,Sj) (ii) a When S isi,SjAll in an unfrozen state, i.e. Fi＝Fj＝0，KL(Si,Sj) Below the threshold curve; when S isi,SjWhen either is frozen, i.e. Fi＝0,Fj1 or Fi＝1,Fj＝0，KL(Si,Sj) The starting threshold curve is marked as h (v) above the threshold curve;
wherein the content of the first and second substances,
Figure BDA0003353569330000042
Fiwhen 0 indicates that the blade is not frozen, FiWhen the number is 1, the unit blade is frozen, and the training set is Q ═ { Q ═ Qk}，
Figure BDA0003353569330000043
k is a constant number, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiFor the set of all the video frames of the unit blade state when the unit blade is in the normal states of icing, melting ice and non-icing respectively, i is 1,2, …, j is 1,2, … and j is not equal to i.
Further, in step 5), the following operations are specifically performed:
acquiring two unit blade state video frames m*And v*And extracting all the satisfaction from the expert data set P
Figure BDA0003353569330000044
And FiData of 0, m*And all satisfy
Figure BDA0003353569330000045
And FiEach machine in the data of 0Group blade state video frame one-time combination
Figure BDA0003353569330000046
Inputting the obtained data into the ice-melting state network to obtain KL divergence value lambdai(ii) a Wherein the content of the first and second substances,
Figure BDA0003353569330000051
ε is a constant, FiWhen 0 indicates that the blade is not frozen, FiWhen 1 indicates that the unit blade is frozen, TiIs a sequence of times t, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps of collecting all the video frames of the state of the unit blades when the unit blades are in normal states of icing, melting ice and non-icing respectively.
Further, in step 6), the following operations are specifically performed:
get t1Time t2Time t and3wind speed and KL divergence value at time t1The wind speed at the moment is v1And KL has a divergence value of lambda1At t2The wind speed at the moment is v2And KL has a divergence value of lambda2At t3The wind speed at the moment is v3And KL has a divergence value of lambda3Using cubic spline function to pair the sequence [ t ]1,t2,t3]And [ lambda ]1-h(v1),λ2-h(v2),λ3-h(v3)]Fitting is carried out, namely the time of the unit entering the starting area can be predicted, and the unit is controlled to be started quickly; wherein h (v) is a startup threshold curve.
The invention provides a unit quick start control system based on icing grade estimation, which comprises:
the unit blade state video shooting module is used for shooting a unit blade state video in real time or at regular time;
the unit blade key frame extraction module is used for extracting unit blade state video frames of the unit blades in normal states of icing, ice melting and non-icing respectively;
the ice melting state calculation module is used for constructing an unsupervised unit blade ice melting data set according to the extracted unit blade state video frames and training an ice melting state network, and the ice melting state network determines a startup threshold and calculates KL divergence;
and the unit starting prediction module predicts the normal starting time of the unit according to the KL divergence value so as to control the unit to start quickly.
Further, the ice-melting state calculation module specifically executes the following operations:
constructing an expert data set:
P＝{Ti,Vi,Bi,Si,Mi,Fi|i＝1,2,…}
Ti＝[t1,t2,…]
Vi＝[v1,v2,…]
Figure BDA0003353569330000061
Mi＝[m1,m2,…]
wherein, FiWhen 0 indicates that the blade is not frozen, FiWhen 1 indicates that the unit blade is frozen, TiIs a sequence of times t, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps of collecting all unit blade state video frames of unit blades in normal states of icing, ice melting and non-icing respectively;
constructing a training set:
sequence V for calculating wind speed ViMean value of wind speed of
Figure BDA0003353569330000062
And according to the mean value of wind speed
Figure BDA0003353569330000063
Ordering the expert data sets; taking the mean value of wind speed of any two groups of data
Figure BDA0003353569330000064
And
Figure BDA0003353569330000065
where i is 1,2, …, j is 1,2, … and j ≠ i, if the wind speed average of two sets of data
Figure BDA0003353569330000066
And
Figure BDA0003353569330000067
is not greater than epsilon, i.e.
Figure BDA0003353569330000068
And both sets of data are in a non-icing state or one set of data is in a non-icing state, namely Fi+FjWhen the number is less than 2, the two groups of data are constructed into a training sample
Figure BDA0003353569330000069
Training set Q ═ QkWhere k is a constant, ε is a constant, MjIs the set of all the video frames of the state of the blades of the unit in the normal states of icing, ice melting and non-icing respectively SjThe sequence of the generator/impeller rotation speed omega, the torque tau and the variable pitch angle beta is obtained;
the ice-melting state network is of a twin network structure, two samples are randomly extracted from an expert data set P, a unit blade state video frame is selected from each sample and input into a backbone network of the ice-melting state network, and the characteristics of the unit blade state video frame are extracted; and then splicing the characteristics of the two unit blade state video frames, inputting the spliced characteristics and the average wind speed into a full-connection layer of the ice-melting state network, wherein the output result of the ice-melting state network is the measurement of KL divergence of two samples, which is used for measuring the difference of the unit running states, and KL (Si, S) is used for simultaneously measuring the difference of the unit running statesj) As supervision information, thereby training the ice-melting state network;
meanwhile, a scatter diagram of the training set Q is drawn, with the horizontal axis as
Figure BDA0003353569330000071
The longitudinal axis is KL (S)i,Sj) (ii) a When S isi,SjAll in an unfrozen state, i.e. Fi＝Fj＝0，KL(Si,Sj) Below the threshold curve; when S isi,SjWhen either is frozen, i.e. Fi＝0,Fj1 or Fi＝1,Fj＝0，KL(Si,Sj) The starting threshold curve is marked as h (v) above the threshold curve;
wherein the content of the first and second substances,
Figure BDA0003353569330000072
Fiwhen 0 indicates that the blade is not frozen, FiWhen the number is 1, the unit blade is frozen, and the training set is Q ═ { Q ═ Qk}，
Figure BDA0003353569330000073
k is a constant number, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps that a set of all unit blade state video frames of a unit blade in normal states of icing, melting ice and non-icing is obtained, wherein i is 1,2, …, j is 1,2, …, and j is not equal to i;
acquiring two unit blade state video frames m*And v*And extracting all the satisfaction from the expert data set P
Figure BDA0003353569330000074
And FiData of 0, m*And all satisfy
Figure BDA0003353569330000075
And FiCombining each set blade state video frame in data of 0 once
Figure BDA0003353569330000076
Inputting the obtained data into the ice-melting state network to obtain KL divergence value lambdaiWhereinε is a constant.
Further, the unit startup prediction module specifically executes the following operations:
according to the calculation of the ice melting state calculation module, t is taken1Time t2Time t and3the wind speed and KL divergence value at the moment are obtained as t1The wind speed at the moment is v1And KL has a divergence value of lambda1At t2The wind speed at the moment is v2And KL has a divergence value of lambda2At t3The wind speed at the moment is v3And KL has a divergence value of lambda3Using cubic spline function to pair the sequence [ t ]1,t2,t3]And [ lambda ]1-h(v1),λ2-h(v2),λ3-h(v3)]Fitting is carried out, namely the time of the unit entering the starting area can be predicted, and the unit is controlled to be started quickly; wherein h (v) is a startup threshold curve.
Compared with the prior art, the invention has the following advantages and beneficial effects:
the method can actively acquire the state video of the blades of the unit, can recognize and extract the state video frames of the blades of the unit by utilizing the running state of the unit, and avoids manual marking work; meanwhile, the ice melting state of the blades of the unit and the normal starting time of the unit are predicted, and the unit can be started in time, so that the generating capacity of the unit is improved.
Drawings
Fig. 1 is a flow chart of a unit quick start control method.
FIG. 2 is a block diagram of an ice-melt state network.
FIG. 3 is a scatter plot of the training set.
Detailed Description
The present invention will be further described with reference to the following specific examples.
Referring to fig. 1, the method for controlling quick start-up of a unit based on icing level estimation provided by the present embodiment includes the following steps:
1) shooting a state video of the blades of the unit in real time or at regular time by using a network cloud deck camera;
2) establishing a unit blade key frame extraction network, analyzing a newly acquired unit blade state video in real time, extracting unit blade state video frames of a unit blade in normal states of icing, ice melting and non-icing respectively, and specifically executing the following operations:
setting a network backbone of the unit blade key frame extraction network as a Resnet or alexnet deep convolutional network, and setting the output of the unit blade key frame extraction network as single regression output or binary classification output; when the output result of the unit blade key frame extraction network is the angle of the unit blade in the unit blade state video frame, setting an angle threshold value, and when the angle of the unit blade is smaller than the angle threshold value, extracting the unit blade state video frames of the unit blade in normal states of icing, ice melting and non-icing respectively; when the output result of the unit blade key frame extraction network is to identify whether blades exist in all unit blade state video frames, the unit blade state video frames with the blades need to be screened out firstly, and then the positions of the unit blades are analyzed by using a computer vision technology, so that the unit blade state video frames of the unit blades in normal states of icing, ice melting and non-icing are extracted; the computer vision technology is an interframe difference method, a contour extraction method or an optical flow method.
The unit blade key frame extraction network implementation structure is shown in the following table 1:
Figure BDA0003353569330000091
table 1 unit blade key frame extraction network implementation structure
3) According to the extracted unit blade state video frames, an unsupervised unit blade ice melting data set is constructed, the unsupervised unit blade ice melting data set comprises an expert data set and a training set, and the following operations are specifically executed:
constructing an expert data set:
P＝{Ti,Vi,Bi,Si,Mi,Fi|i＝1,2,…}
Ti＝[t1,t2,…]
Vi＝[v1,v2,…]
Figure BDA0003353569330000092
Mi＝[m1,m2,…]
wherein, FiWhen 0 indicates that the blade is not frozen, FiWhen 1 indicates that the unit blade is frozen, TiIs a sequence of times t, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps of (1) collecting all unit blade state video frames of unit blades in normal states of icing, ice melting and non-icing respectively, wherein i is a constant;
constructing a training set:
sequence V for calculating wind speed ViMean value of wind speed of
Figure BDA0003353569330000101
And according to the mean value of wind speed
Figure BDA0003353569330000102
Ordering the expert data sets; taking the mean value of wind speed of any two groups of data
Figure BDA0003353569330000103
And
Figure BDA0003353569330000104
where i is 1,2, …, j is 1,2, … and j ≠ i, if the wind speed average of two sets of data
Figure BDA0003353569330000105
And
Figure BDA0003353569330000106
is not greater than epsilon, i.e.
Figure BDA0003353569330000107
And both sets of data are in a non-icing state or one set of data is in a non-icing state, namely Fi+FjWhen the number is less than 2, the two groups of data are constructed into a training sample
Figure BDA0003353569330000108
Training set Q ═ QkWhere k is a constant, ε is a constant, MjIs the set of all the video frames of the state of the blades of the unit in the normal states of icing, ice melting and non-icing respectively SjIs a sequence of generator/impeller speed omega, torque tau and pitch angle beta.
4) According to the built unsupervised unit blade ice melting data set, training an ice melting state network and determining a startup threshold curve, and specifically executing the following operations:
referring to fig. 2, the ice-melting state network has a twin network structure, two samples are randomly extracted from an expert data set P, a set blade state video frame is selected from each sample and input into a backbone network of the ice-melting state network, and the characteristics of the set blade state video frame are extracted; and then splicing the characteristics of the two unit blade state video frames, inputting the spliced characteristics and the average wind speed into a full-connection layer of the ice-melting state network, wherein the output result of the ice-melting state network is the measurement of KL divergence of two samples, which is used for measuring the difference of the unit running states, and KL (S) is used for simultaneously measuring the difference of the unit running statesi,Sj) As supervision information, thereby training the ice-melting state network;
the ice-melting state network implementation structure is shown in the following table 2:
Figure BDA0003353569330000109
Figure BDA0003353569330000111
TABLE 2 Ice-melt State network implementation Structure
Meanwhile, referring to FIG. 3, the training is plottedThe horizontal axis of the scattergram of training set Q is
Figure BDA0003353569330000112
The longitudinal axis is KL (S)i,Sj) (ii) a When S isi,SjAll in an unfrozen state, i.e. Fi＝Fj＝0，KL(Si,Sj) Below the threshold curve; when S isi,SjWhen either is frozen, i.e. Fi＝0,Fj1 or Fi＝1,Fj＝0，KL(Si,Sj) The starting threshold curve is marked as h (v) above the threshold curve;
wherein the content of the first and second substances,
Figure BDA0003353569330000113
Fiwhen 0 indicates that the blade is not frozen, FiWhen the number is 1, the unit blade is frozen, and the training set is Q ═ { Q ═ Qk}，
Figure BDA0003353569330000114
k is a constant number, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiFor the set of all the video frames of the unit blade state when the unit blade is in the normal states of icing, melting ice and non-icing respectively, i is 1,2, …, j is 1,2, … and j is not equal to i.
5) Predicting the ice melting state of the blades of the unit according to the ice melting state network and the startup threshold value to obtain a KL divergence value, and specifically executing the following operations:
acquiring two unit blade state video frames m*And v*And extracting all the satisfaction from the expert data set P
Figure BDA0003353569330000115
And FiData of 0, m*And all satisfy
Figure BDA0003353569330000116
And FiOnce per unit blade status video frame in data equal to 0Combination of
Figure BDA0003353569330000117
Inputting the obtained data into the ice-melting state network to obtain KL divergence value lambdai(ii) a Wherein the content of the first and second substances,
Figure BDA0003353569330000121
ε is a constant, FiWhen 0 indicates that the blade is not frozen, FiWhen 1 indicates that the unit blade is frozen, TiIs a sequence of times t, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps of collecting all the video frames of the state of the unit blades when the unit blades are in normal states of icing, melting ice and non-icing respectively.
6) According to the KL divergence value, predicting the normal starting time of the unit, further controlling the unit to start quickly, and specifically executing the following operations:
get t1Time t2Time t and3wind speed and KL divergence value at time t1The wind speed at the moment is v1And KL has a divergence value of lambda1At t2The wind speed at the moment is v2And KL has a divergence value of lambda2At t3The wind speed at the moment is v3And KL has a divergence value of lambda3Using cubic spline function to pair the sequence [ t ]1,t2,t3]And [ lambda ]1-h(v1),λ2-h(v2),λ3-h(v3)]Fitting is carried out, namely the time of the unit entering the starting area can be predicted, and the unit is controlled to be started quickly; wherein h (v) is a startup threshold curve.
The following fast starting control system for a unit based on icing level estimation provided by this embodiment includes:
the unit blade state video shooting module is used for shooting a unit blade state video in real time or at regular time;
the unit blade key frame extraction module is used for extracting unit blade state video frames of the unit blades in normal states of icing, ice melting and non-icing respectively;
the ice melting state calculation module is used for constructing an unsupervised unit blade ice melting data set according to the extracted unit blade state video frames and training an ice melting state network, the ice melting state network determines a startup threshold and calculates KL divergence, and the following operations are specifically executed:
constructing an expert data set:
P＝{Ti,Vi,Bi,Si,Mi,Fi|i＝1,2,…}
Ti＝[t1,t2,…]
Vi＝[v1,v2,…]
Figure BDA0003353569330000131
Mi＝[m1,m2,…]
wherein, FiWhen 0 indicates that the blade is not frozen, FiWhen 1 indicates that the unit blade is frozen, TiIs a sequence of times t, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps of collecting all unit blade state video frames of unit blades in normal states of icing, ice melting and non-icing respectively;
constructing a training set:
sequence V for calculating wind speed ViMean value of wind speed of
Figure BDA0003353569330000132
And according to the mean value of wind speed
Figure BDA0003353569330000133
Ordering the expert data sets; taking the mean value of wind speed of any two groups of data
Figure BDA0003353569330000134
And
Figure BDA0003353569330000135
where i is 1,2, …, j is 1,2, … and j ≠ i, if the wind speed average of two sets of data
Figure BDA0003353569330000136
And
Figure BDA0003353569330000137
is not greater than epsilon, i.e.
Figure BDA0003353569330000138
And both sets of data are in a non-icing state or one set of data is in a non-icing state, namely Fi+FjWhen the number is less than 2, the two groups of data are constructed into a training sample
Figure BDA0003353569330000139
Training set Q ═ QkWhere k is a constant, ε is a constant, MjIs the set of all the video frames of the state of the blades of the unit in the normal states of icing, ice melting and non-icing respectively SjThe sequence of the generator/impeller rotation speed omega, the torque tau and the variable pitch angle beta is obtained;
the ice-melting state network is of a twin network structure, two samples are randomly extracted from an expert data set P, a unit blade state video frame is selected from each sample and input into a backbone network of the ice-melting state network, and the characteristics of the unit blade state video frame are extracted; and then splicing the characteristics of the two unit blade state video frames, inputting the spliced characteristics and the average wind speed into a full-connection layer of the ice-melting state network, wherein the output result of the ice-melting state network is the measurement of KL divergence of two samples, which is used for measuring the difference of the unit running states, and KL (Si, S) is used for simultaneously measuring the difference of the unit running statesj) As supervision information, thereby training the ice-melting state network;
meanwhile, a scatter diagram of the training set Q is drawn, with the horizontal axis as
Figure BDA0003353569330000141
The longitudinal axis is KL (S)i,Sj) (ii) a When S isi,SjAre all unknottedIn the ice state, i.e. Fi＝Fj＝0，KL(Si,Sj) Below the threshold curve; when S isi,SjWhen either is frozen, i.e. Fi＝0,Fj1 or Fi＝1,Fj＝0，KL(Si,Sj) The starting threshold curve is marked as h (v) above the threshold curve;
wherein the content of the first and second substances,
Figure BDA0003353569330000142
Fiwhen 0 indicates that the blade is not frozen, FiWhen the number is 1, the unit blade is frozen, and the training set is Q ═ { Q ═ Qk}，
Figure BDA0003353569330000143
k is a constant number, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps that a set of all unit blade state video frames of a unit blade in normal states of icing, melting ice and non-icing is obtained, wherein i is 1,2, …, j is 1,2, …, and j is not equal to i;
acquiring two unit blade state video frames m*And v*And extracting all the satisfaction from the expert data set P
Figure BDA0003353569330000144
And FiData of 0, m*And all satisfy
Figure BDA0003353569330000145
And FiCombining each set blade state video frame in data of 0 once
Figure BDA0003353569330000146
Inputting the obtained data into the ice-melting state network to obtain KL divergence value lambdaiWhere ε is a constant.
The unit startup prediction module predicts the normal startup time of the unit according to the KL divergence value, and then controls the unit to rapidly start, and specifically executes the following operations:
according to the calculation of the ice melting state calculation module, t is taken1Time t2Time t and3the wind speed and KL divergence value at the moment are obtained as t1The wind speed at the moment is v1And KL has a divergence value of lambda1At t2The wind speed at the moment is v2And KL has a divergence value of lambda2At t3The wind speed at the moment is v3And KL has a divergence value of lambda3Using cubic spline function to pair the sequence [ t ]1,t2,t3]And [ lambda ]1-h(v1),λ2-h(v2),λ3-h(v3)]Fitting is carried out, namely the time of the unit entering the starting area can be predicted, and the unit is controlled to be started quickly; wherein h (v) is a startup threshold curve.
The above-described embodiments are only preferred embodiments of the present invention, and not intended to limit the scope of the present invention, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents and as included within the scope of the present invention.

Claims (10)

1. A unit quick start control method based on icing grade estimation is characterized by comprising the following steps:
1) shooting a state video of the blades of the unit in real time or at regular time by using a network cloud deck camera;
2) establishing a unit blade key frame extraction network, analyzing a newly acquired unit blade state video in real time, and extracting unit blade state video frames of the unit blade in normal states of icing, ice melting and non-icing respectively;
3) constructing an unsupervised unit blade ice melting data set according to the extracted unit blade state video frame, wherein the unsupervised unit blade ice melting data set comprises an expert data set and a training set;
4) training an ice melting state network and determining a startup threshold curve according to the constructed unsupervised unit blade ice melting data set;
5) predicting the ice melting state of the blades of the unit according to the ice melting state network and the startup threshold value to obtain a KL divergence value;
6) and predicting the normal starting time of the unit according to the KL divergence value, and further controlling the unit to start quickly.
2. The method for controlling the quick startup of the unit based on the icing level estimation according to claim 1, wherein in step 2), the following operations are specifically performed:
setting a network backbone of the unit blade key frame extraction network as a Resnet or alexnet deep convolutional network, and setting the output of the unit blade key frame extraction network as single regression output or binary classification output; when the output result of the unit blade key frame extraction network is the angle of the unit blade in the unit blade state video frame, setting an angle threshold value, and when the angle of the unit blade is smaller than the angle threshold value, extracting the unit blade state video frames of the unit blade in normal states of icing, ice melting and non-icing respectively; when the output result of the unit blade key frame extraction network is to identify whether blades exist in all unit blade state video frames, the unit blade state video frames with the blades need to be screened out first, then the position of the unit blade needs to be analyzed by using a computer vision technology, and then the unit blade state video frames of the unit blade in normal states of icing, ice melting and non-icing are extracted.
3. The method for controlling the quick startup of the unit based on the icing level estimation as claimed in claim 2, wherein the computer vision technique is an inter-frame difference method, a contour extraction method or an optical flow method.
4. The method for controlling the quick startup of the unit based on the icing level estimation according to claim 1, wherein in step 3), the following operations are specifically performed:
constructing an expert data set:
P＝{Ti,Vi,Bi,Si,Mi,Fi|i＝1,2,…}
Ti＝[t1,t2,…]
Vi＝[v1,v2,…]
Figure FDA0003353569320000021
Mi＝[m1,m2,…]
wherein, FiWhen 0 indicates that the blade is not frozen, FiWhen 1 indicates that the unit blade is frozen, TiIs a sequence of times t, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps of (1) collecting all unit blade state video frames of unit blades in normal states of icing, ice melting and non-icing respectively, wherein i is a constant;
constructing a training set:
sequence V for calculating wind speed ViMean value of wind speed of
Figure FDA0003353569320000022
And according to the mean value of wind speed
Figure FDA0003353569320000023
Ordering the expert data sets; taking the mean value of wind speed of any two groups of data
Figure FDA0003353569320000024
And
Figure FDA0003353569320000025
where i is 1,2, …, j is 1,2, … and j ≠ i, if the wind speed average of two sets of data
Figure FDA0003353569320000026
And
Figure FDA0003353569320000027
difference of (2)A value not greater than ε, i.e
Figure FDA0003353569320000028
And both sets of data are in a non-icing state or one set of data is in a non-icing state, namely Fi+FjWhen the number is less than 2, the two groups of data are constructed into a training sample
Figure FDA0003353569320000029
Training set Q ═ QkWhere k is a constant, ε is a constant, MjIs the set of all the video frames of the state of the blades of the unit in the normal states of icing, ice melting and non-icing respectively SjIs a sequence of generator/impeller speed omega, torque tau and pitch angle beta.
5. The method for controlling the quick startup of the unit based on the icing level estimation according to claim 1, wherein in step 4), the following operations are specifically performed:
the ice-melting state network is of a twin network structure, two samples are randomly extracted from an expert data set P, a unit blade state video frame is selected from each sample and input into a backbone network of the ice-melting state network, and the characteristics of the unit blade state video frame are extracted; and then splicing the characteristics of the two unit blade state video frames, inputting the spliced characteristics and the average wind speed into a full-connection layer of the ice-melting state network, wherein the output result of the ice-melting state network is the measurement of KL divergence of two samples, which is used for measuring the difference of the unit running states, and KL (S) is used for simultaneously measuring the difference of the unit running statesi,Sj) As supervision information, thereby training the ice-melting state network;
meanwhile, a scatter diagram of the training set Q is drawn, with the horizontal axis as
Figure FDA0003353569320000031
The longitudinal axis is KL (S)i,Sj) (ii) a When S isi,SjAll in an unfrozen state, i.e. Fi＝Fj＝0，KL(Si,Sj) Under the threshold curveA method for preparing; when S isi,SjWhen either is frozen, i.e. Fi＝0,Fj1 or Fi＝1,Fj＝0，KL(Si,Sj) The starting threshold curve is marked as h (v) above the threshold curve;
wherein the content of the first and second substances,
Figure FDA0003353569320000032
Fiwhen 0 indicates that the blade is not frozen, FiWhen the number is 1, the unit blade is frozen, and the training set is Q ═ { Q ═ Qk}，
Figure FDA0003353569320000033
k is a constant number, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiFor the set of all the video frames of the unit blade state when the unit blade is in the normal states of icing, melting ice and non-icing respectively, i is 1,2, …, j is 1,2, … and j is not equal to i.
6. The method for controlling the quick startup of the unit based on the icing level estimation according to claim 1, wherein in step 5), the following operations are specifically performed:
acquiring two unit blade state video frames m*And v*And extracting all the satisfaction from the expert data set P
Figure FDA0003353569320000041
And FiData of 0, m*And all satisfy
Figure FDA0003353569320000042
And FiCombining each set blade state video frame in data of 0 once
Figure FDA0003353569320000043
Inputting the obtained data into the ice-melting state network to obtain KL divergence value lambdai(ii) a Wherein the content of the first and second substances,
Figure FDA0003353569320000044
ε is a constant, FiWhen 0 indicates that the blade is not frozen, FiWhen 1 indicates that the unit blade is frozen, TiIs a sequence of times t, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps of collecting all the video frames of the state of the unit blades when the unit blades are in normal states of icing, melting ice and non-icing respectively.
7. The method for controlling the quick startup of the unit based on the icing level estimation according to claim 1, wherein in step 6), the following operations are specifically performed:
get t1Time t2Time t and3wind speed and KL divergence value at time t1The wind speed at the moment is v1And KL has a divergence value of lambda1At t2The wind speed at the moment is v2And KL has a divergence value of lambda2At t3The wind speed at the moment is v3And KL has a divergence value of lambda3Using cubic spline function to pair the sequence [ t ]1,t2,t3]And [ lambda ]1-h(v1),λ2-h(v2),λ3-h(v3)]Fitting is carried out, namely the time of the unit entering the starting area can be predicted, and the unit is controlled to be started quickly; wherein h (v) is a startup threshold curve.
8. A unit quick start control system based on icing level estimation is characterized by comprising:
the unit blade state video shooting module is used for shooting a unit blade state video in real time or at regular time;
the unit blade key frame extraction module is used for extracting unit blade state video frames of the unit blades in normal states of icing, ice melting and non-icing respectively;
the ice melting state calculation module is used for constructing an unsupervised unit blade ice melting data set according to the extracted unit blade state video frames and training an ice melting state network, and the ice melting state network determines a startup threshold and calculates KL divergence;
and the unit starting prediction module predicts the normal starting time of the unit according to the KL divergence value so as to control the unit to start quickly.
9. The system according to claim 8, wherein the ice-melting state calculation module performs the following operations:
constructing an expert data set:
P＝{Ti,Vi,Bi,Si,Mi,Fi|i＝1,2,…}
Ti＝[t1,t2,…]
Vi＝[v1,v2,…]
Figure FDA0003353569320000051
Mi＝[m1,m2,…]
wherein, FiWhen 0 indicates that the blade is not frozen, FiWhen 1 indicates that the unit blade is frozen, TiIs a sequence of times t, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps of collecting all unit blade state video frames of unit blades in normal states of icing, ice melting and non-icing respectively;
constructing a training set:
sequence V for calculating wind speed ViMean value of wind speed of
Figure FDA0003353569320000052
And according to the mean value of wind speed
Figure FDA0003353569320000053
Ordering the expert data sets; taking the mean value of wind speed of any two groups of data
Figure FDA0003353569320000054
And
Figure FDA0003353569320000055
where i is 1,2, …, j is 1,2, … and j ≠ i, if the wind speed average of two sets of data
Figure FDA0003353569320000056
And
Figure FDA0003353569320000057
is not greater than epsilon, i.e.
Figure FDA0003353569320000058
And both sets of data are in a non-icing state or one set of data is in a non-icing state, namely Fi+FjWhen the number is less than 2, the two groups of data are constructed into a training sample
Figure FDA0003353569320000059
Training set Q ═ QkWhere k is a constant, ε is a constant, MjIs the set of all the video frames of the state of the blades of the unit in the normal states of icing, ice melting and non-icing respectively SjThe sequence of the generator/impeller rotation speed omega, the torque tau and the variable pitch angle beta is obtained;
the ice-melting state network is of a twin network structure, two samples are randomly extracted from an expert data set P, a unit blade state video frame is selected from each sample and input into a backbone network of the ice-melting state network, and the characteristics of the unit blade state video frame are extracted; and then splicing the characteristics of the two unit blade state video frames, inputting the spliced characteristics and the average wind speed into a full connection layer of the ice-melting state network, wherein the output result of the ice-melting state network is the measurement of KL divergence of two samples and is used for measuring the unit operationDifference of line state, and simultaneous activation of KL (S)i,Sj) As supervision information, thereby training the ice-melting state network;
meanwhile, a scatter diagram of the training set Q is drawn, with the horizontal axis as
Figure FDA0003353569320000061
The longitudinal axis is KL (S)i,Sj) (ii) a When S isi,SjAll in an unfrozen state, i.e. Fi＝Fj＝0，KL(Si,Sj) Below the threshold curve; when S isi,SjWhen either is frozen, i.e. Fi＝0,Fj1 or Fi＝1,Fj＝0，KL(Si,Sj) The starting threshold curve is marked as h (v) above the threshold curve;
P＝{Ti,Vi,Bi,Si,Mi,Fi|i＝1,2,…}
Ti＝[t1,t2,…]
Vi＝[v1,v2,…]
Figure FDA0003353569320000062
wherein M isi＝[m1,m2,…]，FiWhen 0 indicates that the blade is not frozen, FiWhen the number is 1, the unit blade is frozen, and the training set is Q ═ { Q ═ Qk}，
Figure FDA0003353569320000063
k is a constant number, ViIs a sequence of wind speeds v, SiFor a sequence of generator/impeller speed omega, torque tau and pitch angle beta, MiThe method comprises the steps that a set of all unit blade state video frames of a unit blade in normal states of icing, melting ice and non-icing is obtained, wherein i is 1,2, …, j is 1,2, …, and j is not equal to i;
acquiring two unit blade state video frames m*And v*And extracting all the satisfaction from the expert data set P
Figure FDA0003353569320000064
And FiData of 0, m*And all satisfy
Figure FDA0003353569320000065
And FiCombining each set blade state video frame in data of 0 once
Figure FDA0003353569320000066
Inputting the obtained data into the ice-melting state network to obtain KL divergence value lambdaiWhere ε is a constant.
10. The system according to claim 8, wherein the unit start-up prediction module performs the following operations:
according to the calculation of the ice melting state calculation module, t is taken1Time t2Time t and3the wind speed and KL divergence value at the moment are obtained as t1The wind speed at the moment is v1And KL has a divergence value of lambda1At t2The wind speed at the moment is v2And KL has a divergence value of lambda2At t3The wind speed at the moment is v3And KL has a divergence value of lambda3Using cubic spline function to pair the sequence [ t ]1,t2,t3]And [ lambda ]1-h(v1),λ2-h(v2),λ3-h(v3)]Fitting is carried out, namely the time of the unit entering the starting area can be predicted, and the unit is controlled to be started quickly; wherein h (v) is a startup threshold curve.
CN202111344768.7A 2021-11-15 2021-11-15 Quick unit startup control method and system based on icing level estimation Active CN114060213B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111344768.7A CN114060213B (en) 2021-11-15 2021-11-15 Quick unit startup control method and system based on icing level estimation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111344768.7A CN114060213B (en) 2021-11-15 2021-11-15 Quick unit startup control method and system based on icing level estimation

Publications (2)

Publication Number Publication Date
CN114060213A true CN114060213A (en) 2022-02-18
CN114060213B CN114060213B (en) 2023-03-31

Family

ID=80271823

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111344768.7A Active CN114060213B (en) 2021-11-15 2021-11-15 Quick unit startup control method and system based on icing level estimation

Country Status (1)

Country Link
CN (1) CN114060213B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN203770036U (en) * 2014-01-22 2014-08-13 刘中威 Electric heating ice melting wireless control system for wind driven generator rotor blades
CN105134488A (en) * 2015-08-25 2015-12-09 湘电风能有限公司 Method for starting wind turbine generator
CN111852793A (en) * 2020-07-30 2020-10-30 湖南拓天节能控制技术股份有限公司 Method and device for combined control of fan starting and blade deicing prevention
CN112326213A (en) * 2019-08-05 2021-02-05 株式会社理光 Abnormal data detection method and device and mechanical fault detection method and device

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN203770036U (en) * 2014-01-22 2014-08-13 刘中威 Electric heating ice melting wireless control system for wind driven generator rotor blades
CN105134488A (en) * 2015-08-25 2015-12-09 湘电风能有限公司 Method for starting wind turbine generator
CN112326213A (en) * 2019-08-05 2021-02-05 株式会社理光 Abnormal data detection method and device and mechanical fault detection method and device
CN111852793A (en) * 2020-07-30 2020-10-30 湖南拓天节能控制技术股份有限公司 Method and device for combined control of fan starting and blade deicing prevention

Also Published As

Publication number Publication date
CN114060213B (en) 2023-03-31

Similar Documents

Publication Publication Date Title
CN110261394B (en) Online real-time diagnosis system and method for damage of fan blade
CN111160525B (en) Task unloading intelligent decision-making method based on unmanned aerial vehicle group in edge computing environment
CN109209790A (en) A kind of wind electricity blade icing estimation method based on data modeling
CN111400961B (en) Wind generating set blade fault judging method and device
CN110689642B (en) Abnormal driving distinguishing method and system based on vehicle-mounted OBD data and probability statistics
CN112950634A (en) Method, equipment and system for identifying damage of wind turbine blade based on unmanned aerial vehicle routing inspection
CN114060213B (en) Quick unit startup control method and system based on icing level estimation
CA2927455C (en) Method and system for rotor blade deicing as well as computer program and wind power plant
CN112832960A (en) Fan blade icing detection method based on deep learning and storage medium
CN108999758A (en) Blade icing monitoring method and device for wind driven generator
CN114670787B (en) Control method of vehicle and windscreen wiper and related equipment
CN111042456A (en) Heating snow melting method, system and readable storage medium
LU504690B1 (en) A system for automatic leaf photo acquisition by drones
CN114383341B (en) Active ice and snow melting system for road surface and control method thereof
CN116317800A (en) Permanent magnet synchronous motor temperature prediction method and system based on temperature time sequence input
CN114753980A (en) Method and system for monitoring icing of fan blade
CN112452861B (en) Artificial intelligence-based ice and snow removal adjusting method and device for photovoltaic cleaning robot
CN109322802B (en) Method and system for predicting blade icing of wind driven generator
CN108825452B (en) Method and device for determining blade icing of wind generating set
CN115310517A (en) Aero-engine online anomaly detection method based on improved piecewise linear representation
CN113406107A (en) Fan blade defect detection system
EP3299614A1 (en) Method, device and system for managing a wind farm
CN111079852B (en) Method for detecting icing thickness of insulator of power transmission line
CN113095360A (en) Method for constructing fan icing detection model and application thereof
CN110515400A (en) A kind of discharge counter surface dehumanization method and its device

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