CN112580709B - Self-adaptive state evaluation method for offshore wind turbine generator - Google Patents
Self-adaptive state evaluation method for offshore wind turbine generator Download PDFInfo
- Publication number
- CN112580709B CN112580709B CN202011464899.4A CN202011464899A CN112580709B CN 112580709 B CN112580709 B CN 112580709B CN 202011464899 A CN202011464899 A CN 202011464899A CN 112580709 B CN112580709 B CN 112580709B
- Authority
- CN
- China
- Prior art keywords
- state
- data
- dictionary
- real
- time
- 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
- 238000011156 evaluation Methods 0.000 title claims abstract description 32
- 238000000034 method Methods 0.000 claims abstract description 27
- 230000003044 adaptive effect Effects 0.000 claims description 23
- 239000013598 vector Substances 0.000 claims description 23
- 239000011159 matrix material Substances 0.000 claims description 16
- 230000009467 reduction Effects 0.000 claims description 8
- 238000005315 distribution function Methods 0.000 claims description 7
- 238000012423 maintenance Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 5
- 238000001514 detection method Methods 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 3
- 238000013178 mathematical model Methods 0.000 claims description 2
- 238000010248 power generation Methods 0.000 abstract 1
- 238000004364 calculation method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000013210 evaluation model Methods 0.000 description 5
- 230000002159 abnormal effect Effects 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000000556 factor analysis Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/28—Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
-
- 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
- G06F18/2136—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on sparsity criteria, e.g. with an overcomplete basis
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Wind Motors (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention relates to an offshore wind power generation unit self-adaptive state evaluation method, which comprises the following steps: 1: dynamically dividing historical state data through related data and information to obtain a state data set of the unit in a typical state space; 2: learning an initial dictionary in each state of the unit by adopting a dictionary learning method, and storing the initial dictionary as an important parameter of the model; 3: judging the information difference degree between the real-time state data and the historical data set by utilizing the KL divergence, and sparsely classifying the real-time data in each initial dictionary when the information difference degree is smaller than an information difference threshold value, and evaluating to obtain the current state of the unit; 4: when obvious information difference is generated between the real-time state data and each historical data set, the real-time data is used as an increment set in a state space, dictionary atoms are added on the basis of an initial dictionary to obtain a learned increment dictionary, the real-time data is subjected to sparse classification on the increment dictionary, and the current state of the unit is obtained through evaluation. Compared with the prior art, the method has the advantages of self-adaptive state evaluation and the like.
Description
Technical Field
The invention relates to the technical field of state evaluation of offshore wind turbines, in particular to an adaptive state evaluation method of an offshore wind turbine.
Background
Offshore wind power is one of the most growing renewable energy sources in power systems. By the end of 2019, the installed capacity of global offshore wind power exceeds 29.1GW, wherein the newly added installed capacity is more than 6.1GW. Meanwhile, influenced by marine environment, the offshore wind turbine generator has the advantages of variable operating conditions, poor accessibility, high equipment failure rate and various failure reasons. Under the offshore time-varying operation condition, the wind turbine state evaluation method also needs to track the wind turbine for self-adaptive adjustment, and the accurate evaluation of the current state of the wind turbine in combination with state data is an important challenge for the state evaluation of the offshore wind turbine.
The existing offshore wind turbine state evaluation research mainly has the following three problems: 1) The method is characterized in that independent equivalent analytical models are respectively established from the electrical coupling among the unit components, and the change of mechanical parameters is established from the aspect of component failure mechanism to carry out judgment. 2) The state estimation model is idealized. The state space data of the default unit is unchanged, only historical state data are used for training an evaluation model, and the method is difficult to cover various states of the unit in terms of the whole life cycle. 3) Because the accuracy of the unit state evaluation model is greatly influenced by the number of training data, the training is concentrated, and the number of the state data of the unit is unbalanced, a machine learning method is adopted to cause certain tendency to the recognition result.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provide an offshore wind turbine adaptive state evaluation method, and the method mainly aims to track and match the time-varying state of a turbine set, supplement the state space data of the turbine set and pursue the adaptive identification of the turbine set on the typical state in the whole life cycle on the basis of ensuring that the same faults appearing in the history are accurately identified.
The purpose of the invention can be realized by the following technical scheme:
an adaptive state evaluation method for an offshore wind turbine generator comprises the following steps:
step 1: dynamically dividing historical state data through state monitoring data and field maintenance log information of an offshore wind turbine generator set to obtain a state data set of the generator set in a typical state space;
step 2: learning an initial dictionary in each state of the unit by adopting a dictionary learning method, and storing the initial dictionary as an important parameter of a model;
and step 3: judging the information difference degree between the real-time state data and the historical data set by utilizing the KL divergence, and sparsely classifying the real-time data in each initial dictionary to evaluate to obtain the current state of the unit when the information difference threshold is not met;
and 4, step 4: and when obvious information difference is generated between the real-time state data and each historical data set, taking the real-time data as an increment set in a state space, adding dictionary atoms on the basis of an initial dictionary to obtain a learned increment dictionary, carrying out sparse classification on the real-time data on the increment dictionary to evaluate and obtain the current state of the unit, and updating the dictionary in the step 2 and the data in the state space.
Further, the step 1 specifically includes: through offshore wind turbine generator state detection data and field maintenance log information, historical state data are dynamically divided through balanced iterative protocol clustering, and a state data set of the generator set in a typical state space is obtained, wherein sample data in the state data set of the generator set in the typical state space is linear and vector, and the description formula is as follows:
where, Σ l Is the sample data linear sum vector, x n Is the Nth machine set state data sample with dimension D.
Further, the sample data in the state data set in the unit typical state space is linear and vector, and the description formula is as follows:
in the formula (E) s Is the sample data linear sum vector, x no Is the value of the o-th state parameter in the n-th sample.
Further, the calculation formula of the distances between the clusters in the balanced iterative reduction clustering is as follows:
wherein d is the spacing between clusters, N p And N q Number of unit state data samples, Σ, for clusters p and q, respectively sp Sum-sigma sq Sample data linearity and vector, Σ, for cluster p and cluster q, respectively lp Sum-sigma lp Sample data linearity and vector for cluster p and cluster q, respectively.
Further, the cluster radius in the balanced iterative reduction clustering is calculated by the following formula:
wherein R is the cluster radius.
Further, the cluster centroid in the balanced iterative reduction clustering has a calculation formula as follows:
where x is the cluster centroid.
Further, the step 2 specifically includes: on the basis of dynamic division of the unit state space, the dictionary of each type of state sample is learned in a self-adaptive mode, the unit state characteristics are extracted, and the corresponding description formula is as follows:
e (i) =||z-D (i) x (i) || 2
in the formula, Y (i) Set i-th state data, D (i) And X (i) Respectively, the ith dictionary and the sparse matrix, z is real-time state data, x (i) Column vectors corresponding to the dictionary in the sparse matrix, rho is a value of i when the minimum value is reached in the set, lambda is a balance parameter of the relation between a balance norm and a sparse representation coefficient, and e (1) ,e (2) ,...,e (I) For errors of other respective reconstructed states, e (i) Is the error of the i-th reconstructed state.
Further, in step 3, the KL divergence is used to determine the information difference between the real-time status data and the historical data set, and the mathematical model description formula corresponding thereto is as follows:
in the formula, C now For observing the vector, P (w) and Q (w) respectively have the probability that the random variable value of historical data distribution and real-time data distribution is w, G is an information difference function, and L is a distribution functionNumber, D KL (P | Q) is the KL divergence from distribution P to distribution Q.
Further, in step 3, when the information difference threshold is not satisfied, performing sparse classification on the real-time data in each initial dictionary, and evaluating to obtain the current state of the unit, wherein a corresponding description formula is as follows:
in the formula D (j+1) As a new incremental dictionary, X (j+1) As a new sparse matrix, D (j) And X (j) Respectively, jth dictionary and sparse matrix, Y (j+1) The method comprises the steps that a new unit state data set is obtained, lambda is a balance parameter of a balance norm and sparse representation coefficient relation, T is an information difference degree threshold value, L is a distribution function, z is real-time state data, and e' i =e′ 1 ,e' 2 ,...,e' K And performing incremental self-adaptive learning on each model to obtain the representation error between the real-time data and the reconstructed data of the i state.
Further, when an obvious information difference is generated between the real-time state data and each historical data set in the step 4, the real-time data is used as an increment set in a state space, dictionary atoms are added on the basis of an initial dictionary to obtain a learned increment dictionary, the real-time data is subjected to sparse classification on the increment dictionary, and the current state of the unit is obtained through evaluation, wherein a corresponding description formula is as follows:
in the formula, D (j+1) As a new incremental dictionary, X (j+1) As a new sparse matrix, Y (j+1) The method comprises the steps that a new unit state data set is obtained, lambda is a balance parameter of a balance norm and sparse representation coefficient relation, T is an information difference degree threshold value, L is a distribution function, z is real-time state data, and e' i =e′ 1 ,e' 2 ,...,e' K And performing incremental self-adaptive learning on each model to obtain the representation error between the real-time data and the reconstructed data of the i state.
Compared with the prior art, the invention has the following advantages:
(1) Aiming at the problem of incentive diversity of typical unit faults caused in the whole life cycle of an offshore wind turbine generator, the invention innovatively provides an offshore wind turbine generator adaptive evaluation method based on incremental dictionary learning, information difference between real-time state data and historical data is obtained through KL divergence quantification, and for the real-time data lower than an information difference threshold, an initial dictionary, a sparse matrix and an evaluation unit initial state are solved by adopting a dictionary learning method; otherwise, the real-time data is used as an incremental data set of the adaptive evaluation method in the state space, an offshore wind turbine generator adaptive evaluation model is established, two layers of nesting are carried out, model parameters are updated in an iterative mode, and therefore adaptive evaluation of the offshore wind turbine time-varying state is achieved.
(2) The invention provides an offshore wind turbine state self-adaptive evaluation method aiming at the problems of complex operating environment and various fault types of an offshore wind turbine and combining the time-varying characteristic of high-dimensional nonlinear state data of the wind turbine. According to the method, KL divergence quantization state data information difference is adopted, and a fan state evaluation self-learning model is constructed on the basis of an incremental dictionary learning method, so that self-adaptive evaluation of the offshore fan time-varying state is achieved.
Drawings
FIG. 1 is a schematic flow chart illustrating a method for adaptive state estimation of an offshore wind turbine in an embodiment of the present invention;
fig. 2 is a schematic diagram of state space division and state time evolution of an offshore wind turbine in an embodiment of the present invention, where fig. 2 (a) is a schematic diagram of projection of a turbine state on a two-dimensional space, and fig. 2 (b) is a state evolution process of the turbine state on a time sequence;
FIG. 3 is a graph of singular value changes in an embodiment of the present invention;
FIG. 4 is a diagram illustrating a state evaluation result of an offshore wind turbine in an embodiment of the present invention;
FIG. 5 is a graph of information dissimilarity distribution according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of MES and accumulated runtime as a function of a set of increments, in accordance with an embodiment of the present invention;
FIG. 7 is a diagram illustrating the variation of the optimal sparsity with the number of incremental samples according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, shall fall within the scope of protection of the present invention.
DETAILED DESCRIPTION OF EMBODIMENT (S) OF INVENTION
Fig. 1 shows a flow chart of an adaptive state evaluation method for an offshore wind turbine generator according to the present invention, which includes the following steps:
(1) Dynamically dividing historical state data through information such as state detection data (such as SCADA) and field maintenance logs of an offshore wind turbine generator set to obtain a state data set of the generator set in a typical state space;
(2) Learning an initial dictionary in each state of the unit by adopting a dictionary learning method, and storing the initial dictionary as an important parameter of the model;
(3) And judging the information difference degree between the real-time state data and the historical data set by utilizing the KL divergence, and when the information difference threshold is not met, sparsely classifying the real-time data in each initial dictionary to quickly evaluate the current state of the unit.
(4) When obvious information difference is generated between the real-time state data and each historical data set, the real-time data is used as an increment set in a state space, dictionary atoms are added on the basis of an initial dictionary to obtain a learned increment dictionary, and the real-time data is subjected to sparse classification on the increment dictionary to obtain a unit state. And (3) updating the dictionary in the step (2) and the data in the state space.
Aiming at the four steps, the method is concretely realized as follows:
(1) Dynamic division of unit state space based on balanced iterative protocol clustering
Firstly, dividing the state space data of the unit by Balanced Iterative reduction and Clustering using hierarchy (BIRCH), as shown in the following formula:
where, Σ l Is the sample data linear sum vector, x n And D is the Nth machine set state data sample.
The sample data is linear and vector, and the description formula is as follows:
in the formula (E) s Is the sample data linear sum vector, x no Is the o-th shape in the n-th sampleAnd (4) an attitude parameter value.
The distance between clusters in the iterative reduction clustering is balanced, and the calculation formula is as follows:
wherein d is the spacing between clusters, N p And N q Number of unit state data samples, Σ, for clusters p and q, respectively sp Sum-sigma sq Sample data linearity and vector, Σ, for cluster p and cluster q, respectively lp Sum-sigma lp Sample data for cluster p and cluster q, respectively, are linear and vector.
Balancing the cluster radius in the iterative specification cluster, wherein the calculation formula is as follows:
wherein R is the cluster radius.
Balancing cluster centroids in iterative specification clustering, wherein the calculation formula is as follows:
where x is the cluster centroid.
(2) State feature extraction method for offshore wind turbine generator based on dictionary learning
On the basis of dynamic division of the unit state space, the dictionary of each type of state sample is learned in a self-adaptive mode, and the unit state features are extracted.
e (i) =||z-D (i) x (i) || 2
In the formula, Y (i) Set i-th state data, D (i) And X (i) Respectively, the ith dictionary and the sparse matrix, z is real-time state data, x (i) Column vectors corresponding to the dictionary in the sparse matrix, rho is the value of i when the minimum value is reached in the set, lambda is the balance parameter of the relation between the balance norm and the sparse representation coefficient, e (1) ,e (2) ,...,e (I) For errors of other respective reconstructed states, e (i) Is the error of the i-th reconstructed state.
(3) KL divergence based state increment set decision
The judgment of the new information implied in the real-time data is the first step of the real-time evaluation stage of the unit and is also the basis of subsequent incremental learning, and the KL divergence is utilized to construct the information difference degree of the real-time data and the historical state dataThe model is as follows.
In the formula, C now In order to observe the vector, the system is designed to,p (w) and Q (w) respectively have the probability that the random variable value of historical data distribution and real-time data distribution is w, G is an information difference function, L is a distribution function, and D KL (P | Q) is the KL divergence from distribution P to distribution Q.
(4) Offshore wind turbine adaptive state evaluation model based on incremental dictionary learning
And when the real-time state data containing the new information is identified, the current state of the unit is evaluated in a self-adaptive manner, and the data and the initial dictionary in the state space are updated.
In the formula, D (j+1) As a new incremental dictionary, X (j+1) As a new sparse matrix, D (j) And X (j) Respectively, jth dictionary and sparse matrix, Y (j+1) The method comprises the steps that a new unit state data set is obtained, lambda is a balance parameter of a balance norm and sparse representation coefficient relation, T is an information difference degree threshold value, L is a distribution function, z is real-time state data, and e' i =e′ 1 ,e' 2 ,...,e' K And performing incremental self-adaptive learning on each model to obtain the representation error between the real-time data and the reconstructed data of the i state.
Wherein abnormal stateAnd fault stateIn thatIn the above-mentioned method, the state data of the first cause of machine set fault is recorded asSelect initial dictionary D (0) And sparse matrix X 0 Data setHas learned the dictionary D (j) New incremental dictionary D (j+1) The piecewise function F characterizes the adaptive state estimation model.
Examples of specific applications
The case selects SACDA system and unit history maintenance log data of a 3MW offshore wind power unit between 2011 5 month 1 day and 2018 7 month 27 day. A main frequency converter and a wind wheel with high failure occurrence frequency and a generator with high failure shutdown loss cost are selected as typical components of the unit. Each component is divided into 7 states: the wind turbine generator fault state control method comprises a normal state (1-TN), a main frequency converter abnormal state (2-VA), a main frequency converter fault state (3-VF), a generator abnormal state (4-GA), a generator fault state (5-GF), a wind turbine abnormal state (6-WA) and a wind turbine fault state (7-WF). Constant parameters which are not changed in the running process of the fan are removed, and 84-dimensional state characteristics are adopted to reflect changes of all states of the unit.
In order to realize the dynamic division of the unit state space, the She Pingheng factor λ and the branch balance factor β are both 50, the maximum sample radius threshold T of the leaf node CF is 0.4, the number of categories K is determined by the number of unit states in a unit monitoring period, and is set to 3. The projection of the crew state on the two-dimensional space is shown in fig. 2 (a), and the state evolution process on the time sequence is shown in fig. 2 (b).
(1) Extraction of initial state features of offshore wind turbine generator
The number of dictionary atoms in the state space can be determined according to the information contribution rate of the singular value. The change of the singular value and the carried information under the normal state is obtained by using a dictionary learning method as shown in fig. 3, and the method can cover the original data information by 100% when the first 73 singular values are obtained, namely the dimension of the normal state dictionary is 73. Meanwhile, the number of dictionary atoms and the singular value information contribution rate of the remaining obtained states are shown in table 1.
TABLE 1 number of atoms of each State dictionary at 100% singular value information contribution
(2) Initial state assessment of offshore wind turbine
The sample is SCADA data in a state space, the training set selects first fault data except a normal data set, and the testing set is fault data of the same type which occurs again. The status recognition is shown in table 2 below.
TABLE 2 MSE and State recognition of offshore wind turbine State dictionary
Tab.2 Fault state recognition rate by dictionary learning
Fig. 4 shows that, after data of 2016 year 3, month 3, and 17 to 2016 year 3, month 3, and 4, and 5 are selected from the offshore wind turbine generator set, the validity of the proposed evaluation method is tested and verified.
(3) Adaptive state assessment of offshore wind turbine
Fig. 5 shows a state-to-state spatial data information degree difference matrix trained using historical data, and a threshold value of the information difference degree T is set to 0.65.
On the basis of the initial state feature learning of the unit, the phase current grounding (PD) is used as an increment set of the fault state (3-VF) of the main frequency converter, the machine side Current Abnormity (CA) and the variable pitch battery charging fault (PF) are respectively used as increment sets of the states of the generator fault (5-GF) and the wind wheel fault (7-WF), and the unit identification rate after the increment learning is shown in a table 3.
TABLE 3 Unit State identification accuracy based on incremental dictionary learning
(5) Adaptive state assessment result influence factor analysis
In order to analyze the influence of the increase of the state quantity sets containing new information on the evaluation result, the feedback loss and the pre-charging fault of the 24V power supply of the CC301 cabinet are respectively used as the other two increase sets of the state space of the main frequency converter, the data and the test set of the rest state spaces are not changed, and the comparison and analysis of the identification result and the result of the identification result and the original data are shown in table 4.
TABLE 4 incremental set number increase impact on recognition results
Fig. 6 shows that, as the incremental information data increases, the accumulation operation time increases and positively correlates with how many samples are in the incremental set. Meanwhile, the root Mean Square Error (MSE) of the incremental data set and the reconstructed data after learning by the incremental dictionary tends to increase with the increase of the incremental set.
Fig. 7 shows that the number of samples is increased in an equal proportion on the basis of the reinitialization data set, and the optimal sparsity parameter in the fan adaptive evaluation model gradually converges to a constant value.
The case shows that the method provided by the patent is effective and feasible, and can provide reference for evaluating the offshore wind power state.
While the invention has been described with reference to specific embodiments, the invention is not limited thereto, and various equivalent modifications and substitutions can be easily made by those skilled in the art within the technical scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (7)
1. An adaptive state evaluation method for an offshore wind turbine is characterized by comprising the following steps:
step 1: dynamically dividing historical state data through state monitoring data and field maintenance log information of an offshore wind turbine generator set to obtain a state data set of the generator set in a typical state space;
step 2: learning an initial dictionary in each state of the unit by adopting a dictionary learning method, and storing the initial dictionary as an important parameter of the model;
and step 3: judging the information difference degree between the real-time state data and the historical data set by utilizing the KL divergence, and sparsely classifying the real-time data in each initial dictionary to evaluate to obtain the current state of the unit when the information difference threshold is not met;
and 4, step 4: when obvious information difference is generated between the real-time state data and each historical data set, taking the real-time data as an increment set in a state space, adding dictionary atoms on the basis of an initial dictionary to obtain a learned increment dictionary, carrying out sparse classification on the real-time data on the increment dictionary to evaluate and obtain the current state of the unit, and updating the dictionary in the step 2 and the data in the state space;
in the step 3, the KL divergence is used for judging the information difference degree between the real-time state data and the historical data set, and the corresponding mathematical model description formula is as follows:
in the formula, C now For observing the vector, P (w) and Q (w) are the probability that the random variable value of historical data distribution and real-time data distribution is w, G is an information difference function, L is a distribution function, D is KL (P | Q) is the KL divergence from distribution P to distribution Q;
in step 3, when the information difference threshold is not satisfied, the real-time data is sparsely classified in each initial dictionary, and the current state of the unit is obtained through evaluation, wherein the corresponding description formula is as follows:
in the formula, D (j+1) As a new incremental dictionary, X (j+1) As a new sparse matrix, D (j) And X (j) Respectively, jth dictionary and sparse matrix, Y (j+1) Is a new unit state data set, lambda is a balance parameter of the relation between a balance norm and a sparse representation coefficient, T is an information difference threshold, L is a distribution function, z is real-time state data, e i '=e 1 ',e' 2 ,...,e' K Real-time data and i-state after incremental adaptive learning for each modelA representation error between the reconstructed data of (a);
in the step 4, when an obvious information difference is generated between the real-time state data and each historical data set, the real-time data is used as an increment set in a state space, dictionary atoms are added on the basis of an initial dictionary to obtain a learned increment dictionary, the real-time data is sparsely classified on the increment dictionary, and the current state of the unit is obtained by evaluation, wherein the corresponding description formula is as follows:
in the formula D (j+1) As a new incremental dictionary, X (j+1) As a new sparse matrix, Y (j+1) Is a new unit state data set, lambda is a balance parameter of the relation between a balance norm and a sparse representation coefficient, T is an information difference threshold, L is a distribution function, z is real-time state data, e i '=e 1 ',e' 2 ,...,e' K And performing incremental self-adaptive learning on each model to obtain the representation error between the real-time data and the reconstructed data of the i state.
2. The adaptive state evaluation method for the offshore wind turbine generator system according to claim 1, wherein the step 1 specifically comprises: through offshore wind turbine generator state detection data and field maintenance log information, historical state data are dynamically divided through balanced iterative protocol clustering, and a state data set of the generator set in a typical state space is obtained, wherein sample data in the state data set of the generator set in the typical state space is linear and vector, and the description formula is as follows:
in the formula, sigma l Is the sample data linear sum vector, x n And the dimension is D for the nth machine set state data sample.
3. The adaptive state evaluation method for the offshore wind turbine generator system according to claim 2, wherein the sample data in the state data set in the typical state space of the wind turbine generator system is linear and vector, and the description formula is as follows:
in the formula (E) s Is the sample data linear sum vector, x no Is the value of the o-th state parameter in the n-th sample.
4. The adaptive state estimation method for offshore wind turbines according to claim 2, wherein the cluster-to-cluster distance in the balanced iterative reduction clustering is calculated by the formula:
wherein d is the spacing between clusters, N p And N q Number of unit state data samples, Σ, for clusters p and q, respectively sp Sum-sigma sq Sample data linearity and vector, Σ, for cluster p and cluster q, respectively lp Sum sigma lp Sample data linearity and vector for cluster p and cluster q, respectively.
7. The adaptive state evaluation method for the offshore wind turbine generator system according to claim 1, wherein the step 2 specifically comprises: on the basis of dynamic division of the unit state space, the dictionary of each type of state sample is learned in a self-adaptive mode, the unit state characteristics are extracted, and the corresponding description formula is as follows:
e (i) =||z-D (i) x (i) || 2
in the formula, Y (i) Set i-th state data, D (i) And X (i) Are respectively the firsti dictionaries and sparse matrices, z real-time state data, x (i) Column vectors corresponding to the dictionary in the sparse matrix, rho is a value of i when the minimum value is reached in the set, lambda is a balance parameter of the relation between a balance norm and a sparse representation coefficient, and e (1) ,e (2) ,...,e (I) For errors of other respective reconstructed states, e (i) Is the error of the i-th reconstructed state.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011464899.4A CN112580709B (en) | 2020-12-14 | 2020-12-14 | Self-adaptive state evaluation method for offshore wind turbine generator |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011464899.4A CN112580709B (en) | 2020-12-14 | 2020-12-14 | Self-adaptive state evaluation method for offshore wind turbine generator |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112580709A CN112580709A (en) | 2021-03-30 |
CN112580709B true CN112580709B (en) | 2022-11-29 |
Family
ID=75131879
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011464899.4A Active CN112580709B (en) | 2020-12-14 | 2020-12-14 | Self-adaptive state evaluation method for offshore wind turbine generator |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112580709B (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107704990A (en) * | 2017-08-29 | 2018-02-16 | 中国矿业大学 | A kind of wind power prediction Real-time Error appraisal procedure based on dictionary learning algorithm |
CN111523710A (en) * | 2020-04-10 | 2020-08-11 | 三峡大学 | Power equipment temperature prediction method based on PSO-LSSVM online learning |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107368809B (en) * | 2017-07-20 | 2018-05-18 | 合肥工业大学 | A kind of bearing fault sorting technique based on rarefaction representation and dictionary learning |
CN111666982B (en) * | 2020-05-19 | 2023-04-18 | 上海核工程研究设计院股份有限公司 | Electromechanical equipment fault diagnosis method based on deep neural network |
-
2020
- 2020-12-14 CN CN202011464899.4A patent/CN112580709B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107704990A (en) * | 2017-08-29 | 2018-02-16 | 中国矿业大学 | A kind of wind power prediction Real-time Error appraisal procedure based on dictionary learning algorithm |
CN111523710A (en) * | 2020-04-10 | 2020-08-11 | 三峡大学 | Power equipment temperature prediction method based on PSO-LSSVM online learning |
Also Published As
Publication number | Publication date |
---|---|
CN112580709A (en) | 2021-03-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109842373B (en) | Photovoltaic array fault diagnosis method and device based on space-time distribution characteristics | |
CN111237134B (en) | Offshore double-fed wind driven generator fault diagnosis method based on GRA-LSTM-stacking model | |
CN111537219B (en) | Fan gearbox performance detection and health assessment method based on temperature parameters | |
CN110530650B (en) | Method for monitoring performance state of heavy-duty gas turbine based on generalized regression neural network and box diagram analysis | |
CN110879377B (en) | Metering device fault tracing method based on deep belief network | |
CN104915747A (en) | Electricity generation performance evaluation method of generator set and equipment thereof | |
CN110362045B (en) | Marine doubly-fed wind turbine generator fault discrimination method considering marine meteorological factors | |
CN111814849A (en) | DA-RNN-based wind turbine generator key component fault early warning method | |
CN110365053A (en) | Short-term wind power forecast method based on delay optimisation strategy | |
CN115578084A (en) | Wind turbine generator set frequency converter fault early warning method based on deep convolution self-encoder | |
CN115034493B (en) | Wind farm black start path optimizing method considering unit running state | |
CN112364446A (en) | Engine whole performance attenuation prediction method based on EC-RBELM algorithm | |
CN115146739A (en) | Power transformer fault diagnosis method based on stacked time series network | |
CN117473411A (en) | Bearing life prediction method based on improved transducer model | |
CN116928038A (en) | Fault early warning method and device based on main bearing temperature of offshore wind turbine generator | |
CN115526258A (en) | Power system transient stability evaluation method based on Spearman correlation coefficient feature extraction | |
CN118188342A (en) | Fan-oriented fault early warning and life prediction method | |
CN112734141B (en) | Diversified load interval prediction method and device | |
CN111062516A (en) | Fan output prediction method based on GMDH (Gaussian mixture distribution) multivariate processing | |
CN114037194A (en) | Hydroelectric power plant power generation load prediction system and method based on machine learning | |
CN112580709B (en) | Self-adaptive state evaluation method for offshore wind turbine generator | |
CN108536877B (en) | Method and system for evaluating vibration degree of wind generating set | |
CN116578908A (en) | Cross-domain fault diagnosis method, system, medium and equipment for wind power gear box | |
CN116151799A (en) | BP neural network-based distribution line multi-working-condition fault rate rapid assessment method | |
CN110889646A (en) | Method for checking correctness of operating parameters of generator set adapting to working condition change |
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 |