CN114781244A - Grouping and parameter optimization method in wind power plant - Google Patents
Grouping and parameter optimization method in wind power plant Download PDFInfo
- Publication number
- CN114781244A CN114781244A CN202210306444.2A CN202210306444A CN114781244A CN 114781244 A CN114781244 A CN 114781244A CN 202210306444 A CN202210306444 A CN 202210306444A CN 114781244 A CN114781244 A CN 114781244A
- Authority
- CN
- China
- Prior art keywords
- clustering
- model
- equivalent
- algorithm
- power plant
- 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.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000005457 optimization Methods 0.000 title claims abstract description 22
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 45
- 238000002156 mixing Methods 0.000 claims abstract description 24
- 230000009467 reduction Effects 0.000 claims abstract description 22
- 238000004088 simulation Methods 0.000 claims abstract description 16
- BVEXVMARIREFTJ-TXMWGMRISA-N methyl (2r)-3-[(2s,6r,8s,11r)-2-[(e,2r)-4-[(2s,2'r,4r,4as,6r,8ar)-4-hydroxy-2-[(1s,3s)-1-hydroxy-3-[(2s,3r,6s)-3-methyl-1,7-dioxaspiro[5.5]undecan-2-yl]butyl]-3-methylidenespiro[4a,7,8,8a-tetrahydro-4h-pyrano[3,2-b]pyran-6,5'-oxolane]-2'-yl]but-3-en-2-yl] Chemical compound O1[C@H](C[C@@](C)(O)C(=O)OC)CC[C@@H](O)[C@@]21C=C(C)C[C@@H]([C@H](C)\C=C\[C@@H]1O[C@@]3(O[C@H]4[C@H](O)C(=C)[C@@H]([C@@H](O)C[C@H](C)[C@@H]5[C@@H](CC[C@@]6(OCCCC6)O5)C)O[C@@H]4CC3)CC1)O2 BVEXVMARIREFTJ-TXMWGMRISA-N 0.000 claims abstract description 10
- 238000004364 calculation method Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 14
- 238000012549 training Methods 0.000 claims description 7
- 230000011218 segmentation Effects 0.000 claims description 6
- 150000001875 compounds Chemical class 0.000 claims description 2
- 230000008602 contraction Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims description 2
- 238000005192 partition Methods 0.000 claims 1
- 229910052799 carbon Inorganic materials 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 23
- 230000004044 response Effects 0.000 description 15
- 238000003064 k means clustering Methods 0.000 description 9
- 239000007787 solid Substances 0.000 description 6
- 230000004927 fusion Effects 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 4
- 230000001052 transient effect Effects 0.000 description 4
- 238000007637 random forest analysis Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000012217 deletion Methods 0.000 description 2
- 230000037430 deletion Effects 0.000 description 2
- 230000036039 immunity Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/06—Wind turbines or wind farms
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
With the proposal of a double-carbon target, the scale of a wind power plant is gradually increased, and the dynamic characteristic of the wind power plant has great influence on the stability of a novel power system. The invention provides a method for optimizing clustering and parameters in a wind power plant, which aims to improve equivalent precision and wide multi-working-condition applicability of a wind power plant simulation model. Firstly, performing dimensionality reduction selection on a grouping index by adopting XGboost-Blending; then, a clustering method based on DBSCAN-DTW is provided for cluster division; and finally, selecting a fan leading parameter, converting the equivalent parameter calculation problem into a multi-target nonlinear optimization model, and solving by adopting a multi-target equilibrium optimization algorithm MEOA based on an improved reference point. A simulation model is built based on matlab/simulink, and the accuracy and the multi-working-condition wide applicability of the equivalent model after clustering by using the method are verified.
Description
Technical Field
The patent relates to the problem of establishing a dynamic simulation model of a wind power plant, and provides a grouping and parameter optimization method in the wind power plant.
Background
With the proposal of the double-carbon target, the scale of the wind power plant is gradually increased, the dynamic characteristic of the wind power plant has great influence on the stability of a novel power system, and a simulation model which accurately reflects the dynamic characteristic of the wind power plant needs to be constructed. The invention provides a grouping and parameter optimizing method in a wind power plant. The K-means clustering is easily affected by noise data, dynamic response time is different due to different wind speeds of fans in a wind power plant, noise influence can be eliminated by density-based clustering of the DBSCAN algorithm, the DTW algorithm takes Euclidean distance between normalization paths as similarity indexes among the fans, and the two algorithms are combined to be beneficial to solving the problem.
Disclosure of Invention
The invention discloses a grouping and parameter optimizing method in a wind power plant.
The method comprises the following steps of 1: performing dimensionality reduction selection on the grouping index by adopting XGboost-Blending;
and 2, step: the clustering method based on DBSCAN-DTW is used for cluster division;
and 3, step 3: converting the equivalent parameter calculation problem into a multi-target nonlinear optimization model, and solving by adopting a multi-target equilibrium optimization algorithm MEOA;
and 4, step 4: a simulation model is built based on matlab/simulink, and the accuracy and the multi-working-condition wide applicability of the equivalent model after clustering by using the method are verified.
Optionally, the step 1: the XGboost-Blending is adopted to perform dimensionality reduction selection on the clustering index, and the method specifically comprises the following steps:
the index selection result of the XGboost can be explained, and the influence of different characteristics on the result can be clearly seen. XGBoost is an integration algorithm based on Classification And Regression Tree (CART) model. The objective function is expressed as:
wherein, L is a training error, namely a loss function about a predicted value and a true value, and represents the matching degree of the model to a training set; y isiIs the true value; omega (f)t) The model complexity is represented by a regular term defined by an L2 norm, and the more complex the model is, the larger the value of the regular term is; c is a constant term.
The loss function is:
to avoid overfitting and dimension disasters, dimension reduction selection needs to be carried out on the features. The wind driven generator characteristic extraction method comprises the following steps:
1) the wind driven generator sample x is input, and the loss function L is input.
2) And building a tree by using a greedy algorithm, learning a new function, and fitting the residual error of the last prediction.
3) L is iteratively trained so that the smaller the error, the better.
4) And defining a regular term by adopting an L2 norm, calculating the complexity, and dividing the tree into a structure s and a weight m.
5) After node iteration, an optimal value of the difference between the loss before node segmentation and the loss after node segmentation (namely, the gain of node splitting) is obtained, and the optimal value can be used for calculating the average gain of the features to represent the importance degree of the features, and then more important features are selected to realize dimension reduction:
in the formula (I), the compound is shown in the specification,representing the score of the left sub-tree,representing the score of the right sub-tree,representing the score obtained without segmentation and μ representing the complexity of the new node.
6) And repeating the steps until enough trees are generated, so that the predicted value is closest to the true value, and finishing the algorithm.
The Blending of the Blending model can overcome data crossing, and the Blending model and the XGboost meta-model not only can strengthen the learning effect, but also can not cause excessive redundancy of the whole model, so the invention adopts the Blending of the Blending model XGboost to carry out clustering index dimensionality reduction. And recording the G value of each feature when the feature is split, and finally dividing the total G value of the feature by the number of times that the feature is used for splitting the node to obtain the quantitative score of the contribution degree of the feature. And deleting the features one by one according to the sequence of the contribution degrees from low to high, clustering again, and traversing to obtain an index selection scheme corresponding to the clustering condition with the highest profile value.
Optionally, the step 2: the clustering method based on DBSCAN-DTW is proposed for cluster division, and specifically comprises the following steps:
the DTW is adopted in the DBSCAN to calculate the similarity of time sequence data, and the points on different sequences are aligned with the points through the extension and contraction of the sequences, so that the accumulated minimum distance between the points on two sequences with different lengths is calculated, as shown in a formula (4).
Assuming that the sequence-normalized path is R, and k represents the length of the path, the normalized path is R ═ (R)1,r2,…rk) The regular path distance function isThe defined warping path needs to satisfy certain constraints: 1) boundary property: the start and end points of the two sequences P and Q must correspond, i.e.R1=(1,1),Rk(m, n). 2) Monotonicity: the points on the sequences P and Q must be monotonic so that the two sequences do not intersect. 3) Continuity: the points in the sequence can only be matched with adjacent points, and can not be matched in a spanning way, i.e. 0 is less than or equal to i-i' is less than or equal to 1.
Similarity among the multidimensional time sequences is obtained through DTW, and the similarity needs to be substituted into a DBSCAN clustering algorithm for clustering. After two parameters, namely clustering radius Epsilon and minimum value MinPts of the number of samples in a category, in the DBSCAN algorithm are set, one point in the samples can be selected arbitrarily as a core point, all samples meeting the condition that the density can reach the density are found as a category, all points in the category are ensured to be in the Epsilon neighborhood, a sample set is set as E, and a point m is selected arbitrarily, as shown in formula (5).
Epsilon(m,M)={m∈E|d(m,M)≤Epsilon} (5)
Optionally, the step 3: converting the equivalent parameter calculation problem into a multi-target nonlinear optimization model, and solving by adopting a multi-target equilibrium optimization algorithm MEOA, which specifically comprises the following steps:
calculating parameters of a generator, a transformer and a shafting of the equivalent unit by adopting a capacity weighting method; calculating equivalent wind speed on the basis of the principle that the total wind energy input by the wind turbines in each cluster before and after equivalence is equal; and converting the trunk type topological structure into a radial topological structure according to an equal power loss method, and then calculating equivalent parameters of the current collection system.
Selecting equivalent precision of active power, reactive power and voltage to construct an objective function:
wherein:
and solving the multi-target nonlinear optimization model by adopting a multi-target equilibrium optimization algorithm MEOA based on the improved reference point.
Optionally, the step 4: a simulation model is built based on matlab/simulink, the accuracy and the multi-working-condition wide applicability of the equivalent model after clustering by using the method are verified, and the method specifically comprises the following steps:
the method comprises the steps of constructing a wind power plant model containing 16 DFIGs by using matlab/simulink, acquiring characteristic time sequence data by setting three-phase short circuit faults at the outlet of the wind power plant at a certain time period, realizing an algorithm by python, and training the model by adopting a Linux server.
Compared with the prior art, the technology has the following beneficial effects:
aiming at the problem of low equivalent precision of a wind power plant simulation model, the invention provides a method for carrying out clustering index dimension reduction by Blending XGboost, optimizing DBSCAN cluster partitioning by adopting DTW (dynamic time warping) and carrying out clustering result fusion.
(1) The DTW considers that the dynamic response time of the DFIG of the same model in the wind power plant is different under different wind speeds, and effectively solves the problem of data partial deletion of the actual wind power plant.
(2) DBSCAN avoids similar DFIG outliers based on density, and its noise immunity is also significantly enhanced.
(3) Blending is combined with XGboost to select grouping indexes, strong correlation among variables is eliminated, and the speed and accuracy of grouping are improved.
(4) The equivalent parameter calculation problem is converted into a multi-target nonlinear optimization model, and the MEOA is adopted for solving, so that the equivalent precision is improved, and the method has the advantages of multi-working-condition wide applicability.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a general flow chart of an embodiment of the method of the present invention;
FIG. 2 is a CART basic structure diagram according to the present invention;
FIG. 3 is a schematic view of a Blending fusion step according to the present invention;
FIG. 4 is a wind farm simulation of the present invention;
fig. 5 is a simulation diagram of the transient response of the active power of the fan 1 and the fan 2 according to the present invention;
fig. 6 is a similarity timing chart of the fan 1 and the fan 2 according to the present invention;
FIG. 7 is a diagram of the k-means clustering results before noise is added in the present invention;
FIG. 8 is a diagram of k-means clustering results after noise is added in the present invention;
FIG. 9 is a clustering diagram of the present invention;
FIG. 10 is a DBSCAN clustering flow chart according to the present invention;
FIG. 11 is a diagram of a DBSCAN clustering result before noise is added in the present invention;
FIG. 12 is a diagram of a DBSCAN clustering result after noise is added in the invention;
FIG. 13 is a flowchart of the python algorithm of the present invention;
FIG. 14 is a diagram of different wind speed disturbance scenarios according to the present invention;
FIG. 15 is a schematic view of a wind farm of the present invention;
FIG. 16 is a schematic view of a four-machine equivalent wind farm of the present invention;
FIG. 17 is a graph of voltage response at grid tie points;
fig. 18 is a graph of an active power response at a grid connection point;
fig. 19 is a graph of reactive power response at a grid connection point;
FIG. 20 is a schematic diagram of a five-machine equivalent wind farm 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 only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
With the proposal of the double-carbon target, the scale of the wind power plant is gradually increased, the dynamic characteristic of the wind power plant has great influence on the stability of a novel power system, and a simulation model which accurately reflects the dynamic characteristic of the wind power plant needs to be constructed. The invention provides a grouping and parameter optimizing method in a wind power plant. The K-means clustering is susceptible to noise data, dynamic response time is different due to different wind speeds of fans in a wind power plant, noise influence can be eliminated by density-based clustering of the DBSCAN algorithm, the DTW algorithm takes Euclidean distance between normalization paths as similarity indexes between the fans, and the two algorithms are combined to solve the problem.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a general flowchart of the method of the present invention, and as shown in fig. 1, a method for identifying a harmonic resonant frequency gray box of a multi-grid-connected inverter system based on apparent impedance includes the following steps:
step 1: performing dimensionality reduction selection on the grouping index by adopting XGboost-Blending;
step 2: the clustering method based on DBSCAN-DTW is used for cluster division;
and 3, step 3: calculating multi-machine equivalent parameters, converting the multi-machine equivalent parameters into a multi-target nonlinear optimization model, and solving by adopting an improved reference point-based multi-target equilibrium optimization algorithm MEOA;
and 4, step 4: a simulation model is built based on matlab/simulink, and the accuracy and the multi-working-condition wide applicability of the equivalent model after clustering by using the method are verified.
FIG. 2 is a diagram showing the basic structure of CART of the present invention. The XGboost is an integrated algorithm based on a Classification And Regression Tree (CART) model, And the CART is constructed by the following steps: 1) finding a split point: starting from a root node (sample set), traversing the features, and calculating the gain of the model when the sample set is divided into two parts by taking the feature value as a threshold value. 2) Splitting and stopping: and recording the corresponding characteristics and the gain value thereof when the gain is maximum, and dividing the tree into left and right nodes according to the value until the gains of all the split points are less than 0. 3) Results and scores: each leaf node of the last layer corresponds to a subset of samples (i.e., the last classification result), and the gain of each feature can be calculated according to the previous step.
FIG. 3 is a schematic diagram of the Blending fusion step of the present invention. "→" in the figure indicate division so that,the input is represented by a representation of the input,representing the input. DP, DT and DA represent the original prediction data set, the sub-training set test set, MO, respectively1、MO2、…、MOVRepresenting V XGboost metamodels, DA _ P, DP _ P representing an index contribution degree result on an output metamodel, DA _ OUT representing that the DA _ P and an actual result corresponding to original DA data form a new data set, and MODARepresenting a meta-model, MO, based on a data set DA _ OUTDAP represents the meta model MODAOutput of index contribution, MOPRepresentation based on MODAP and DP P form a meta-model of the data set, and the final index degree represents the meta-model MOPAnd outputting the index contribution degree.
FIG. 4 is a wind farm simulation of the present invention. The wind farm consists of 30 DFIGs rated at 1.5 MW. The voltage 575V at the DFIG terminal is boosted to 25kV on site by a unit-to-unit wiring mode, every 6 transformers in a field are connected through an overhead line, and are transmitted to a 25kV/220kV transformer substation and an external power grid, and the initial wind speed data of a fan is shown in an appendix table 1.
TABLE 1 disturbance component parameters of gust disturbance experiment
Fig. 5 is a simulation diagram of the active power transient response of the fan 1 and the fan 2 according to the present invention. When the three-phase short circuit fault occurs in 12s, the active power of the wind turbine generator falls. After a short transient (tens of milliseconds), the steady state value during the fault is reached very quickly. After 12.1s fault clearing, the active power of the fan is subjected to downward and upward overshoot for a certain time, and then is recovered to a normal working state according to a certain slope after being subjected to short-time transient (tens of milliseconds). Fan 2 (red solid line) has reached steady state at 12.5s while fan 1 (blue solid line) is still in the recovery process.
FIG. 6 is a diagram of the k-means clustering results before noise is added. The clustering dataset is X ═ 1,1.1, 1.2,1.2,1.45,1.38,1.68,1.72,1.75,3,3.3,3.15,3.2,4,3.9,4.1,4.15], Y ═ 1,1.1,0.9, 1.25,1.35,1.55,1.5,1.65,1.68,3,3.1,3.05,2.9,4,4.1,3.9,3.95], and the noise point is (2, 2.5). One color representation is classified into a group, cluster 1, cluster 2, cluster 3, and cluster 4 are graphically represented in blue, red, green, and yellow, respectively.
FIG. 7 is a diagram of the k-means clustering result after adding noise according to the present invention. When the noise points are added, the clustering results change, but the algorithm classifies the noise points into cluster 2, and incorrectly classifies the two points in the box that should belong to cluster 2 into cluster 1.
FIG. 8 is a clustering diagram according to the present invention. The DBSCAN algorithm classifies all samples into three categories: core points (represented by black solid dots), boundary points (solid dot colors correspond to each cluster color); noise points (indicated by solid grey dots).
FIG. 9 is a DBSCAN clustering flow chart according to the present invention. After the two parameters are set, one point in the samples can be selected as a core point, all samples meeting the condition that the density can reach are found as a category, and all the points in the category are ensured to be in the Epsilon neighborhood.
Fig. 10 is a diagram of the clustering result of DBSCAN before adding noise according to the present invention. One color representation is classified into clusters, cluster 1, cluster 2, cluster 3, and cluster 4, represented by solid dots of blue, purple, green, and yellow, respectively.
FIG. 11 is a diagram of the clustering result of DBSCAN after adding noise according to the present invention. Cluster 1, cluster 2, cluster 3, and cluster 4 are represented by blue, dark green, yellow graphical solid dots, respectively, and noise is represented by purple solid dots. After the noise is added, the clustering result is not influenced, and the algorithm can automatically identify the noise point.
FIG. 12 is a flowchart of the python algorithm of the present invention. And calculating the optimal sorting path between fans according to the acquired fan data by adopting DTW (dynamic time warping), calculating the similarity according to the sorted Euclidean distance, using the similarity as a clustering basis of DBSCAN (direct space-based clustering controller area network), and continuously adjusting parameters to obtain optimal grouping. At the moment, the calculation dimensionality is extremely high, and strong correlation problems and redundancy among features may exist, so that clustering indexes are selected by Blending XGboost, feature dimensionality reduction is realized, DBSCAN-DTW clustering is carried out again, and clustering results are output.
FIG. 13 is a diagram of different wind speed disturbance scenes according to the present invention. The model was examined in different wind speed scenarios, including gusts (blue solid line), gradual winds (red solid line) and integrated winds (yellow solid line).
FIG. 14 is a schematic view of a wind farm of the present invention. A wind power plant consisting of 16 DFIGs with rated power of 1.5MW is built by using a matlab/simulink simulation platform. The voltage 690V at the DFIG terminal is boosted to 35kV on site in a unit wiring mode from one machine to one machine, and is transmitted to a 35kV/220kV transformer substation through an overhead line and is transmitted to an external power grid.
FIG. 15 is a schematic diagram of a four-machine equivalent wind farm according to the present invention. The clustering index and the DBSCAN-DTW clustering cluster are selected through the XGboost, the equivalent parameters of the DFIG are calculated through a capacity weighting method, and an equivalent schematic diagram obtained by the equivalent parameters of a collecting network in the wind power plant is obtained through a loss invariant principle.
Fig. 16 is a voltage response graph at the grid-connection point. Voltage response curve diagrams of an original model and a DBSCAN-DTW clustering algorithm for XGboost dimension reduction are respectively represented by a blue solid line and a green solid line, and the experiment is simultaneously carried out by using a K-means clustering algorithm (red solid line), a DBSCAN-DTW clustering algorithm (yellow solid line) and a DBSCAN-DTW clustering algorithm for random forest dimension reduction (purple solid line) as a comparison.
Fig. 17 is a graph of an active power response at a grid connection point. Active power response curves of an original model and the DBSCAN-DTW clustering algorithm for XGboost dimensionality reduction are respectively represented by a blue solid line and a green solid line, and the experiment is simultaneously carried out by using a K-means clustering algorithm (red solid line), a DBSCAN-DTW clustering algorithm (yellow solid line) and a DBSCAN-DTW clustering algorithm for random forest dimensionality reduction (purple solid line) as a comparison.
Fig. 18 is a reactive power response graph at a grid connection point. The reactive power response curves of the original model and the DBSCAN-DTW clustering algorithm for XGboost dimensionality reduction are respectively represented by a blue solid line and a green solid line, and the experiment is simultaneously carried out by using a K-means clustering algorithm (red solid line), a DBSCAN-DTW clustering algorithm (yellow solid line) and a DBSCAN-DTW clustering algorithm for random forest dimensionality reduction (purple solid line) as a comparison.
FIG. 19 is a schematic diagram of a five-machine equivalent wind farm of the present invention. And the equivalent model of the wind power plant is divided into groups by a clustering algorithm under the disturbance of the 1 st group of gusts and under the disturbance of comprehensive winds.
FIG. 20 is a diagram illustrating the fusion clustering results of the present invention. And acquiring data under a large number of wind speed scenes by modifying parameters, and clustering and fusing clustering results under different working conditions to obtain clustering results.
Aiming at the problem of low equivalent precision of a wind power plant simulation model, the invention provides a method for performing grouping index dimension reduction by Blending XGboost, optimizing DBSCAN (direct dynamic range) cluster partitioning by adopting DTW (dynamic time warping), and performing grouping result fusion.
(1) The DTW considers that the dynamic response time of the DFIG of the same model in the wind power plant is different under different wind speeds, and effectively solves the problem of data partial deletion of the actual wind power plant.
(2) DBSCAN avoids similar DFIG outliers based on density, and its noise immunity is also significantly enhanced.
(3) Blending is combined with XGboost to select grouping indexes, strong correlation among variables is eliminated, and the speed and accuracy of grouping are improved.
(4) The equivalent parameter calculation problem is converted into a multi-target nonlinear optimization model, and the MEOA is adopted for solving, so that the equivalent precision is improved, and the method has the advantages of multi-working-condition wide applicability.
In the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
The principle and the implementation mode of the invention are explained by applying a specific example, and the description of the embodiment is only used for helping to understand the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the foregoing, the description is not to be taken in a limiting sense.
Claims (5)
1. A method for grouping and parameter optimization in a wind power plant is characterized by comprising the following steps:
step 1: performing dimensionality reduction selection on the grouping index by adopting XGboost-Blending;
step 2: the clustering method based on DBSCAN-DTW is used for cluster division;
and step 3: converting the equivalent parameter calculation problem into a multi-target nonlinear optimization model, and solving by adopting a multi-target equilibrium optimization algorithm MEOA;
and 4, step 4: a simulation model is built based on matlab/simulink, and the accuracy and the multi-working-condition wide applicability of the equivalent model after clustering by using the method are verified.
2. The method for performing clustering index dimension reduction by using Blending fused XGboost according to claim 1, wherein the step 1: the XGboost-Blending is adopted to perform dimensionality reduction selection on the clustering index, and the method specifically comprises the following steps:
the index selection result of the XGboost can be explained, and the influence of different characteristics on the result can be clearly seen. XGBoost is an integrated algorithm based on Classification And Regression Tree (CART) model. The objective function is expressed as:
wherein, L is a training error, namely a loss function about a predicted value and a true value, and represents the matching degree of the model to a training set; y isiIs the true value; omega (f)t) The regular term defined by the L2 norm represents the complexity of the model, and the more complex the model is, the larger the value of the regular term is; c is a constant term.
The loss function is:
to avoid overfitting and dimension disasters, dimension reduction selection needs to be carried out on the features. The wind driven generator characteristic extraction method comprises the following steps:
1) input wind turbine sample x, loss function L.
2) And building a tree by using a greedy algorithm, learning a new function, and fitting the residual error of the last prediction.
3) L is iteratively trained so that the smaller the error, the better.
4) And defining a regular term by adopting an L2 norm, calculating the complexity, and dividing the tree into a structure s and a weight m.
5) After node iteration, an optimal value of the difference between the loss before node segmentation and the loss after node segmentation (namely, the gain of node splitting) is obtained, and the optimal value can be used for calculating the average gain of the features to represent the importance degree of the features, and then more important features are selected to realize dimension reduction:
in the formula (I), the compound is shown in the specification,representing the score of the left sub-tree,representing the score of the right sub-tree,representing the score obtained without segmentation and μ representing the complexity of the new node.
6) And repeating the steps until enough trees are generated, so that the predicted value is closest to the true value, and finishing the algorithm.
The Blending of the Blending model can overcome data crossing, and the Blending model and the XGboost meta model can enhance the learning effect and can not cause excessive redundancy of the whole model, so that the clustering index dimensionality reduction is performed by the Blending of the Blending model and the XGboost. And recording the G value of each feature when the feature is split, and finally dividing the total G value of the feature by the number of times that the feature is used for splitting the node to obtain the quantitative score of the contribution degree of the feature. And deleting the features one by one according to the sequence of the contribution degrees from low to high, clustering again, and traversing to obtain an index selection scheme corresponding to the clustering condition with the highest profile value.
3. The DTW optimized DBSCAN partition cluster according to claim 1, wherein the step 2: the clustering method based on DBSCAN-DTW is provided for cluster division, and specifically comprises the following steps:
DTW is adopted in DBSCAN to calculate the similarity of time sequence data, and points on different sequences are aligned with points through extension and contraction of the sequences, so that the accumulated minimum distance between the points on two sequences with different lengths is calculated, as shown in a formula (4).
Assuming that the sequence-normalized path is R and k denotes the length of the path, the normalized path is R ═ (R)1,r2,…rk) The regular path distance function isThe defined regular path needs to satisfy certain constraint conditions: 1) boundary property: the start and end points of the two sequences P and Q must correspond, i.e.R1=(1,1),Rk(m, n). 2) Monotonicity: the points on the sequences P and Q must be monotonic so that the two sequences do not intersect. 3) Continuity: the points in the sequence can only be matched with adjacent points, and can not be matched in a spanning way, i.e. 0 is less than or equal to i-i' is less than or equal to 1.
The similarity among the multidimensional time sequences is obtained through DTW, and the similarity needs to be substituted into a DBSCAN clustering algorithm for clustering. After two parameters, namely clustering radius Epsilon and minimum MinPts of the number of samples in one category, in the DBSCAN algorithm are set, one point in the samples can be selected randomly as a core point, all samples meeting the condition that the density can reach the density are found as one category, all the points in the category are ensured to be in the Epsilon neighborhood, a sample set is set as E, and a point m is selected randomly as shown in a formula (5).
Epsilon(m,M)={m∈E|d(m,M)≤Epsilon} (5)
4. The method for clustering and fusing the clustering results under the multiple working conditions according to claim 1, wherein the step 3: converting the equivalent parameter calculation problem into a multi-target nonlinear optimization model, and solving by adopting a multi-target equilibrium optimization algorithm MEOA, which specifically comprises the following steps:
calculating parameters of a generator, a transformer and a shafting of the equivalent unit by adopting a capacity weighting method; calculating equivalent wind speed on the basis of the principle that the total wind energy input by the wind turbines in each cluster before and after equivalence is equal; and converting the trunk type topological structure into a radial topological structure according to an equal power loss method, and then calculating equivalent parameters of the current collection system.
Selecting equivalent precision of active power, reactive power and voltage to construct a target function:
wherein:
and solving the multi-target nonlinear optimization model by adopting a multi-target equilibrium optimization algorithm MEOA based on the improved reference point.
5. The authentication method according to claim 1, wherein said step 4: a simulation model is built based on matlab/simulink, the accuracy and the multi-working-condition wide applicability of the equivalent model after clustering by using the method are verified, and the method specifically comprises the following steps:
the method comprises the steps of constructing a wind power plant model containing 16 DFIGs by using matlab/simulink, acquiring characteristic time sequence data by setting three-phase short circuit faults at the outlet of the wind power plant at a certain time period, realizing an algorithm by python, and training the model by adopting a Linux server.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210306444.2A CN114781244A (en) | 2022-03-25 | 2022-03-25 | Grouping and parameter optimization method in wind power plant |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210306444.2A CN114781244A (en) | 2022-03-25 | 2022-03-25 | Grouping and parameter optimization method in wind power plant |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114781244A true CN114781244A (en) | 2022-07-22 |
Family
ID=82424610
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210306444.2A Withdrawn CN114781244A (en) | 2022-03-25 | 2022-03-25 | Grouping and parameter optimization method in wind power plant |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114781244A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115310533A (en) * | 2022-08-05 | 2022-11-08 | 中远海运科技股份有限公司 | AIS-based offshore wind farm identification method and system |
-
2022
- 2022-03-25 CN CN202210306444.2A patent/CN114781244A/en not_active Withdrawn
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115310533A (en) * | 2022-08-05 | 2022-11-08 | 中远海运科技股份有限公司 | AIS-based offshore wind farm identification method and system |
CN115310533B (en) * | 2022-08-05 | 2024-01-19 | 中远海运科技股份有限公司 | AIS-based offshore wind farm identification method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111157899B (en) | Method for estimating SOC of battery based on model fusion idea | |
CN109142946A (en) | Transformer fault detection method based on ant group algorithm optimization random forest | |
CN110969290B (en) | Runoff probability prediction method and system based on deep learning | |
CN111478314B (en) | Transient stability evaluation method for power system | |
CN110909977A (en) | Power grid fault diagnosis method based on ADASYN-DHSD-ET | |
CN112581315A (en) | Wind power field clustering method based on extreme gradient dynamic density clustering | |
CN109672221B (en) | Direct-drive wind power plant dynamic equivalence method for subsynchronous oscillation analysis | |
CN110837915B (en) | Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning | |
CN110009141B (en) | Climbing event prediction method and system based on SDAE feature extraction and SVM classification model | |
CN111725802B (en) | Method for judging transient stability of alternating current-direct current hybrid power grid based on deep neural network | |
CN116842337A (en) | Transformer fault diagnosis method based on LightGBM (gallium nitride based) optimal characteristics and COA-CNN (chip on board) model | |
CN112651576A (en) | Long-term wind power prediction method and device | |
CN115618249A (en) | Low-voltage power distribution station area phase identification method based on LargeVis dimension reduction and DBSCAN clustering | |
CN114781244A (en) | Grouping and parameter optimization method in wind power plant | |
CN111401792A (en) | Dynamic safety assessment method based on extreme gradient lifting decision tree | |
CN115329899A (en) | Clustering equivalent model construction method, system, equipment and storage medium | |
CN108694475B (en) | Short-time-scale photovoltaic cell power generation capacity prediction method based on hybrid model | |
CN117148034A (en) | Power distribution network fault line selection method based on empirical wavelet transformation and graph annotation force network | |
CN117150898A (en) | Power transmission section limit transmission capacity assessment method based on parameter optimization ensemble learning | |
CN115021269B (en) | Two-stage optimal power flow solving method based on data driving | |
CN116826743A (en) | Power load prediction method based on federal graph neural network | |
CN116488151A (en) | Short-term wind power prediction method based on condition generation countermeasure network | |
Wang et al. | Cluster division in wind farm through ensemble modelling | |
CN113466630B (en) | Power distribution network fault cause classification method based on RSSPN model | |
CN112465271B (en) | Energy storage battery type selection method for energy storage stabilizing wind power fluctuation scene |
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 | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20220722 |
|
WW01 | Invention patent application withdrawn after publication |