CN116565877A - Automatic voltage partition control method based on spectral cluster analysis - Google Patents
Automatic voltage partition control method based on spectral cluster analysis Download PDFInfo
- Publication number
- CN116565877A CN116565877A CN202310408317.8A CN202310408317A CN116565877A CN 116565877 A CN116565877 A CN 116565877A CN 202310408317 A CN202310408317 A CN 202310408317A CN 116565877 A CN116565877 A CN 116565877A
- Authority
- CN
- China
- Prior art keywords
- voltage
- operation region
- reactive power
- power
- data set
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000005192 partition Methods 0.000 title claims abstract description 33
- 230000003595 spectral effect Effects 0.000 title claims abstract description 25
- 238000007621 cluster analysis Methods 0.000 title claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 38
- 238000011217 control strategy Methods 0.000 claims abstract description 26
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 18
- 239000011159 matrix material Substances 0.000 claims description 47
- 239000003990 capacitor Substances 0.000 claims description 34
- 239000013598 vector Substances 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 8
- 238000005295 random walk Methods 0.000 claims description 6
- 230000009471 action Effects 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000013459 approach Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 2
- 230000005856 abnormality Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 7
- 238000013473 artificial intelligence Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 5
- 210000002569 neuron Anatomy 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000003064 k means clustering Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000005251 gamma ray Effects 0.000 description 2
- 210000002364 input neuron Anatomy 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 210000004205 output neuron Anatomy 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/12—Circuit arrangements for ac mains or ac distribution networks for adjusting voltage in ac networks by changing a characteristic of the network load
-
- 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
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
-
- 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
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/30—Reactive power compensation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Power Engineering (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a voltage automatic partition control method based on spectral cluster analysis, belonging to the technical field of voltage control of an electric power system; the problem of local voltage abnormality caused by continuous out-of-limit of the power grid voltage is solved; the method comprises the following steps: making a voltage partition control strategy according to the power grid operation data; making voltage control strategy label data according to the voltage partition control operation area; clustering the input data by adopting a spectral cluster analysis algorithm according to the tag data set to obtain an input sample data set corresponding to each type of tag; normalizing the input sample data set; establishing a voltage automatic control model based on a DBN (direct current network), training the voltage automatic control model based on the DBN by adopting an input sample data set, inputting voltage real-time acquisition data into the DBN, classifying and identifying the voltage data based on the DBN, and issuing a voltage control instruction in real time; the invention is applied to automatic control of the power grid voltage.
Description
Technical Field
The invention provides a voltage automatic partition control method based on spectral cluster analysis, and belongs to the technical field of voltage control of power systems.
Background
Compared with the traditional voltage control mode, the AVC control mode has the characteristics of global optimization, high accuracy, nonlinear control and the like, and the AVC control technology is gradually realized and put into application to different degrees along with the continuous development and perfection of the power grid computer technology and the communication network technology in China. Because the power grid voltage control has complex nonlinearity, instantaneity and accuracy requirements, and the artificial intelligence technology has the functions of autonomous learning and knowledge acquisition, the method is suitable for processing the problems of uncertainty, nonlinearity and the like, and has great potential in the research and application of a voltage control method.
The artificial intelligence technology still faces new problems in deep application of the voltage automatic control method, firstly, the voltage control strategy is difficult to formulate by utilizing large-scale power grid operation data, and the neural network is still dependent on the data quantity and the data characteristics of training data samples, although the classification of the result can be effectively realized by the characteristic extraction and the training learning of the data. The expert system can construct the expert system on the original data according to the expert knowledge experience base so as to guide the formation of the sample data. However, when the system scale is large and the rules are complex, the calculation speed and result of the expert system are affected, and meanwhile, the learning and updating of new rules are seriously insufficient. Secondly, the selection of the artificial intelligence method can influence the speed and the result of automatic voltage control, the algorithms such as the traditional neural network have the defects of overlong training time, slower learning algorithm speed and the like, and the data dimension reduction tools such as an automatic encoder and the like are generally adopted for carrying out the dimension reduction and compression of the data, but the neuron number, the layer number and the structural parameters of the traditional neural network cannot be changed, the number of parameters of the traditional neural network is rapidly increased due to the large-scale training of sample data, the training time is overlong, and meanwhile, the local optimum is easily trapped along with the increase of the layer number. Third, the large-scale complex raw data is in a label-free state, and is difficult to classify rapidly. The artificial intelligent network model is adopted for training and learning sample data, classification labels are required to be established, and meanwhile, the sample data is correspondingly classified, but the original data is large in scale, multiple in dimension and complex in structure, so that fine classification is difficult, and the method cannot be further applied to an artificial intelligent network.
Therefore, the invention provides a voltage automatic partition control method based on spectral clustering analysis, and provides a novel power grid dispatching operation data classification method, wherein an artificial intelligence technology is adopted to carry out the voltage automatic control method, label data is formulated according to a voltage control strategy, the label data is adopted to carry out large data classification, and a supervised sample database is trained and learned based on an artificial intelligence model, so that the purpose of power grid voltage automatic control is achieved.
Disclosure of Invention
The invention provides a voltage automatic partition control method based on spectral cluster analysis, which aims to solve the problem of local voltage abnormality caused by continuous out-of-limit of power grid voltage.
In order to solve the technical problems, the invention adopts the following technical scheme: a voltage automatic partition control method based on spectral cluster analysis comprises the following steps:
s1: making a voltage partition control strategy according to the power grid operation data, wherein the voltage partition control strategy realizes the division of a voltage partition control operation area according to voltage out-of-limit control, power factors and reactive power;
s2: making voltage control strategy label data according to the voltage partition control operation areas, wherein each voltage partition control operation area corresponds to one voltage control strategy label, and a label data set is obtained;
s3: clustering the input data by adopting a spectral cluster analysis algorithm according to the tag data set to obtain an input sample data set corresponding to each type of tag;
s4: normalizing the input sample data set to obtain a normalized input sample data set, and dividing the normalized input sample data set into a training set and a testing set;
s5: establishing a voltage automatic control model based on a DBN network, training the voltage automatic control model based on the DBN network by adopting an input sample data set, and inputting voltage real-time acquisition data into the DBN network; based on the DBN network, classifying and identifying the voltage data, and issuing a voltage control instruction in real time; performing capacitor switching and transformer gear lifting according to the voltage control instruction; and the voltage is recovered to be normal, and the power factor is qualified.
In the step S1, the division of the voltage partition control operation area is realized according to the voltage out-of-limit control, the power factor and the reactive power, wherein the specific division of the voltage partition control operation area is divided into the following 9 types of operation areas, which are respectively:
operation region 1: the lower limit of the voltage is higher than the lower limit, the reactive power is too high, and the power factor is unqualified;
operation region 2: the lower limit of the voltage is reached, the reactive power is in the normal range, and the power factor is qualified;
operation region 3: the lower limit of the voltage is exceeded, reactive power is too low and even is negative, reactive power is sent to the power grid, and the power factor is qualified;
operation region 4: the voltage is in a qualified range, the reactive power is too high, and the power factor is unqualified;
operation region 5: the voltage is in a qualified range, reactive power is too low and even is negative, reactive power is sent to a power grid, and power factors are qualified;
operation region 6: the voltage is higher than the upper limit, the reactive power is too high, and the voltage power factor is unqualified;
operation region 7: the upper limit of the voltage is exceeded, the reactive power is in the normal range, and the power factor is qualified;
operation region 8: the voltage is higher than the upper limit, reactive power is too low and even is negative, reactive power is sent to the power grid, and the power factor is qualified;
operation region 9: the voltage, reactive power and power factor are all within normal allowable ranges.
The voltage control strategy labels sequentially formulated for the 9 types of operation areas in the step S2 are as follows:
operation region 1: firstly, a capacitor is put in, and then the main transformer is shifted up;
operation region 2: a main gear shift up;
operation region 3: firstly, performing main transformer upshift, and then exiting the capacitor;
operation region 4: putting a capacitor;
operation region 5: exiting the capacitor;
operation region 6: firstly, performing main transformer downshift, and then putting a capacitor;
operation region 7: a main gear shift down;
operation region 8: the capacitor is withdrawn first, and then the main transformer is shifted down;
operation region 9: normal, no action policy.
The input data in the step S3 are input sample data sets constructed according to a historical operation database of a transformer substation of a power grid, voltage, power factors and reactive power data values of m points in succession at a certain moment are taken to construct single sample points, and n groups of single sample point data are taken to construct the input sample data sets;
clustering the input sample data set based on spectral cluster analysis, and classifying the input sample set according to the label result, wherein the specific flow is as follows:
inputting a sample data set, constructing a similarity matrix and a degree matrix of the input sample data set, calculating the Laplacian matrix of the similarity matrix and the degree matrix and the Laplacian matrix of the random walk, calculating the characteristic values of the Laplacian matrix of the random walk, arranging the characteristic values from small to large, taking the first k characteristic values to construct a characteristic vector, forming k characteristic vectors into a matrix U, extracting row elements of the matrix U, sequencing according to the column direction, reconstructing a new solving space matrix Y, clustering the matrix Y by adopting a k-means algorithm, and remapping a clustering structure into the original input sample data set, wherein a clustering center is selected according to voltage, power factors and reactive power ranges corresponding to 9 voltage control strategy labels.
The training process of the DBN network in the step S5 is as follows:
setting network parameters and learning rate;
inputting the normalized input sample data set as an input vector into a first-layer RBM to complete unsupervised training;
the feature vector extracted after the first layer RBM training is used as an input vector to train the next layer RBM;
training each layer of RBM is completed sequentially according to the steps, and output characteristics of the top layer of RBM are obtained; obtaining a local optimal parameter of each layer of RBM in the RBM training process of each layer;
and setting a Softmax classifier at the top layer of the DBN network, and sending the features extracted by the RBM to the Softmax classifier to be combined with a voltage control strategy label for classification training.
And performing supervised adjustment on the whole network parameters of the pre-trained DBN network by using a sample with a label and using an error back propagation algorithm to enable the network performance to approach global optimum, obtaining the trained DBN network, testing the performance of the DBN network by using a data sample in a test set, and outputting a voltage control strategy classification result.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention provides a voltage control 'nine-grid' strategy, effectively judges the voltage running state according to voltage running data, provides a targeted strategy and establishes a firm research foundation for the data driving realization of voltage automatic control.
2. The invention provides a voltage automatic control method based on deep learning, which adopts power grid operation data to drive and control voltage, and the strong learning capacity and adaptability of the deep learning improve the accuracy and robustness of the voltage control.
3. Compared with the traditional clustering algorithm, the method can cluster and converge on the global optimal solution in any sample space, and ensures the accuracy of voltage control.
4. Compared with the traditional neural network, the method can adjust network parameters through joint distribution between training samples and labels, so that the classification recognition capacity of the whole network is optimal, and the accuracy of automatic voltage control is ensured.
5. The invention establishes the voltage automatic control model, can be effectively applied to the power grid, and can send out the command of voltage automatic control according to the state of the real-time acquisition operation data of the power system, thereby improving the real-time automatic control performance of the voltage.
Drawings
The invention is further described below with reference to the accompanying drawings:
FIG. 1 is a schematic diagram illustrating the division of the voltage partition control operation area according to the present invention;
FIG. 2 is a data sample clustering flow chart based on a spectral clustering algorithm;
FIG. 3 is a clustering flow chart of a data set Y based on a k-means clustering algorithm of the invention;
FIG. 4 is a training flow diagram of the DBN network of the present invention;
fig. 5 is a flowchart of the automatic voltage control based on the DBN network according to the present invention.
Detailed Description
As shown in fig. 1 to 5, the method for controlling the automatic voltage partition based on spectral cluster analysis provided by the invention comprises the following steps:
1. voltage operation database is built based on collection of power grid SCADA system:
E＝{e 1 ，……，e i ，……e n }；
wherein element e i The method comprises the following steps:
wherein: u (u) i 、λ i And Q i For the voltage, power factor and reactive power of the ith sample, m points are taken for each sample point.
2. Formulating tag data according to a voltage control strategy:
I 1 : firstly, a capacitor is put in, and then the main transformer is shifted up;
I 2 : a main gear shift up;
I 3 : firstly, performing main transformer upshift, and then exiting the capacitor;
I 4 : putting a capacitor;
I 5 : exiting the capacitor;
I 6 : firstly, performing main transformer downshift, and then putting a capacitor;
I 7 : main transformerDownshifting;
I 8 : the capacitor is withdrawn first, and then the main transformer is shifted down;
I 9 : normal, no action policy;
constructing a tag dataset: i= { I 1 ,I 2 ,……I 9 }。
3. Sample clustering is performed on input data samples based on spectral cluster analysis.
4. Training the data after spectral cluster analysis based on the DBN network.
5. And establishing a voltage automatic control model based on the DBN network.
6. And collecting data in real time according to the voltage and inputting the data into the DBN network.
7. And classifying and identifying voltage data based on the DBN network, and issuing a voltage control instruction in real time.
8. And switching the capacitor and lifting the gear of the transformer according to the voltage control instruction.
9. And the voltage is recovered to be normal, and the power factor is qualified.
The method comprises the following specific implementation steps:
firstly, making a voltage control strategy according to power grid operation data, and providing a voltage partition control operation area. The voltage partitioning strategy mainly considers voltage out-of-limit control, power factor and reactive power balance, and the division of the operation area is shown in fig. 1. In the control operation region, U H And U L Representing upper and lower operating limits of the voltage; q (Q) H And Q L Representing an allowable range limit for reactive power;
the operation area 9 is a voltage, reactive power and power factor which are all in the normal allowable range, and the voltage is automatically controlled to have no action strategy;
the operation area 5 is that the voltage is in a qualified range, the reactive power is too low or even is negative (i.e. reactive power is sent to the power grid), and if the power factor is qualified, the capacitor should be withdrawn;
the operation area 4 is that the voltage is in a qualified range, the reactive power is too high, and if the power factor is unqualified, a capacitor is put into;
the lower limit of the voltage is set in the operation area 1, reactive power is too high, and if the power factor is unqualified, a capacitor is put into the operation area, and then the main transformer is shifted up;
the lower limit of the voltage is the running area 3, the reactive power is too low or even is negative (i.e. reactive power is sent to the power grid), the power factor is qualified, the main transformer is shifted up, and then the capacitor is withdrawn;
the operation area 6 is that the voltage is higher than the upper limit, the reactive power is too high, the voltage power factor is unqualified, the main transformer is subjected to downshifting, and then the main transformer is put into a capacitor;
the running area 8 is the upper limit of the voltage, the reactive power is too low or even is negative (i.e. reactive power is sent to the power grid), if the power factor is qualified, the capacitor is withdrawn, and then the main transformer is downshifted;
the running area 2 is the lower limit of the voltage, the reactive power is in the normal range, and if the power factor is qualified, the main transformer is shifted up;
the operation area 7 is that the voltage is higher than the upper limit, the reactive power is in the normal range, the power factor is qualified, and the main transformer is shifted down.
And making a voltage control strategy label according to the voltage partition control operation area, wherein the method comprises the following steps of:
operation region 1: firstly, a capacitor is put in, and then the main transformer is shifted up;
operation region 2: a main gear shift up;
operation region 3: firstly, performing main transformer upshift, and then exiting the capacitor;
operation region 4: putting a capacitor;
operation region 5: exiting the capacitor;
operation region 6: firstly, performing main transformer downshift, and then putting a capacitor;
operation region 7: a main gear shift down;
operation region 8: the capacitor is withdrawn first, and then the main transformer is shifted down;
operation region 9: normal, no action policy.
And clustering the input data according to the voltage control strategy labels to obtain an input sample data set corresponding to each type of label.
The input sample data set is constructed according to a historical operation database of a certain transformer substation of the power grid, voltage, power factors and reactive power data values of m points in succession at a certain moment are taken to construct a single sample point, and the expression is as follows:
e i ＝[u i λ i Q i ]；
in the above formula: e, e i Represents the ith sample data, u i 、λ i And Q i The voltage, power factor and reactive power of the ith sample, respectively.
Taking m points for each sample point, the expression is:
then n sets of sample point data are taken to construct an input sample dataset expressed as:
E＝{e 1 ，……，e i ，……e n }。
based on spectral cluster analysis, clustering is carried out on the sample data set E, and input data samples are classified according to label results, wherein the specific flow is as follows:
1. inputting a sample data set E;
2. constructing a similarity matrix W for describing sample characteristics, wherein W is the similarity matrix, and representing a sample point e i Similarity to other sample points, each element W in W ij 0, and W is N is a matrix, wherein each element W in W ij The calculation formula of (2) is as follows:
in the above formula: d (e) i ,e j ) For Euclidean distance between two sample points, sigma is a scale parameter to control the similarity matrix element w ij Along with the change of Euclidean distance of two sample points;
3. building a degree matrix D:
D ij the sum of each row of elements of the similarity matrix W, therefore, the degree matrix D is a diagonal matrix, expressed as follows:
4. laplacian matrix L and randomly walked Laplacian matrix L rw ：
The expression of the laplace matrix is: l=d-W;
the expression of the random walk laplacian matrix is: l (L) rw ＝D -1/2 LD -1/2 ；
In the above formula: d (D) -1/2 Is the inverse square root of the degree matrix D, L rw Is a symmetric laplace matrix, namely: for satisfying any w ij Similarity matrix W which is not less than 0, and random walk Laplacian matrix L rw A symmetric semi-positive definite matrix;
5. calculate L rw And the characteristic equation is: i gamma I-L rw |＝0；
In the above formula: i is an identity matrix, and gamma is a matrix L rw Is written as: gamma ray 1 、γ 2 、……、γ n N eigenvalues are arranged from small to large: gamma ray 1 ≤γ 2 ≤…≤γ k ；
Taking the first k eigenvalues to construct eigenvectors { gamma } 1 ，……，γ k }；
6. Forming k eigenvectors into a matrix U: u= { U 1 ，u 2 ，……u k }；
Where u=γ, U is a matrix of n×k;
7. extracting each row of elements y of U i (i=1, … …, n) reconstruct the new solution matrix Y, i.e.:
Y＝{y 1 ，y 2 ，……y n }；
8. clustering the matrix Y by adopting a K-Means clustering algorithm to obtain clustersResults: { c 1 ，c 2 ，c 3 ，……，c p Mapping the clustering result back to the original input sample dataset, i.e.: if the ith row of the matrix U is clustered into subclass p, then the ith data in the original input sample dataset belongs to subclass p.
The K-Means clustering algorithm is as follows:
1. setting a clustering center and the number of clusters p, wherein in the invention, p=9, the clustering center is selected according to the voltage, the power factor and the reactive power range corresponding to each label, and the clustering center is recorded as: { c 1 ，c 2 ，c 3 ，……，c p }；
2. The Euclidean distance between the elements in the matrix Y and each clustering center is calculated:
wherein y is i ∈Y，i＝1,……n；c j For the cluster center, j=1, … … p, d ij Is the element y i Euclidean distance from the clustering center;
3. according to element y i Clustering the Euclidean distance to each clustering center:
min{d i1 ，d i2 ，……d ip }；
element y i Dividing into clusters with minimum Euclidean distance;
4. recalculating and adjusting the cluster centers:
wherein: c (C) j The j-th cluster set after clustering;
5. re-calculating Euclidean distance for clustering, and repeating the steps 2 and 3;
6. and converging the minimized square error as a clustering objective function judgment algorithm:
in the above formula: j is a clustering objective function.
7. When the function J reaches the minimum, obtaining a clustering optimal result, and outputting a clustering center and clustering data: { C 1 ，C 2 ，……C p }。
Obtaining an input sample training data set through spectral cluster analysis: e' = { C 1 ，C 2 ，……C p }；
Wherein p=9.
The output tag dataset is: i' = { I 1 ，I 2 ，……I 9 }；
Wherein I is 1 、……I 9 Respectively representing 9 kinds of tag data.
After the input and output sample data sets are constructed, data preprocessing is carried out, and the accuracy and the speed of an algorithm can be influenced during training operation due to inconsistent dimensions of voltage, power factor and reactive power adopted by the invention, so that normalization processing is carried out, namely:
wherein: e, e min U is taken min 、λ min 、Q min ；e max U is taken max 、λ max 、Q max For u respectively i 、λ i And Q i And (5) carrying out normalization processing.
The normalized input sample dataset is: e (E) * ＝{C 1 * ，C 2 * ，……C 9 * }。
Data training was performed using a deep belief network (Deep belief network, DBN) and classification was performed by connecting softmax regression classifications at the top layer of the DBN network.
The automatic voltage control method based on DBN is as follows:
1. setting network parameters according to the normalized input sample data set E: n input layer neurons; the momentum parameter beta is usually 0.05; the learning rate alpha is usually 0.001;
2. inputting the normalized input sample data set E as an input vector into a first layer RBM to complete unsupervised training;
3. the feature vector extracted after the first layer RBM training is used as an input vector to train the next layer RBM;
4. sequentially completing RBM training of each layer according to 2 and 3 to obtain output characteristics of RBM of the top layer; in the training process of each layer of RBM, obtaining the local optimal parameters of each layer of RBM, wherein the calculation formula is as follows:
w new ＝w+v；
wherein: beta and alpha are super parameters of the DBN network, beta is momentum parameter, alpha is learning rate, v is gradient,the weight gradient is w is a DBN network weight parameter;
5. and setting a Softmax classifier at the top layer of the DBN network, and sending the features extracted by the RBM to the Softmax classifier to be combined with a voltage control strategy label for classification training.
Calculating the probability of the feature belonging to each category by using a probability calculation function, and finishing the classification task; the probability value function is calculated as follows:
in the above formula: q represents a tag class, q=9. X is x i Representing the ith input sample data sequence, y i And the label value of the classification result corresponding to the ith input sample data is represented. h is a θ (x i ) And (3) representing the probability value of the ith sample data corresponding to various output results, wherein θ is a built-in parameter of the Softmax network.
Wherein, the cross entropy loss function in the Softmax classifier training is shown as follows:
wherein m is the number of training samples, q represents the number of categories, and y ij Representing the input sample as x i Output of network at time t ij Representing the target output class of the i-th sample.
6. And (3) using a labeled sample, performing supervised fine tuning on the whole network parameters of the pre-trained DBN model by using an error back propagation algorithm, so that the network performance approaches global optimum, wherein the main formula is as follows:
the loss function of the back propagation algorithm is:
wherein r is j Representing the true value, y, of the jth neuron j The predicted value of the j-th neuron output is represented, and p represents the number of output neurons.
Weight adjustment between neurons: Δw ij k ＝η(r j -y j )y j (1-y j )x i ；
Wherein Deltaw ij k An adjustment value representing the connection weight between the ith input neuron and the jth output neuron in the kth network layer, eta being the adjustment step size, x i Representing the value of the ith input neuron.
The adjustment value of the final connection weight is as follows:
7. and obtaining a trained DBN model, testing the performance of the DBN model by utilizing data samples in a data test set, and outputting a voltage control strategy classification result.
According to the invention, three main factors of voltage deviation, power factor and reactive power of a power grid are comprehensively considered to control the voltage, and the voltage of different operation areas is controlled by adjusting the main gear and switching of a capacitor, so that the voltage is ensured to operate in a stable range. And (3) compiling and dividing label data by adopting voltage control instructions of different voltage operation areas, and training voltage, reactive power and power factor data of a power grid node (transformer substation) based on a deep confidence network so as to achieve the aim of accurately identifying control instruction labels. Aiming at the defects that the original training data volume is too large and cannot be classified rapidly, the method adopts spectral clustering analysis to perform the dimension reduction and clustering of the original data, performs clustering analysis according to a voltage operation area, improves the accuracy and reliability of automatic voltage control, and controls the power grid voltage in a stable range in real time.
The specific structure of the invention needs to be described that the connection relation between the component modules adopted by the invention is definite and realizable, and besides the specific description in the embodiment, the specific connection relation can bring about corresponding technical effects, and on the premise of not depending on execution of corresponding software programs, the technical problems of the invention are solved, the types of the components, the modules and the specific components, the connection modes of the components and the expected technical effects brought by the technical characteristics are clear, complete and realizable, and the conventional use method and the expected technical effects brought by the technical characteristics are all disclosed in patents, journal papers, technical manuals, technical dictionaries and textbooks which can be acquired by a person in the field before the application date, or the prior art such as conventional technology, common knowledge in the field, and the like, so that the provided technical scheme is clear, complete and the corresponding entity products can be reproduced or obtained according to the technical means.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.
Claims (6)
1. A voltage automatic partition control method based on spectral cluster analysis is characterized in that: the method comprises the following steps:
s1: making a voltage partition control strategy according to the power grid operation data, wherein the voltage partition control strategy realizes the division of a voltage partition control operation area according to voltage out-of-limit control, power factors and reactive power;
s2: making voltage control strategy label data according to the voltage partition control operation areas, wherein each voltage partition control operation area corresponds to one voltage control strategy label, and a label data set is obtained;
s3: clustering the input data by adopting a spectral cluster analysis algorithm according to the tag data set to obtain an input sample data set corresponding to each type of tag;
s4: normalizing the input sample data set to obtain a normalized input sample data set, and dividing the normalized input sample data set into a training set and a testing set;
s5: establishing a voltage automatic control model based on a DBN network, training the voltage automatic control model based on the DBN network by adopting an input sample data set, and inputting voltage real-time acquisition data into the DBN network; based on the DBN network, classifying and identifying the voltage data, and issuing a voltage control instruction in real time; performing capacitor switching and transformer gear lifting according to the voltage control instruction; and the voltage is recovered to be normal, and the power factor is qualified.
2. The automatic voltage partition control method based on spectral cluster analysis according to claim 1, wherein the method comprises the following steps: in the step S1, the division of the voltage partition control operation area is realized according to the voltage out-of-limit control, the power factor and the reactive power, wherein the specific division of the voltage partition control operation area is divided into the following 9 types of operation areas, which are respectively:
operation region 1: the lower limit of the voltage is higher than the lower limit, the reactive power is too high, and the power factor is unqualified;
operation region 2: the lower limit of the voltage is reached, the reactive power is in the normal range, and the power factor is qualified;
operation region 3: the lower limit of the voltage is exceeded, reactive power is too low and even is negative, reactive power is sent to the power grid, and the power factor is qualified;
operation region 4: the voltage is in a qualified range, the reactive power is too high, and the power factor is unqualified;
operation region 5: the voltage is in a qualified range, reactive power is too low and even is negative, reactive power is sent to a power grid, and power factors are qualified;
operation region 6: the voltage is higher than the upper limit, the reactive power is too high, and the voltage power factor is unqualified;
operation region 7: the upper limit of the voltage is exceeded, the reactive power is in the normal range, and the power factor is qualified;
operation region 8: the voltage is higher than the upper limit, reactive power is too low and even is negative, reactive power is sent to the power grid, and the power factor is qualified;
operation region 9: the voltage, reactive power and power factor are all within normal allowable ranges.
3. The automatic voltage partition control method based on spectral cluster analysis according to claim 2, wherein the method comprises the following steps: the voltage control strategy labels sequentially formulated for the 9 types of operation areas in the step S2 are as follows:
operation region 1: firstly, a capacitor is put in, and then the main transformer is shifted up;
operation region 2: a main gear shift up;
operation region 3: firstly, performing main transformer upshift, and then exiting the capacitor;
operation region 4: putting a capacitor;
operation region 5: exiting the capacitor;
operation region 6: firstly, performing main transformer downshift, and then putting a capacitor;
operation region 7: a main gear shift down;
operation region 8: the capacitor is withdrawn first, and then the main transformer is shifted down;
operation region 9: normal, no action policy.
4. The automatic voltage partition control method based on spectral cluster analysis according to claim 3, wherein: the input data in the step S3 are input sample data sets constructed according to a historical operation database of a transformer substation of a power grid, voltage, power factors and reactive power data values of m points in succession at a certain moment are taken to construct single sample points, and n groups of single sample point data are taken to construct the input sample data sets;
clustering the input sample data set based on spectral cluster analysis, and classifying the input sample set according to the label result, wherein the specific flow is as follows:
inputting a sample data set, constructing a similarity matrix and a degree matrix of the input sample data set, calculating the Laplacian matrix of the similarity matrix and the degree matrix and the Laplacian matrix of the random walk, calculating the characteristic values of the Laplacian matrix of the random walk, arranging the characteristic values from small to large, taking the first k characteristic values to construct a characteristic vector, forming k characteristic vectors into a matrix U, extracting row elements of the matrix U, sequencing according to the column direction, reconstructing a new solving space matrix Y, clustering the matrix Y by adopting a k-means algorithm, and remapping a clustering structure into the original input sample data set, wherein a clustering center is selected according to voltage, power factors and reactive power ranges corresponding to 9 voltage control strategy labels.
5. The automatic voltage partition control method based on spectral cluster analysis according to claim 4, wherein the method comprises the following steps: the training process of the DBN network in the step S5 is as follows:
setting network parameters and learning rate;
inputting the normalized input sample data set as an input vector into a first-layer RBM to complete unsupervised training;
the feature vector extracted after the first layer RBM training is used as an input vector to train the next layer RBM;
training each layer of RBM is completed sequentially according to the steps, and output characteristics of the top layer of RBM are obtained; obtaining a local optimal parameter of each layer of RBM in the RBM training process of each layer;
and setting a Softmax classifier at the top layer of the DBN network, and sending the features extracted by the RBM to the Softmax classifier to be combined with a voltage control strategy label for classification training.
6. The automatic voltage partition control method based on spectral cluster analysis according to claim 5, wherein the method comprises the following steps: and performing supervised adjustment on the whole network parameters of the pre-trained DBN network by using a sample with a label and using an error back propagation algorithm to enable the network performance to approach global optimum, obtaining the trained DBN network, testing the performance of the DBN network by using a data sample in a test set, and outputting a voltage control strategy classification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310408317.8A CN116565877A (en) | 2023-04-17 | 2023-04-17 | Automatic voltage partition control method based on spectral cluster analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310408317.8A CN116565877A (en) | 2023-04-17 | 2023-04-17 | Automatic voltage partition control method based on spectral cluster analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116565877A true CN116565877A (en) | 2023-08-08 |
Family
ID=87495579
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310408317.8A Pending CN116565877A (en) | 2023-04-17 | 2023-04-17 | Automatic voltage partition control method based on spectral cluster analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116565877A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118011280A (en) * | 2024-04-09 | 2024-05-10 | 国网山东省电力公司东营供电公司 | Voltage quality on-line monitoring and analyzing method and system |
-
2023
- 2023-04-17 CN CN202310408317.8A patent/CN116565877A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118011280A (en) * | 2024-04-09 | 2024-05-10 | 国网山东省电力公司东营供电公司 | Voltage quality on-line monitoring and analyzing method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110378799B (en) | Alumina comprehensive production index decision method based on multi-scale deep convolution network | |
| CN108205717A (en) | A kind of photovoltaic generation power Multiple Time Scales Forecasting Methodology | |
| CN109063911A (en) | A kind of Load aggregation body regrouping prediction method based on gating cycle unit networks | |
| CN112990558B (en) | Meteorological temperature and illumination prediction method based on deep migration learning | |
| CN116503676B (en) | Picture classification method and system based on knowledge distillation small sample increment learning | |
| CN111461921B (en) | Load modeling typical user database updating method based on machine learning | |
| CN110400018B (en) | Operation control method, system and device for coal-fired power plant pulverizing system | |
| CN115481788B (en) | Phase change energy storage system load prediction method and system | |
| CN112308298B (en) | Multi-scenario performance index prediction method and system for semiconductor production line | |
| CN110717581A (en) | Short-term load prediction method based on temperature fuzzy processing and DBN | |
| CN116565877A (en) | Automatic voltage partition control method based on spectral cluster analysis | |
| CN114219181A (en) | Wind power probability prediction method based on transfer learning | |
| CN108345866A (en) | A kind of pedestrian's recognition methods again based on depth characteristic study | |
| CN114626504A (en) | Model compression method based on group relation knowledge distillation | |
| CN107577994A (en) | A kind of pedestrian based on deep learning, the identification of vehicle auxiliary product and search method | |
| CN114936599A (en) | Base station energy consumption abnormity monitoring method and system based on wavelet decomposition and migration discrimination | |
| CN113422371B (en) | Distributed power supply local voltage control method based on graph convolution neural network | |
| CN110991689A (en) | Distributed photovoltaic power generation system short-term prediction method based on LSTM-Morlet model | |
| CN110570091A (en) | Load identification method based on improved F-score feature selection and particle swarm BP neural network | |
| CN113887794A (en) | Method and device for reactive power optimization of power distribution network | |
| CN117574776A (en) | Task planning-oriented model self-learning optimization method | |
| CN117407758A (en) | Cloud-edge collaborative load library self-learning method | |
| CN115021269A (en) | Two-stage optimal power flow solving method based on data driving | |
| CN114254828A (en) | Power load prediction method based on hybrid convolution feature extractor and GRU | |
| CN116646927B (en) | Wind power prediction method based on segmented filtering and longitudinal and transverse clustering |
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 |