CN113935408B - Wind power generation equipment state monitoring method based on simplified core principal component network - Google Patents
Wind power generation equipment state monitoring method based on simplified core principal component network Download PDFInfo
- Publication number
- CN113935408B CN113935408B CN202111156322.1A CN202111156322A CN113935408B CN 113935408 B CN113935408 B CN 113935408B CN 202111156322 A CN202111156322 A CN 202111156322A CN 113935408 B CN113935408 B CN 113935408B
- Authority
- CN
- China
- Prior art keywords
- vector
- matrix
- power generation
- wind power
- generation equipment
- 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.)
- Active
Links
- 238000010248 power generation Methods 0.000 title claims abstract description 47
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000012544 monitoring process Methods 0.000 title claims abstract description 30
- 238000004458 analytical method Methods 0.000 claims abstract description 9
- 239000013598 vector Substances 0.000 claims description 87
- 239000011159 matrix material Substances 0.000 claims description 44
- 238000005070 sampling Methods 0.000 claims description 20
- 238000004364 calculation method Methods 0.000 claims description 9
- 210000002569 neuron Anatomy 0.000 claims description 9
- 230000001133 acceleration Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000007621 cluster analysis Methods 0.000 claims description 4
- 238000012423 maintenance Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 102000002274 Matrix Metalloproteinases Human genes 0.000 claims description 2
- 108010000684 Matrix Metalloproteinases Proteins 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims description 2
- 238000004422 calculation algorithm Methods 0.000 abstract description 11
- 238000000513 principal component analysis Methods 0.000 abstract description 6
- 238000012549 training Methods 0.000 abstract description 4
- 238000000605 extraction Methods 0.000 abstract description 2
- 238000003064 k means clustering Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 101100272279 Beauveria bassiana Beas gene Proteins 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- F03D17/00—Monitoring or testing of wind motors, e.g. diagnostics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- 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)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Sustainable Energy (AREA)
- Combustion & Propulsion (AREA)
- General Health & Medical Sciences (AREA)
- Mechanical Engineering (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Chemical & Material Sciences (AREA)
- Sustainable Development (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Probability & Statistics with Applications (AREA)
- Wind Motors (AREA)
Abstract
The invention discloses a method for monitoring the state of wind power generation equipment based on a simplified core principal component network. Specifically, the method of the invention uses the edge data points and the clustering center data points to construct the hidden layer of the principal component network by carrying out sample distribution characteristic analysis on the training data set of the wind power generation equipment, thereby greatly reducing the calculated amount when carrying out online monitoring and ensuring the application feasibility and the real-time performance of the state monitoring method based on the simplified principal component network. In addition, the method calculates corresponding outputs through the two output layer coefficient matrixes respectively, and compared with the traditional kernel principal component analysis algorithm which has only one output, the method further ensures the diversity of the feature extraction.
Description
Technical Field
The invention relates to a method for monitoring the state of wind power generation equipment, in particular to a method for monitoring the state of wind power generation equipment based on a simplified core principal component network.
Background
Due to the fluctuation of primary energy prices such as petroleum and the like and the environmental protection characteristic of wind power generation, the advantages of wind power generation are more and more obvious. The five top-ranked countries of the world's wind farm installation are China, the United states, germany, spanish and India, respectively. The wind energy resources in China are rich, the wind energy reserve is 32 hundred million kilowatts, the developable installed capacity is about 2.5 hundred million kilowatts, and the world is top. The installed capacity of wind power in China is rapidly increased in recent years, and the development of domestic wind power generation is promoted. Wind power generation equipment (i.e., wind turbines) is a type of power machinery that converts wind energy into mechanical energy, and meanwhile, the current main form of utilization of wind energy is power generation, which has grown most rapidly in new energy and renewable energy industries.
In general, most wind power generation equipment operates in areas with severe weather, and the operating conditions are extremely poor, so that the wind power generation set has more faults, and a large number of faults cause the wind power generation set to stop operating or cannot reach rated output. Therefore, the running state of the wind power generation equipment is monitored in real time, and running abnormality is found in time so as to quickly organize equipment maintenance, and the method has important practical significance for improving the wind power utilization rate and the wind power generation service quality. In the existing wind power generation equipment set, a plurality of sensors are usually installed in a matched mode, and data information such as the rotating speed of a generator, generated electric power, acceleration and the like is measured and fed back in real time. The data collected by these sensors provides a solid data base for implementing data-driven wind power plant condition monitoring. In current smart manufacturing and big data wind tides, it is very desirable to use these sampled data to implement a solution for monitoring the condition of a wind power plant.
However, the working state of the wind driven generator is directly affected by the wind speed of the external environment, and the working state of the wind driven generator is continuously changed along with the change of the wind speed. Because the intermittent, nonlinear and time-series variation characteristics of wind power are not manually and accurately predictable or controllable, the operating condition of the wind power generator is directly affected by wind speed, which presents challenges for implementing data-driven monitoring of the condition of the wind power generation equipment. The development of equipment fault diagnosis technology in the field of wind power equipment at home is still in an try stage, the development of large-scale wind power generation sets in recent years is very fast in home, but the state monitoring system products of the corresponding large-scale wind power generation sets are still in a primary stage, and the state monitoring system products of the large-scale wind power generation sets in home are still immature, and independent real-time intelligent monitoring equipment is lacked.
In recent years, in the field of data-driven process monitoring, nuclear principal component analysis algorithms have been successfully applied to monitor the operational state of nonlinear industrial processes. While the kernel principal component analysis algorithm can monitor the operating state by mining nonlinear features of the sampled data, there is a key problem: the computational complexity of implementing on-line monitoring is proportional to the size of the training data set, but the larger the training data set is, the more advantageous the implementation of state monitoring. Therefore, if the state monitoring of the wind power generation equipment is implemented by using the kernel principal component analysis algorithm, the online calculation amount is a problem that must be considered in order to ensure the feasibility and the real-time performance of the practical application.
Disclosure of Invention
The main technical problems to be solved by the invention are as follows: how to build a simplified principal component network model and on the basis of this to implement real-time status monitoring of wind power plants. Specifically, the method of the invention uses the edge data points and the clustering center data points to construct the hidden layer of the principal component network by carrying out sample distribution characteristic analysis on the training data set of the wind power generation equipment, thereby greatly reducing the calculated amount when carrying out online monitoring and ensuring the application feasibility and the real-time performance of the state monitoring method based on the simplified principal component network.
The technical scheme adopted by the method for solving the problems is as follows: a wind power generation equipment state monitoring method based on a reduced core principal component network comprises the following steps:
Step (1): after the data which can be measured in real time by the wind power generation equipment are determined, under the normal running state of the wind power generation equipment, acquiring the data according to the inherent sampling time interval of the data acquisition system of the wind power generation equipment; the 11 data which can be measured in real time at each sampling moment are specifically and sequentially: wind speed, rotor speed, generator speed, mechanical torque, generated power, blade pitch angle, blade azimuth, blade root moment, top horizontal axis acceleration, top longitudinal axis acceleration, and yaw error.
Step (2): n sample data vectors X 1,x2,...,xn with wind speed between cut-in wind speed (typically 3 meters per second) and cut-out wind speed (typically 25 meters per second) are formed into a matrix x= [ X 1,x2,...,xn ], and each row vector in X e R 11×n is subjected to normalization processing to obtain a new matrixWherein 11 data in the i-th sample data vector x i∈R11×1 are sequentially arranged in the order of step (1), i∈ {1,2,.. N }, R 11×n represents a real matrix of 11×n dimensions, R represents a real set, and R 11×1 represents a real vector of 11×1 dimensions.
Step (3): for new matrixAfter the analysis of the principal components of the kernel, the kernel parameters delta and the kernel score matrix Y epsilon R D×n are reserved, and the specific implementation process is shown in the steps (3.1) to (3.5).
Step (3.1): according to the formulaCalculating a new matrixIn the ith column vectorAnd the j-th column vectorThe square distance ζ (i, j) between; where j e {1,2,., n }, the upper label T represents the transpose of the matrix or vector.
Step (3.2): the kernel parameter delta is determined according to equation ① as follows:
Step (3.3): the ith row and jth column element K (i, j) in the kernel matrix K e R n×n is calculated according to the formula ② shown below:
in the above formula, e represents a natural constant, i.e., e is about 2.718281828.
Step (3.4): after n eigenvalues lambda 1≥λ2≥...≥λn of the kernel matrix K and corresponding eigenvectors alpha 1,α2,...,αn are calculated, recording the number of eigenvalues not smaller than lambda 0 in lambda 1,λ2,...,λn as D; where λ 0 represents the average of n eigenvalues, i.e., λ 0=(λ1+λ2+...+λn)/n, the length of the eigenvector α 1,α2,...,αn is 1.
Step (3.5): after D eigenvectors α 1,α2,...,αD are assembled into a feature transformation matrix a= [ α 1,α2,...,αD ], a kernel score matrix y∈r D×n is calculated according to the formula y=a T K.
Step (4): for new matrixColumn vectors in (a)And (3) carrying out edge characteristic point analysis so as to obtain E edge point vectors xi 1,ξ2,...,ξE, wherein the specific implementation process is shown in the steps (4.1) to (4.5).
Step (4.1): initializing i=1.
Step (4.2): will beMiddle and column vectorsColumn vectors with square distances less than delta between are recorded asWherein N i representsThe number of column vectors satisfying the condition ζ (i, j) < δ.
Step (4.3): calculating column vectors according to the formulaNormal vector f i:
In the above-mentioned method, the step of, Representation calculationAnd (3) withDistance between, b=1, 2,..n i.
Step (4.4): the column vector is calculated according to equation ④ as followsThe corresponding edge point index g i:
In the above, b e {1, 2.. The number N i},θb is a binary number, and the value rule is as follows:
Step (4.5): judging whether the condition i is less than n; if yes, after setting i=i+1, returning to the step (4.2); if not, the edge point indexes g 1,g2,...,gn are arranged in descending order according to the numerical value, and the column vectors corresponding to the maximum E edge point indexes are sequentially recorded as edge point vectors xi 1,ξ2,...,ξE.
Step (5): for new matrixColumn vectors in (a)And (3) performing cluster analysis to obtain C cluster center point vectors h 1,h2,...,hC.
It should be noted that, the plurality of clustering algorithms that can be used for performing the clustering analysis in the step (5) above are not limited to the k-means clustering (k-means clustering) algorithm adopted in the specific embodiment.
Step (6): respectively sequentially arranging column vectorsAs an input vector, the hidden layer neuron output vector z 1,z2,...,zn of the reduced core principal component network is calculated according to the formula ④ as follows:
where z i∈R(E+C)×1 represents the i-th hidden layer neuron output vector, i E {1,2,., n }, R (E+C)×1 represents the (E+C). Times.1-dimensional real number vector,
Step (7): and combining Z 1,z2,...,zn into a hidden layer output matrix Z= [ Z 1,z2,...,zn ] with (E+C) multiplied by n dimensions, and calculating according to a formula B 1=(ZTZ)-1ZYT to obtain a first output coefficient matrix B 1 of the reduced core principal component network.
Step (8): computing a matrixAfter E+C eigenvalues eta 1≥η2≥...≥ηE+C and corresponding eigenvectors beta 1,β2,...,βE+C respectively, recording the number of eigenvalues which are not smaller than eta 0 in eta 1,η2,...,ηE+C as d; where η 0 denotes the average of the e+c eigenvalues, η 0=(η1+η2+...+ηE+C)/(e+c), and the length of the eigenvector β 1,β2,...,βE+C is 1.
Step (9): after β 1,β2,...,βd is combined into the second output layer coefficient matrix B 2=[β1,β2,...,βd of the reduced core principal component network, a first monitor index vector Q 1 and a second monitor index vector Q 2 are calculated according to the following formula ⑤, respectively:
In the above equation, diag { } represents an operation of converting matrix diagonal elements in curly brackets into column vectors, AndThe inverse matrices of Λ 1=B1 TZZTB1 and Λ 2=B2 TZZTB2 are represented, respectively.
Step (10): after recording the maximum value elements in Q 1 and Q 2 as the control upper limits Q 1 and Q 2, respectively, the off-line modeling process is ended and step (11) is performed.
Step (11): 11 data of the wind power generation equipment at the latest sampling moment are collected by utilizing a data collection system of the wind power generator, a sample data vector x epsilon R 11×1 is formed, and whether the first data in x is between the cut-in wind speed and the cut-out wind speed is judged; if yes, go to step (12); if not, the wind driven generator is in an offline state, and the step (11) is repeated to continuously use the sample data vector at the latest sampling moment to monitor the state; wherein, 11 data in x are sequentially arranged according to the sequence in the step (1).
Step (12): performing the same normalization processing as in the step (2) on the elements of each row in x to obtain data vectors
Step (13): will beWhen an input vector is made, the hidden layer neuron output vector z ε R (E+C)×1 is calculated using the following equation ⑥:
step (14): according to the formula AndAfter the monitoring indexes J 1 and J 2 are obtained through calculation, whether the conditions are met or not is judged: j 1≤q1 and J 2≤q2; if yes, the wind power generation equipment at the current sampling moment operates normally, and the step (11) is returned; if not, go to step (15).
Step (15): returning to the step (11), and continuously using the sample data vector at the latest sampling moment to monitor the state of the wind power generation equipment; if all the monitoring indexes of the 6 continuous latest sampling moments do not meet the judging conditions in the step (14), triggering an abnormal working alarm and timely arranging personnel for maintenance; otherwise, the wind power generation equipment normally operates, and the step (11) is returned to continue to monitor the state of the wind power generation equipment.
The main advantages of the process according to the invention compared to the conventional process are detailed below.
As can be seen from the above step (13), the online calculation of the method of the present invention mainly derives from the formula ⑥, i.e. a vector containing E+C data is calculated. While the traditional kernel principal component analysis algorithm needs to be aimed at when implementing online monitoringA column vector containing n data is calculated as follows:
Since n is equal to the total number of sample data vectors, and E and C represent the number of edge feature points and cluster center points, respectively, e+c is typically much smaller than n. Therefore, the online calculation amount of the method is greatly reduced.
In addition, the method calculates corresponding outputs through the two output layer coefficient matrixes respectively, and compared with the traditional kernel principal component analysis algorithm which has only one output, the method has the advantage that the diversity of the characteristic extraction is optimally ensured. Finally, the present description will further explain the advantages of the method according to the invention by means of a specific implementation procedure.
Drawings
FIG. 1 is a schematic flow chart of the method of the present invention.
Fig. 2 is a schematic diagram of a network structure of a reduced core principal component involved in the method of the present invention.
Detailed Description
The invention will be described in detail below with reference to the drawings and the detailed description.
The invention discloses a method for monitoring the state of wind power generation equipment based on a reduced core principal component network, and a specific implementation mode of the method is described below by combining an implementation flow diagram shown in FIG. 1.
Step (1): after the data which can be measured in real time by the wind power generation equipment are determined, the data are collected according to fixed sampling time intervals in the normal running state of the wind power generation equipment.
Step (2): n sample data vectors X 1,x2,...,xn with wind speed between 3 m/s and 25 m/s are formed into matrix X= [ X 1,x2,...,xn ], and each row vector in X epsilon R 11×n is normalized to obtain new matrix
Step (3): for new matrixAfter the analysis of the principal components of the kernel, the kernel parameters delta and the kernel score matrix Y epsilon R D×n are reserved, and the specific implementation process is shown in the steps (3.1) to (3.5).
Step (4): for new matrixColumn vectors in (a)And (3) carrying out edge characteristic point analysis so as to obtain E edge point vectors xi 1,ξ2,...,ξE, wherein the specific implementation process is shown in the steps (4.1) to (4.5).
Step (5): for new matrixColumn vectors in (a)And (3) performing cluster analysis to obtain C cluster center point vectors h 1,h2,...,hC.
In this embodiment, step (5) uses a k-means clustering (k-means clustering) algorithm. It should be noted that, the implementation of step (5) of the method of the present invention is not limited to the use of k-means clustering algorithm, as long as the column vectors can be implementedA cluster analysis algorithm that clusters into C clusters can be used.
Step (6): respectively sequentially arranging column vectorsAs an input vector, the hidden layer neuron output vector z 1,z2,...,zn of the reduced core principal component network is calculated according to the aforementioned formula ④.
Step (7): and combining Z 1,z2,...,zn into a hidden layer output matrix Z= [ Z 1,z2,...,zn ] with (E+C) multiplied by n dimensions, and calculating according to a formula B 1=(ZTZ)-1ZYT to obtain a first output coefficient matrix B 1 of the reduced core principal component network.
Step (8): computing a matrixAfter E+C eigenvalues eta 1≥η2≥...≥ηE+C and corresponding eigenvectors beta 1,β2,...,βE+C respectively, recording the number of eigenvalues which are not smaller than eta 0 in eta 1,η2,...,ηE+C as d;
Step (9): after β 1,β2,...,βd is combined into the second output layer coefficient matrix B 2=[β1,β2,...,βd of the reduced core principal component network, a first monitor index vector Q 1 and a second monitor index vector Q 2 are calculated according to the foregoing formula ⑤, respectively.
Through the two output coefficient matrices B 1 and B 2 in the step (7) and the step (9), a reduced core principal component network structure as shown in fig. 2 can be built. It can be seen from fig. 2 that the number of hidden layer neurons of the reduced core principal component network involved in the method of the present invention is equal to e+c, while the output layer neurons are divided into two groups, using B 1 and B 2 as their corresponding coefficient matrices, respectively.
Step (10): after recording the maximum value elements in Q 1 and Q 2 as the control upper limits Q 1 and Q 2, respectively, the off-line modeling process is ended and step (11) is performed.
Step (11): collecting a sample data vector x epsilon R 11×1 at the latest sampling moment by using a data collecting system of the wind driven generator, and judging whether first data in x is between cut-in wind speed and cut-out wind speed; if yes, go to step (12); if not, the wind driven generator is in an offline state, and the step (11) is repeated to continuously use the sample data vector at the latest sampling moment to monitor the state; wherein, 11 data in x are sequentially arranged according to the sequence in the step (1).
Step (12): performing the same normalization processing as in the step (2) on the elements of each row in x to obtain data vectors
Step (13): will beAs an input vector, the hidden layer neuron output vector z e R (E+C)×1 is calculated using the above formula ⑥.
Step (14): according to the formulaAndAfter the monitoring indexes theta 1 and theta 2 are obtained through calculation, judging whether the conditions are met or not: θ 1≤q1 and θ 2≤q2; if yes, the wind power generation equipment at the current sampling moment operates normally, and the step (11) is returned; if not, go to step (15).
Step (15): returning to the step (11), and continuously using the sample data vector at the latest sampling moment to monitor the state of the wind power generation equipment; if the monitoring indexes of the continuous 6 sampling moments do not meet the judging conditions in the step (14), triggering an abnormal alarm and timely arranging personnel for maintenance; otherwise, the wind power generation equipment normally operates, and the step (11) is returned to continue to monitor the state of the wind power generation equipment.
Claims (1)
1. The wind power generation equipment state monitoring method based on the reduced core principal component network is characterized by comprising the following steps of:
step (1): after the data which can be measured in real time by the wind power generation equipment are determined, collecting the data according to a fixed sampling time interval in the normal running state of the wind power generation equipment; the 11 data which can be measured in real time at each sampling moment are specifically and sequentially: wind speed, rotor speed, generator speed, mechanical torque, generated power, blade pitch angle, blade azimuth, blade root moment, top horizontal axis acceleration, top longitudinal axis acceleration, and yaw error;
Step (2): n sample data vectors X 1,x2,…,xn with wind speed between the cut-in wind speed and the cut-out wind speed are formed into a matrix X= [ X 1,x2,…,xn ], and each row vector in X epsilon R 11×n is subjected to standardization processing to obtain a new matrix Wherein, 11 data in the ith sample data vector x i∈R11×1 are sequentially arranged according to the order in the step (1), i∈ {1,2, …, n }, R 11×n represents a real matrix of 11×n dimensions, R represents a real set, and R 11×1 represents a real vector of 11×1 dimensions;
Step (3): for new matrix After the analysis of the principal components of the core is implemented, the core parameters delta and the core score matrix Y epsilon R D×n are reserved, and the specific implementation process is shown in the steps (3.1) to (3.5);
step (3.1): according to the formula Calculating a new matrixIn the ith column vectorAnd the j-th column vectorThe square distance ζ (i, j) between; wherein j e {1,2, …, n }, the upper label T represents the transpose of the matrix or vector;
step (3.2): the kernel parameter delta is determined according to equation ① as follows:
Step (3.3): the ith row and jth column element K (i, j) in the kernel matrix K e R n×n is calculated according to the formula ② shown below:
in the above formula, e represents a natural constant;
Step (3.4): after n eigenvalues lambda 1≥λ2≥…≥λn of the kernel matrix K and corresponding eigenvectors alpha 1,α2,…,αn are calculated, recording the number of eigenvalues not smaller than lambda 0 in lambda 1,λ2,…,λn as D; wherein the length of the feature vector alpha 1,α2,…,αn is equal to 1, lambda 0=(λ1+λ2+…+λn)/n;
Step (3.5): d eigenvectors alpha 1,α2,…,αD are built into a feature transformation matrix A= [ alpha 1,α2,…,αD ], and then a kernel score matrix Y epsilon R D×n is obtained by calculation according to the formula Y=A T K;
step (4): for new matrix Column vectors in (a)Performing edge feature point analysis to obtain E edge point vectors xi 1,ξ2,…,ξE, wherein the specific implementation process is shown in the steps (4.1) to (4.5);
step (4.1): initializing i=1;
Step (4.2): will be Middle and column vectorsN i column vectors with square distance less than delta are recorded as
Step (4.3): the column vector is calculated according to equation ③ as followsNormal vector f i:
In the above-mentioned method, the step of, Representation calculationAnd (3) withDistance between, b=1, 2, …, N i;
step (4.4): the column vector is calculated according to equation ④ as follows The corresponding edge point index g i:
in the above formula, the value rule of θ b is as follows:
Step (4.5): judging whether i is smaller than n; if yes, after setting i=i+1, returning to the step (4.2); if not, the edge point indexes g 1,g2,…,gn are arranged in a descending order according to the numerical value, and the column vectors corresponding to the maximum E edge point indexes are sequentially recorded as edge point vectors xi 1,ξ2,…,ξE;
Step (5): for new matrix Column vectors in (a)Performing cluster analysis to obtain C cluster center point vectors h 1,h2,…,hC;
Step (6): respectively sequentially arranging column vectors As an input vector, the hidden layer neuron output vector z 1,z2,…,zn of the reduced core principal component network is calculated according to the formula ⑥ as follows:
in the above formula, z i∈R(E+C)×1 represents a group corresponding to I E {1,2, …, n }, R (E+C)×1 represents a real vector of (E+C). Times.1 dimension,
Step (7): combining Z 1,z2,…,zn into a hidden layer output matrix Z= [ Z 1,z2,…,zn ] with (E+C) multiplied by n dimensions, and calculating according to a formula B 1=(ZTZ)-1ZYT to obtain a first output coefficient matrix B 1 of the reduced core principal component network;
Step (8): computing a matrix After E+C eigenvalues eta 1≥η2≥…≥ηE+C and corresponding eigenvectors beta 1,β2,…,βE+C respectively, recording the number of eigenvalues which are not smaller than eta 0 in eta 1,η2,…,ηE+C as d; wherein the length of the feature vector beta 1,β2,…,βE+C is 1, eta 0=(η1+η2+…+ηE+C)/(E+C);
Step (9): after β 1,β2,…,βd is combined into the second output layer coefficient matrix B 2=[β1,β2,…,βd of the reduced core principal component network, a first monitor index vector Q 1 and a second monitor index vector Q 2 are calculated according to the following formula ⑦, respectively:
In the above equation, diag { } represents an operation of converting matrix diagonal elements in curly brackets into column vectors, AndInverse matrices representing Λ 1=B1 TZZTB1 and Λ 2=B2 TZZTB2, respectively;
Step (10): after the maximum value elements in Q 1 and Q 2 are respectively recorded as control upper limits Q 1 and Q 2, ending the offline modeling process and implementing the step (11);
Step (11): 11 pieces of data of the wind power generation equipment at the latest sampling moment are collected and form a sample data vector x epsilon R 11×1, and whether the first data in x is between the cut-in wind speed and the cut-out wind speed is judged; if yes, go to step (12); if not, the wind driven generator is in an offline state, and the step (11) is repeated to continuously use the sample data vector at the latest sampling moment to monitor the state; wherein, 11 data in x are sequentially arranged according to the sequence in the step (1);
step (12): performing the same normalization processing as in the step (2) on the data of each row in x to obtain a data vector
Step (13): will beWhen an input vector is made, the hidden layer neuron output vector z ε R (E+C)×1 is calculated using the following equation ⑧:
step (14): according to the formula AndAfter the monitoring indexes J 1 and J 2 are obtained through calculation, whether the conditions are met or not is judged: j 1≤q1 and J 2≤q2; if yes, the wind power generation equipment at the current sampling moment operates normally, and the step (11) is returned; if not, executing the step (15);
step (15): returning to the step (11), and continuously using the sample data vector at the latest sampling moment to monitor the state of the wind power generation equipment; if all the monitoring indexes of the 6 continuous latest sampling moments do not meet the judging conditions in the step (14), triggering an abnormal alarm and timely arranging personnel for maintenance; otherwise, the wind power generation equipment normally operates, and the step (11) is returned to continue to monitor the state of the wind power generation equipment.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111156322.1A CN113935408B (en) | 2021-09-22 | 2021-09-22 | Wind power generation equipment state monitoring method based on simplified core principal component network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111156322.1A CN113935408B (en) | 2021-09-22 | 2021-09-22 | Wind power generation equipment state monitoring method based on simplified core principal component network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113935408A CN113935408A (en) | 2022-01-14 |
CN113935408B true CN113935408B (en) | 2024-07-02 |
Family
ID=79277691
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111156322.1A Active CN113935408B (en) | 2021-09-22 | 2021-09-22 | Wind power generation equipment state monitoring method based on simplified core principal component network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113935408B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111914886A (en) * | 2020-06-13 | 2020-11-10 | 宁波大学 | Nonlinear chemical process monitoring method based on online brief kernel learning |
CN111914889A (en) * | 2020-06-13 | 2020-11-10 | 宁波大学 | Rectifying tower abnormal state identification method based on brief kernel principal component analysis |
CN111914471A (en) * | 2020-06-13 | 2020-11-10 | 宁波大学 | Rectification tower fault detection method based on rapid nuclear independent component analysis |
-
2021
- 2021-09-22 CN CN202111156322.1A patent/CN113935408B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111914886A (en) * | 2020-06-13 | 2020-11-10 | 宁波大学 | Nonlinear chemical process monitoring method based on online brief kernel learning |
CN111914889A (en) * | 2020-06-13 | 2020-11-10 | 宁波大学 | Rectifying tower abnormal state identification method based on brief kernel principal component analysis |
CN111914471A (en) * | 2020-06-13 | 2020-11-10 | 宁波大学 | Rectification tower fault detection method based on rapid nuclear independent component analysis |
Also Published As
Publication number | Publication date |
---|---|
CN113935408A (en) | 2022-01-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Tao et al. | Wind turbine blade icing diagnosis using hybrid features and Stacked-XGBoost algorithm | |
CN104978608B (en) | A kind of wind electric powder prediction device and prediction technique | |
Qian et al. | Data-driven condition monitoring approaches to improving power output of wind turbines | |
CN101858312B (en) | System and method thereof for evaluating real-time running state of wind generating set | |
CN108680358A (en) | A kind of Wind turbines failure prediction method based on bearing temperature model | |
CN109063276B (en) | Wind power plant dynamic equivalent modeling method suitable for long-time domain wind speed fluctuation | |
CN110889535B (en) | Multi-point wind speed prediction method in wind power plant based on convolution cyclic neural network | |
CN105069236B (en) | Consider the broad sense load joint probability modeling method of wind power plant node space correlation | |
CN109523084A (en) | A kind of ultrashort-term wind power prediction method based on pivot analysis and machine learning | |
CN111832812A (en) | Wind power short-term prediction method based on deep learning | |
CN111091236A (en) | Multi-classification deep learning short-term wind power prediction method classified according to pitch angles | |
CN106762452A (en) | Fan master control system fault diagnosis and on-line monitoring method based on data-driven | |
CN114215706A (en) | Wind turbine generator blade cracking fault early warning method and device | |
CN110544003A (en) | temperature prediction-based wind power plant wind turbine generator set frequency converter state evaluation method | |
CN115221785A (en) | Neural network prediction model construction method, prediction method and system | |
Xiaolan et al. | One-month ahead prediction of wind speed and output power based on EMD and LSSVM | |
CN113009338B (en) | Interturn short circuit fault diagnosis method for stator of offshore wind power variable pitch motor | |
CN113935408B (en) | Wind power generation equipment state monitoring method based on simplified core principal component network | |
CN112986821A (en) | Fault diagnosis method for broken blade of variable pitch motor rotor of offshore wind turbine generator | |
Pang et al. | Multi kernel fusion convolutional neural network for wind turbine fault diagnosis | |
CN113887705B (en) | Photovoltaic panel running state monitoring method based on sparse RBF neural network | |
CN113946608B (en) | Wind generating set running state monitoring method based on online RBF neural network | |
CN111539577A (en) | Short-term wind power prediction method based on wind speed change rate and Gaussian process regression | |
Qian et al. | A weighted kNN fault detection based on multistep index and dynamic neighborhood scale under complex working conditions | |
Li | Research on New Energy Power Generation Power Prediction Method Based on Machine Learning |
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 |