CN113496255A - Power distribution network hybrid observation point distribution method based on deep learning and decision tree driving - Google Patents
Power distribution network hybrid observation point distribution method based on deep learning and decision tree driving Download PDFInfo
- Publication number
- CN113496255A CN113496255A CN202110617672.7A CN202110617672A CN113496255A CN 113496255 A CN113496255 A CN 113496255A CN 202110617672 A CN202110617672 A CN 202110617672A CN 113496255 A CN113496255 A CN 113496255A
- Authority
- CN
- China
- Prior art keywords
- distribution network
- power distribution
- active power
- sample
- topology
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000003066 decision tree Methods 0.000 title claims abstract description 44
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000013135 deep learning Methods 0.000 title claims abstract description 16
- 238000005259 measurement Methods 0.000 claims abstract description 25
- 230000008878 coupling Effects 0.000 claims abstract description 15
- 238000010168 coupling process Methods 0.000 claims abstract description 15
- 238000005859 coupling reaction Methods 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 11
- 238000004458 analytical method Methods 0.000 claims description 15
- 238000013016 damping Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000011218 segmentation Effects 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 238000005192 partition Methods 0.000 claims description 2
- 238000005457 optimization Methods 0.000 abstract description 4
- 238000013136 deep learning model Methods 0.000 abstract 2
- 210000002569 neuron Anatomy 0.000 description 8
- 230000000007 visual effect Effects 0.000 description 7
- 230000001186 cumulative effect Effects 0.000 description 3
- 239000012535 impurity Substances 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000002347 injection Methods 0.000 description 2
- 239000007924 injection Substances 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012847 principal component analysis method Methods 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 230000002547 anomalous effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000000243 solution Substances 0.000 description 1
- 238000000528 statistical test 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
- 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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Human Resources & Organizations (AREA)
- General Health & Medical Sciences (AREA)
- Water Supply & Treatment (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a power distribution network hybrid observation point distribution method based on deep learning and decision tree driving, which relates to the technical field of power system optimization, and is characterized by firstly constructing a decision tree model to quantitatively analyze the influence importance of each node characteristic of a power distribution network on the topology observability of the power distribution network, and selecting part of node measurement characteristics with high importance and measurement data to form a hybrid measurement sample set for deep learning model training according to the sequence of the node characteristic importance from large to small; and then analyzing the topology observability change performance index of the power distribution network under different mixed measurement schemes by adopting a PCA-DBN coupling deep learning model, and finally determining the optimal point distribution scheme of the mixed measurement device according to the selected nodes when the power distribution network is completely observable and the mixed measurement scheme. The method breaks through the planning idea of the traditional measuring device optimization point distribution method, and adopts data driving methods such as deep learning and decision trees to realize an economical and efficient hybrid observation point distribution planning scheme for active power distribution network topology identification.
Description
Technical Field
The invention relates to the technical field of power system optimization, in particular to a power distribution network hybrid observation and point distribution method based on deep learning and decision tree driving.
Background
With the development of distribution network automation, more and more AMI and PMU equivalent measurement devices are put into a distribution network, and data support is provided for solving complex nonlinear problems such as active distribution network state estimation and topology identification by adopting a data science method represented by deep learning. However, the data-based data science method has higher requirements on the quality and quantity of sample data, and particularly in a large-scale power distribution network, because high-precision automatic power distribution measurement devices such as AMI (advanced metering infrastructure) and PMU (power management unit) are expensive, the measurement devices are very limited in configuration quantity, the quantity and precision of real-time measurement data still cannot meet the requirements of data required by the current algorithm, and pseudo measurement data based on prediction has to be used in most scenes; although more advanced measurement devices are already put into part of the power grid, the distribution positions of the power grid are not reasonable enough, the collected measurement data contain more redundant information, the cost of each measurement device is different, and data incompatibility caused by data delay, different data periods, large data precision difference and the like among the measurement data is a completely considerable important factor influencing the power distribution network. In order to consider the economy and reliability of power data sources and improve the observability of a power distribution network under the power distribution network state estimation and topology identification scenes, a new measurement system needs to be planned and configured in a power distribution network system or an existing measurement system needs to be improved, and the observability of the power distribution network topology is realized by fully utilizing mixed measurement data. Under the current environment of rapid development of the power distribution network, powerful support of related technologies is urgently needed to guide topology identification hybrid observation point distribution and planning of the active power distribution network.
Disclosure of Invention
In view of the technical defects, the invention provides a power distribution network hybrid observation point distribution method based on deep learning and decision tree driving.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a power distribution network hybrid observation point distribution method based on deep learning and decision tree driving comprises the following steps:
s1, obtaining an active power distribution network running state offline sample T;
s2, constructing a decision tree model based on the off-line samples of the running state of the active power distribution network, inputting the off-line samples obtained in S1 into the decision tree model for analysis and calculation to obtain the feature importance of the nodes, arranging the feature importance according to the descending order, selecting the measured data of the nodes corresponding to the first n feature importance of the sequence, and selecting a candidate mixed observation stationing sample set;
s3, inputting the alternative mixed observation stationing sample set obtained in the step S2 into a PCA-DBN coupling topology recognition model, analyzing the observability of the topology of the active power distribution network, judging whether the topology of the active power distribution network is completely observable, if the topology of the active power distribution network is completely observable, entering the step S4, if the topology of the active power distribution network is not completely observable, selecting the measurement data of the nodes corresponding to the first n +1 feature importance degrees of the sequence to form an alternative mixed observation stationing scheme, and inputting the alternative mixed observation stationing scheme into the PCA-DBN coupling topology recognition model again;
and S4, obtaining a mixed observation point distribution scheme.
Preferably, the process of obtaining the offline sample of the operation state of the active power distribution network in S1 is as follows:
s10, selecting a power distribution network operation topological structure set G meeting three operation characteristics of radial, loop-free and island-free of an active power distribution networkT=(GT1,GT2,…,GTm);
S11, from the feasible topology set GTIn select topology GTa;
S12, based on feasible topology GTaGenerating different active power distribution network operation scenes according to a typical daily load fluctuation curve and a photovoltaic output curve and considering photovoltaic and ZIP loads;
and S13, calculating the load flow distribution under different active traveling scenes, and recording the active distribution network operation state data to form an active distribution network operation state offline sample, wherein the offline sample comprises the active power P of each node of the active distribution network, the node voltage V and the voltage phase angle difference theta as the input of the sample, and the correspondingly input distribution network tie switch and section switch states as the sample output.
Preferably, the step of constructing the decision tree model in S2 includes the following steps:
carrying out supervised learning on an active power distribution network running state offline sample T which takes X as an input variable and Y as an output variable, wherein the active power distribution network running state offline sample is as follows: t { (x)1,y1),(x2,y2),…,(xn,yn)};
The mathematical expression of the kini coefficient of the active power distribution network running state off-line sample T is as follows:
the active power distribution network running state offline sample is provided with K categories CkThe number of samples of the Kth category;
the decision tree model is a binary decision tree recursively constructed from a root node through an active power distribution network running state offline sample based on a minimum criterion of a kini coefficient, and the decision tree model is established in the following process:
s20, inputting an active power distribution network running state off-line sample T and a Gini coefficient threshold value gini;
S21, calculating each characteristic of the active power distribution network operation state offline sample and the loss function of the division node, and randomly selecting the jth characteristic x of the active power distribution network operation state offline sample TjCut variable as sample and xjThe value s is used as a dividing node of the sample, and the off-line sample T of the running state of the active power distribution network is divided into two subdata sets Ts1And Ts2The mathematical expression of the partition principle is as follows:
the loss function being after divisionTwo subdata sets Ts1And Ts2The mean square error of (a), the corresponding feature segmentation variable x when the loss function reaches the set minimum valuejAnd a dividing node s, the calculation formula of which is as follows:
in the formula c1Representing a sub data set Ts1Average value of output, c2Representing a sub data set Ts2Average value of the output, wherein
S22, respectively calculating the damping coefficients of the two division nodes, and judging whether the damping coefficients of the two division nodes are smaller than a damping coefficient threshold value giniWhen the damping coefficients of the two divided nodes are not less than the damping coefficient threshold value giniIf so, the two divided nodes respectively return to the step S21 to continue the node recursive division; when the Gini coefficient of one of the divided nodes is smaller than the threshold value g of the Gini coefficientiniThen the node stops the node recursive partitioning, and the node is not smaller than the threshold value g of the Gini coefficientiniThe other divided node returns to step S21 to continue the node recursive division until the kini coefficient recursively reaches the other divided node and is less than the threshold value g of the kini coefficientiniObtaining K subspaces of all divided nodes;
s23, dividing the input active power distribution network operation state offline samples into K subspaces, wherein each subspace comprises partial sample data and an average value c of subspace output valuesKAnd obtaining a decision tree model, wherein the mathematical expression of the decision tree model is as follows:
in the formula I (x is equal to T)sK) An indicator function representing the model.
Preferably, the offline samples obtained in step S1 are input to the decision tree model for analysis and calculation in step S2 as follows:
in the formula, N represents the total number of off-line samples of the running state of the active power distribution network, and NtNumber of offline samples, N, representing active distribution network operating state of current nodetRRepresenting the left sub-tree, N, of the current nodetLThe number of the active power distribution network running state offline samples of the left sub-tree of the current node is represented, H represents the purity of the current node, and H represents the purity of the current noderightRepresenting the left sub-tree of the current node, HleftIndicating the degree of purity of the left sub-tree of the current node.
Preferably, in step S3, the alternative mixed observation point distribution sample set obtained in step S2 is input into the PCA-DBN coupled topology recognition model, and the process of analyzing the observability of the topology of the active power distribution network is as follows:
s30, preprocessing the alternative mixed observation stationing sample set;
s31, selecting principal component variables meeting the requirement of topology identification accuracy by adopting PCA, and simultaneously reducing the noise and data dimensionality of node voltage data;
s32, constructing a DBN network, and applying the DBN to train and learn the nonlinear relation between the node voltage of the sample and the network topology;
s33, after the training is finished, storing the PCA-DBN coupling model after the training;
and S34, inputting the preprocessed alternative mixed observation stationing sample set into the trained PCA-DBN coupling model topology recognizer for power distribution network topology identification analysis, and obtaining an analysis result of the topology observability of the active power distribution network.
The invention has the beneficial effects that: analyzing the influence degree of node measurement data on the observability of the power distribution network quantitatively through a data-driven decision tree model, preliminarily screening the positions of mixed observation distribution points, and formulating different mixed observation distribution points according to the importance degree of the nodes;
then based on a mixed observation stationing scheme, a PCA-DBN deep learning method is adopted to carry out observability analysis on the topology of the active power distribution network, a principal component analysis method is used to adaptively select the principal characteristics in the measured data, the measured redundant information is removed, and the learning efficiency of the DBN can be effectively improved;
the DBN is adopted to process non-sequence sample data, and the problem of complex nonlinearity in active power distribution network topology identification is effectively solved. And finally, determining an economical and efficient mixed observation point arrangement planning scheme according to observability analysis results of different mixed observation point arrangement schemes.
Drawings
FIG. 1 is a schematic view; the invention provides; an IEEE33 node standard power distribution system;
FIG. 2 is a schematic view; the invention provides; system node feature importance percentage;
FIG. 3 is a schematic view; the invention provides; system node feature importance;
FIG. 4 is a schematic view; the invention provides; the measurement device distribution quantity influences the topology observability of the power distribution network;
FIG. 5 is a schematic view; the invention provides; a system mixed observation stationing planning scheme;
FIG. 6 is a schematic view; the invention provides; the flow chart is schematic.
Detailed Description
The technical solutions in the embodiments of the present invention are clearly and completely described below with reference to the accompanying drawings, and other advantages and effects of the present invention will be readily apparent to those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
A power distribution network hybrid observation point distribution method based on deep learning and decision tree driving comprises the following steps:
s1, obtaining an active power distribution network running state offline sample T;
s2, constructing a decision tree model based on the off-line samples of the running state of the active power distribution network, inputting the off-line samples obtained in S1 into the decision tree model for analysis and calculation to obtain the feature importance of the nodes, arranging the feature importance according to the descending order, selecting the measured data of the nodes corresponding to the first n feature importance of the sequence, and forming an alternative mixed observation stationing sample set;
s3, inputting the alternative mixed observation stationing sample set obtained in the step S2 into a PCA-DBN coupling topology recognition model, analyzing the observability of the topology of the active power distribution network, judging whether the topology of the active power distribution network is completely observable, if the topology of the active power distribution network is completely observable, entering the step S4, if the topology of the active power distribution network is not completely observable, selecting the measurement data of the nodes corresponding to the first n +1 feature importance degrees of the sequence to form an alternative mixed observation stationing scheme, and inputting the alternative mixed observation stationing scheme into the PCA-DBN coupling topology recognition model again;
and S4, obtaining a mixed observation point distribution scheme.
Preferably, the process of obtaining the offline sample of the operation state of the active power distribution network in S1 is as follows:
s10, selecting a power distribution network operation topological structure set satisfying three operation characteristics of radial, loop-free and island-free of an active power distribution network as GT=(GT1,GT2,…,GTm);
S11, from the feasible topology set GTIn select topology GTa;
S12, based on feasible topology GTaGenerating different power distribution network operation scenes according to the typical daily load fluctuation curve and the photovoltaic output curve; meanwhile, in order to simulate more real operation conditions of the active power distribution network system, large-scale photovoltaic access power distribution networks are considered, and all load types are ZIP loads.
And S13, calculating the load flow distribution under different active traveling scenes, and recording the active distribution network operation state data to form an active distribution network operation state offline sample, wherein the offline sample comprises the active power P of each node of the active distribution network, the node voltage V and the voltage phase angle difference theta as the input of the sample, and the correspondingly input distribution network tie switch and section switch states as the sample output.
The input reasons for selecting the active power P of the node, the node voltage V and the voltage phase angle difference theta as samples are as follows:
for any distribution network determined by the network topology structure, the steady-state operation state of the power grid can also be determined by solving the power flow. The power flow equation of the power system is shown as follows:
in the formula PiAnd QiRespectively representing injected active and reactive power, V, of node iiRepresenting the sum voltage magnitude, θ, of node iijRepresenting the voltage angle difference between node i and node j. According to the power flow equation, the other two variables can be determined for the power distribution network determined by the network topology as long as any two of P, Q, V and theta are known.
Preferably, the step of constructing the decision tree model in S2 includes the following steps:
carrying out supervised learning on an active power distribution network operation state offline sample T which takes X as an input variable and Y as an output variable, wherein the active power distribution network operation state offline sample T { (X)1,y1),(x2,y2),…,(xn,yn) And (4) assuming that K types of off-line samples of the running state of the active power distribution network exist, wherein the number of the Kth type samples is Ck. In the process of establishing the decision tree, the operation state offline samples of the active power distribution network need to be divided, and the jth characteristic x of the operation state offline samples T of the active power distribution network is usually selectedjAnd the value s is used as a segmentation variable and a division node of the sample, and the off-line sample of the running state of the active power distribution network is divided into twoSubdata set Ts1And Ts2The division principle is as follows:
the present invention selects features based on the criterion of minimum kini coefficients. The expression of the Keyny coefficient of the active power distribution network operation state off-line sample T is as follows:
the magnitude of the kiney coefficient represents the purity of the model, and a smaller kiney coefficient indicates a better selected feature of the model. The decision tree model is a binary decision tree recursively constructed by taking the minimum kini coefficient of off-line sample data of the active power distribution network operation state as a criterion from a root node, and the construction process of the decision tree model is as follows:
s20, inputting an active power distribution network running state off-line sample T and a Gini coefficient threshold value gini;
S21, calculating a loss function of each characteristic of the active power distribution network running state offline sample and the division node, wherein the loss function is divided into two sub data sets Ts1And Ts2Selecting a corresponding feature segmentation variable xj and a corresponding segmentation node s when the loss function is minimum, wherein a calculation formula is shown as the following formula:
in the formula c1And c2Respectively representing two sub data sets Ts1And Ts2Average value of the output, wherein
S22, dividing the input active distribution network operation state offline sample into two sub-samples through the step S21Data set Ts1And Ts2;
S23, respectively calculating the damping coefficients of the two division nodes, and judging whether the damping coefficients of the two division nodes are smaller than a damping coefficient threshold value giniWhen the damping coefficients of the two divided nodes are not less than the damping coefficient threshold value giniIf so, the two divided nodes respectively return to the step S21 to continue the node recursive division; when the Gini coefficient of one of the divided nodes is smaller than the threshold value g of the Gini coefficientiniThen the node stops the node recursive partitioning, and the node is not smaller than the threshold value g of the Gini coefficientiniThe other divided node returns to step S21 to continue the node recursive division until the kini coefficient recursively reaches the other divided node and is less than the threshold value g of the kini coefficientiniObtaining K subspaces of all divided nodes;
s24, finally dividing the input active power distribution network operation state offline samples into K subspaces, wherein each subspace comprises partial sample data and an average value c of subspace output valuesKThe final decision tree model can be expressed as
In the formula I (x is equal to T)sK) An indicator function representing the model.
For a sample set containing multiple features, the decision tree can calculate a division standard value of each feature in the sample, the division standard value is used as an importance degree calculation index of each feature, and the contribution degree of each feature in the sample to a target variable can be determined through the feature importance degree.
Preferably, in step S2, the feature importance of each node of the active power distribution network is calculated by the decision tree model as follows:
in which N and NtRespectively representing the total quantity of the off-line samples of the running state of the active power distribution network and the quantity, N, of the off-line samples of the running state of the active power distribution network of the current nodetRAnd NtLRespectively representing the quantity of offline samples of the operating states of the active power distribution network of the current node left sub-tree and the current node left sub-tree. H represents the purity of the current node, the impurity is the sum of squares of errors of off-line samples of the running state of the active power distribution network of the node, and HrightAnd HleftAnd respectively representing the impurity degrees of the left sub-tree of the current node and the left sub-tree of the current node, wherein the smaller the impurity degree H of the output node is, the higher the feature importance degree of the node is.
Preferably, in step S3, the alternative mixed observation point distribution sample set obtained in step S2 is input into the PCA-DBN coupled topology recognition model, and the process of analyzing the observability of the topology of the active power distribution network is as follows:
and S30, preprocessing the sample set of the alternative mixed observation stationed points.
The preprocessing method for the alternative mixed observation stationing sample set is as follows:
data quality issues mainly include data loss, erroneous data, and high levels of measurement noise. When missing data is processed, a k neighbor algorithm is adopted to complement and modify sample data. According to the algorithm, k adjacent samples of the missing data samples are selected according to the characteristic of high similarity between adjacent data of the samples, and the missing data is corrected by using the number of bits of the k adjacent samples. The use of median values also enables us to deal with bad data. The euclidean distance is used to determine the distance between sample points as shown in the following equation:
in the formula Xi=(xi1,xi2,…,xim) Representing the first m-dimensional data of sample i, element xirRepresenting the attribute of the r-th dimension of sample i. The anomalous data in the sample will be represented by the mean absolute deviation described belowThe index (MAD) was subjected to hypothesis testing as shown below:
wherein MAD is mean (| x)i-xmedian|). With the robust statistical test method based on MAD, we can identify bad data, and these identified data will be replaced by the values obtained by the k-nearest neighbor algorithm. Meanwhile, because the types of the operation data of the power grid are more, the dimensions of the operation data are different and distributed, and the operation data need to be standardized firstly. Suppose the operating data obtained from the grid is X ═ X (X)1,x2,x3,…,xm) And has been subjected to a standardization process. The normalization process is shown below:
and S31, adopting PCA to select principal component variables meeting the requirement of the topology identification accuracy, and simultaneously reducing the noise of the node voltage data and the data dimension.
The main characteristics in the measured data are selected in a self-adaptive mode by applying a principal component analysis method, redundant information of the measured data is removed, and the learning efficiency of the DBN can be effectively improved. The principle component variable mode which meets the requirement of the topological identification accuracy is selected by adopting PCA as follows:
Z=XXT
Z=λω
Y=Xω
wherein Z is a covariance matrix of the power grid operation data X, and the eigenvalue and eigenvector of the covariance matrix Z are respectively lambda ═ lambda1,λ2,…,λm) And ω ═ ω (ω)1,ω2,…,ωm) And a characteristic value λ1≥λ2≥…≥λmArranged from large to small, and Y is the main component of the sample data.
Calculating the variance contribution rate delta of the kth principal component variablekAnd cumulative varianceThe contribution φ is shown as:
selecting proper magnitude of the cumulative variance contribution rate according to actual requirements, determining the number of the principal components F according to the set cumulative variance contribution rate, and expressing the selected principal component set as F ═ { F ═ F1,f2,f3,…,fpAnd p is the number of the selected principal components.
S32, constructing a DBN network, and applying the DBN to train and learn the nonlinear relation between the node voltage of the sample and the network topology.
Constructing a DBN network, and applying the DBN to train and learn the nonlinear relation mode of the node voltage of the sample and the network topology as follows:
the DBN is a probabilistic generative model composed of multiple Restricted Boltzmann Machine (RBM) stacks. Each RBM comprises a visual layer V and an implied layer H, the neurons of the visual layer and the implied layer are connected, and the neurons in the visual layer and the neurons in the implied layer are not connected. The joint probability, joint probability density, activation probability of nerve units and model parameters of the visual layer and the hidden layer in the RBM in the DBN model are shown as follows:
wherein E (v, h | theta) represents the joint probability of the visible layer and the hidden layer in the RBM, P (v, h | theta) is the joint probability density, and P (h | theta)j1| v) is the activation probability of the cell, L (θ) is the maximum likelihood function of the model parameter θ, viRepresenting the status of the nodes of the visual layer, hjRepresenting the state of the hidden layer node, aiAnd bjRespectively representing the corresponding bias values, w, of the nodes of the visual layer and the hidden layerijAnd representing the connection weight between the visual layer and the hidden layer, wherein p is the number of the main components.
The parameter θ is updated using a Nesterov acceleration adaptive moment estimation (Nadam) optimizer. The adaptive moment estimation (Adam) algorithm is improved by a Nesterov Acceleration Gradient (NAG) algorithm. Adam's update rules are as follows:
mt=β1mt-1+(1-β1)gt
where eta is the learning rate, t is the number of updates, gtIs a gradient, mtIs an estimate of the first moment of the gradient, vtIs an estimate of the second moment of the gradient,andcorrecting the estimate, beta, for deviations in the momentum vector1,β2And epsilon is a correction parameter,the estimate is corrected for the current momentum vector bias.
The structure of the DBN is mainly determined by the number of neurons per layer and the number of hidden layers. The scale and the precision of the deep belief network are determined by the number of hidden layer neurons and the number of hidden layer neurons, and too many hidden layer neurons can cause the problems of overlarge DBN scale, large calculation amount, long model training time and the like; while too few hidden layer neurons reduce prediction accuracy. The invention determines the structure of the DBN by a grid search method.
And S33, stopping training after reaching the set standard, and storing the PCA-DBN coupling model after training.
The PCA-DBN coupling model training setting standard adopts reconstruction errors to carry out quantitative analysis, and the definition and the judgment standard of the reconstruction errors are shown as the following formula:
where n is the number of samples, pr is the calculated value of the network, d is the true value,and calculating an average value of the network, wherein delta is a preset reconstruction error value, and L is the number of layers of the hidden layer.
And finally, carrying out observability analysis on the topology of the active power distribution network under different mixed observation stationing schemes by using a PCA-DBN deep learning method, and determining the optimal mixed observation stationing planning scheme by taking the complete observability of the active power distribution network as a target.
And S34, inputting the real-time measurement data into the trained PCA-DBN coupling model topology recognizer for power distribution network topology identification analysis.
As shown in fig. 1, an IEEE33 node standard distribution network is described as an example. The types of loads in the system are all ZIP loads, and the proportion of constant power, constant impedance and constant current loads is 50%, 30% and 20% respectively. Three photovoltaic power stations with a rated capacity of 0.3MW are placed at node 17, node 21 and node 32. 350 topological structures meeting the running characteristics of radial, ring-free and island-free operation of the power distribution network are selected in total, 35000 data samples are formed, and the sample data comprises node voltage amplitude, phase angle and injection power.
Firstly, according to the proposed decision tree optimization model, the importance of the node characteristics of the power distribution network is analyzed and calculated, and the importance percentage of the node characteristics is shown in fig. 2. According to the graph result, the amplitude and the phase angle difference of the node voltage have large influence on the observability of the power distribution network topological structure, and the injection power of the node has small influence on the observability of the power distribution network topological structure. From the analysis of the operation angle of the power system, the voltage and the reactive power of the power grid are strongly related, and the phase angle and the active power are strongly related, so that the operation topology of the power grid can be reflected by selecting two variables of the node voltage amplitude and the phase angle, and the topology observability of the power distribution network is realized.
Further analysis, the characteristic importance indexes of each node are sequentially arranged from large to small according to the importance of the node, as shown in fig. 3, it can be seen from the graph that the difference of the characteristic importance of different nodes is large, for example, the importance of the voltage amplitude and the phase angle of the 0 th node is 0, because the node 0 is a balanced node, and the voltage amplitude and the phase angle are not changed; secondly, the importance of the voltage amplitude characteristic and the voltage phase angle characteristic of each node are different, for example, the importance of the voltage amplitude characteristic of the node 7 is smaller than that of the phase angle, and the importance of the voltage amplitude characteristic of the node 21 and the node 18 is larger than that of the phase angle.
The number of nodes is reduced from all the nodes according to the descending order of the importance degree of the nodes, the measured data of the corresponding nodes are selected to form a training sample set, the training sample set is input into a PCA-DBN coupling topology recognition model and other machine learning methods, the influence of the distribution position and the distribution number of the measuring device on the topology observability of the power distribution network is analyzed, and the comparison result is shown in FIG. 4. The comparison shows that the PCA-DBN model requires the least number of the arranged measuring devices under the condition that the topology of the power distribution network is completely considerable, and when the number of the arranged measuring devices is more than 11, the identification rate of the topology of the power distribution network is stabilized to be more than 99%, and the topology of the power distribution network can be considered to be completely considerable at the moment; when the number of the installed measuring devices is less than 11, the descending range of the topology identification rate of the power distribution network is large.
The distribution quantity and the topology observability of the power distribution network measuring device are considered comprehensively, the first 11 nodes with the maximum node importance degree are selected as the optimal distribution positions of the high-precision PMU measuring device of the IEEE33 node system, and the selected nodes are combined into Pcase33I.e., {7,21,18,17,24,4,3,6,9,11,32 }. Further, the voltage and phase angle data of the nodes 7, 17, 3,6,9,11 and 32 and the voltage data of the nodes 21,18, 24 and 4 are selected as the input of the model, the topology of the power distribution network still can reach 98% of the accuracy of topology identification, and the observability of the topology of the power distribution network is ensured. Considering both economy and topology observability, the optimal configuration scheme is to arrange high-precision PMU measuring devices capable of measuring node voltages and phase angles only at nodes 7, 17, 3,6,9,11 and 32, and arrange voltage mutual inductance devices PT with lower price at nodes 21,18, 24 and 4, as shown in FIG. 5.
The foregoing is illustrative of the preferred embodiments of this invention, and it is to be understood that the invention is not limited to the precise form disclosed herein and that various other combinations, modifications, and environments may be resorted to, falling within the scope of the concept as disclosed herein, either as described above or as apparent to those skilled in the relevant art. And that modifications and variations may be effected by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (5)
1. A power distribution network hybrid observation point distribution method based on deep learning and decision tree driving is characterized by comprising the following steps:
s1, obtaining an active power distribution network running state offline sample T;
s2, constructing a decision tree model based on the off-line samples of the running state of the active power distribution network, inputting the off-line samples obtained in S1 into the decision tree model for analysis and calculation to obtain the feature importance of the nodes, arranging the feature importance according to the descending order, selecting the measured data of the nodes corresponding to the first n feature importance of the sequence, and forming an alternative mixed observation stationing sample set;
s3, inputting the alternative mixed observation stationing sample set obtained in the step S2 into a PCA-DBN coupling topology recognition model, analyzing the observability of the topology of the active power distribution network, judging whether the topology of the active power distribution network is completely observable, if the topology of the active power distribution network is completely observable, entering the step S4, if the topology of the active power distribution network is not completely observable, selecting the measurement data of the nodes corresponding to the first n +1 feature importance degrees of the sequence to form an alternative mixed observation stationing sample set, and inputting the alternative mixed observation stationing sample set into the PCA-DBN coupling topology recognition model again;
and S4, obtaining a mixed observation point distribution scheme.
2. The deep learning and decision tree driving-based hybrid observation and stationing method for the power distribution network according to claim 1, wherein the process of obtaining the offline samples of the operation status of the active power distribution network in S1 is as follows:
s10, selecting a power distribution network operation topological structure set G meeting three operation characteristics of radial, loop-free and island-free of an active power distribution networkT=(GT1,GT2,…,GTm);
S11, from the feasible topology set GTIn select topology GTa;
S12, based on feasible topology GTaGenerating different active power distribution network operation scenes according to a typical daily load fluctuation curve and a photovoltaic output curve and considering photovoltaic and ZIP loads;
and S13, calculating load flow distribution under different active traveling scenes, and recording the active distribution network operation state data to form an active distribution network operation state offline sample, wherein the offline sample comprises active power P of each node of the active distribution network, node voltage V and voltage phase angle difference theta as sample inputs, and correspondingly input distribution network tie switch and section switch states as sample outputs.
3. The deep learning and decision tree driving based power distribution network hybrid observation stationing method according to claim 1, wherein the step of constructing the decision tree model in S2 comprises the steps of:
carrying out supervised learning on an active power distribution network running state offline sample T which takes X as an input variable and Y as an output variable, wherein the active power distribution network running state offline sample is as follows: t { (x)1,y1),(x2,y2),…,(xn,yn)};
The mathematical expression of the kini coefficient of the active power distribution network running state off-line sample T is as follows:
the active power distribution network running state offline sample is provided with K categories CkThe number of samples of the Kth category;
the decision tree model is a binary decision tree recursively constructed from a root node through an active power distribution network running state offline sample based on a minimum criterion of a kini coefficient, and the decision tree model is established in the following process:
s20, inputting an active power distribution network running state off-line sample T and a Gini coefficient threshold value gini;
S21, calculating the running state of the active power distribution network to be off-lineEach characteristic of the samples and the loss function of the divided nodes are selected randomly, and the jth characteristic x of the active power distribution network operation state offline sample T is selected randomlyjCut variable as sample and xjThe value s is used as a dividing node of the sample, and the off-line sample T of the running state of the active power distribution network is divided into two subdata sets Ts1And Ts2The mathematical expression of the partition principle is as follows:
the loss function is divided into two sub data sets Ts1And Ts2The mean square error of (a), the corresponding feature segmentation variable x when the loss function reaches the set minimum valuejAnd a dividing node s, the calculation formula of which is as follows:
in the formula c1Representing a sub data set Ts1Average value of output, c2Representing a sub data set Ts2Average value of the output, wherein
S22, respectively calculating the damping coefficients of the two division nodes, and judging whether the damping coefficients of the two division nodes are smaller than a damping coefficient threshold value giniWhen the damping coefficients of the two divided nodes are not less than the damping coefficient threshold value giniIf so, the two divided nodes respectively return to the step S21 to continue the node recursive division; when the Gini coefficient of one of the divided nodes is smaller than the threshold value g of the Gini coefficientiniThen the node stops the node recursive partitioning, and the node is not smaller than the threshold value g of the Gini coefficientiniReturns to step S21 to continue the recursive division of the nodes until recursion to the kiney system of another divided nodeNumber less than the threshold value g of the kini coefficientiniObtaining K subspaces of all divided nodes;
s23, dividing the input active power distribution network operation state offline samples into K subspaces, wherein each subspace comprises partial sample data and an average value c of subspace output valuesKAnd obtaining a decision tree model, wherein the mathematical expression of the decision tree model is as follows:
in the formula I (x is equal to T)sK) An indicator function representing the model.
4. The deep learning and decision tree driving based hybrid observation point distribution method for the power distribution network according to claim 1, wherein the offline samples obtained in step S1 are input into the decision tree model for analysis and calculation in the following manner in step S2:
in the formula, N represents the total number of off-line samples of the running state of the active power distribution network, and NtNumber of offline samples, N, representing active distribution network operating state of current nodetRRepresenting the left sub-tree, N, of the current nodetLThe number of the active power distribution network running state offline samples of the left sub-tree of the current node is represented, H represents the purity of the current node, and H represents the purity of the current noderightRepresenting the left sub-tree of the current node, HleftIndicating the degree of purity of the left sub-tree of the current node.
5. The deep learning and decision tree driving-based power distribution network hybrid observation stationing method according to claim 1, wherein the step S3 is implemented by inputting the candidate hybrid observation stationing sample set obtained in the step S2 into a PCA-DBN coupled topology recognition model, and the process of analyzing the observability of the topology of the active power distribution network is as follows:
s30, preprocessing the alternative mixed observation stationing sample set;
s31, selecting principal component variables meeting the requirement of topology identification accuracy by adopting PCA, and simultaneously reducing the noise and data dimensionality of node voltage data;
s32, constructing a DBN network, and applying the DBN to train and learn the nonlinear relation between the node voltage of the sample and the network topology;
s33, after the training is finished, storing the PCA-DBN coupling model after the training;
and S34, inputting the preprocessed alternative mixed observation stationing sample set into the trained PCA-DBN coupling model topology recognizer for power distribution network topology identification analysis, and obtaining an analysis result of the topology observability of the active power distribution network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110617672.7A CN113496255B (en) | 2021-05-31 | 2021-05-31 | Power distribution network mixed observation point distribution method based on deep learning and decision tree driving |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110617672.7A CN113496255B (en) | 2021-05-31 | 2021-05-31 | Power distribution network mixed observation point distribution method based on deep learning and decision tree driving |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113496255A true CN113496255A (en) | 2021-10-12 |
CN113496255B CN113496255B (en) | 2023-04-25 |
Family
ID=77997653
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110617672.7A Active CN113496255B (en) | 2021-05-31 | 2021-05-31 | Power distribution network mixed observation point distribution method based on deep learning and decision tree driving |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113496255B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114692513A (en) * | 2022-05-26 | 2022-07-01 | 国网山西省电力公司晋城供电公司 | New energy bearing capacity assessment method and early warning method based on deep learning |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111861256A (en) * | 2020-07-30 | 2020-10-30 | 国网经济技术研究院有限公司 | Active power distribution network reconstruction decision method and system |
US20200373726A1 (en) * | 2018-02-05 | 2020-11-26 | The University Of Sussex | Method and system for nonlinear optical process optimization via temporal pulse splitting |
CN115000956A (en) * | 2022-07-18 | 2022-09-02 | 南京工程学院 | Power distribution network topology second-order identification method based on data model joint driving |
-
2021
- 2021-05-31 CN CN202110617672.7A patent/CN113496255B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200373726A1 (en) * | 2018-02-05 | 2020-11-26 | The University Of Sussex | Method and system for nonlinear optical process optimization via temporal pulse splitting |
CN111861256A (en) * | 2020-07-30 | 2020-10-30 | 国网经济技术研究院有限公司 | Active power distribution network reconstruction decision method and system |
CN115000956A (en) * | 2022-07-18 | 2022-09-02 | 南京工程学院 | Power distribution network topology second-order identification method based on data model joint driving |
Non-Patent Citations (2)
Title |
---|
YOUBO LIU 等: "Real-time Topology Estimation for Active Distribution System using Graph-bank Tracking Bayesian Networks" * |
李洋麟: "基于数据驱动的电力***小干扰稳定评估与预警" * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114692513A (en) * | 2022-05-26 | 2022-07-01 | 国网山西省电力公司晋城供电公司 | New energy bearing capacity assessment method and early warning method based on deep learning |
Also Published As
Publication number | Publication date |
---|---|
CN113496255B (en) | 2023-04-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107992976B (en) | Hot topic early development trend prediction system and prediction method | |
CN115270965B (en) | Power distribution network line fault prediction method and device | |
CN106055918A (en) | Power system load data identification and recovery method | |
CN110766137A (en) | Power electronic circuit fault diagnosis method based on longicorn whisker optimized deep confidence network algorithm | |
CN106971240A (en) | The short-term load forecasting method that a kind of variables choice is returned with Gaussian process | |
CN110929847A (en) | Converter transformer fault diagnosis method based on deep convolutional neural network | |
CN112508442B (en) | Transient stability assessment method and system based on automatic and interpretable machine learning | |
CN111768000A (en) | Industrial process data modeling method for online adaptive fine-tuning deep learning | |
CN112557034B (en) | Bearing fault diagnosis method based on PCA _ CNNS | |
CN110888058A (en) | Algorithm based on power battery SOC and SOH joint estimation | |
CN111680726B (en) | Transformer fault diagnosis method and system based on neighbor component analysis and k neighbor learning fusion | |
CN113702895A (en) | Online quantitative evaluation method for error state of voltage transformer | |
CN111401599A (en) | Water level prediction method based on similarity search and L STM neural network | |
CN113919545A (en) | Photovoltaic power generation power prediction method and system with integration of multiple data models | |
CN113780420B (en) | GRU-GCN-based method for predicting concentration of dissolved gas in transformer oil | |
CN111680875A (en) | Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model | |
CN116842459B (en) | Electric energy metering fault diagnosis method and diagnosis terminal based on small sample learning | |
CN115860177A (en) | Photovoltaic power generation power prediction method based on combined machine learning model and application thereof | |
CN115600729A (en) | Grid load prediction method considering multiple attributes | |
CN113850320A (en) | Transformer fault detection method based on improved support vector machine regression algorithm | |
CN116050605A (en) | Power load prediction method based on neural network and random forest method | |
CN113884807B (en) | Power distribution network fault prediction method based on random forest and multi-layer architecture clustering | |
CN113496255B (en) | Power distribution network mixed observation point distribution method based on deep learning and decision tree driving | |
Ullah et al. | Adaptive data balancing method using stacking ensemble model and its application to non-technical loss detection in smart grids | |
CN113298296A (en) | Method for predicting day-ahead load probability of power transmission substation from bottom to top |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |