CN112561303A - Power system dynamic analysis method based on integrated learning and power grid topological change - Google Patents

Power system dynamic analysis method based on integrated learning and power grid topological change Download PDF

Info

Publication number
CN112561303A
CN112561303A CN202011455569.9A CN202011455569A CN112561303A CN 112561303 A CN112561303 A CN 112561303A CN 202011455569 A CN202011455569 A CN 202011455569A CN 112561303 A CN112561303 A CN 112561303A
Authority
CN
China
Prior art keywords
topological structure
parameters
training
power
limit
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
Application number
CN202011455569.9A
Other languages
Chinese (zh)
Inventor
陈磊
刘显壮
傅一苇
徐飞
胡伟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Original Assignee
Tsinghua University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tsinghua University filed Critical Tsinghua University
Priority to CN202011455569.9A priority Critical patent/CN112561303A/en
Publication of CN112561303A publication Critical patent/CN112561303A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Evolutionary Computation (AREA)
  • Marketing (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Educational Administration (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Development Economics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Primary Health Care (AREA)
  • Water Supply & Treatment (AREA)
  • Public Health (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Supply And Distribution Of Alternating Current (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a dynamic analysis method of an electric power system based on integrated learning and power grid topological change, which comprises the following steps: collecting current tide parameters and topological structure parameters of a power grid to be tested; inputting the power flow parameters and the topological structure parameters into a prediction model, and outputting a limit cutting time prediction value of the power grid to be tested; the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed. The method provided by the invention realizes that the accuracy of the safety analysis is improved while the calculation burden of the dynamic safety analysis of the power system is reduced.

Description

Power system dynamic analysis method based on integrated learning and power grid topological change
Technical Field
The invention relates to the technical field of electric power system safety analysis, in particular to an electric power system dynamic analysis method based on integrated learning and power grid topology change.
Background
In the actual operation of the power grid, the situation of topology change often exists, however, this is not considered in a general dynamic security analysis model, and this brings a certain problem to the practicability of the model. A separate data-driven model can be established for each topology, however, the method is time-consuming and labor-consuming, and the implicit connection and information between the topologies are omitted.
Generally, in the power flow description of the power system, the values of the element variables are continuous, and we refer to the continuous quantities. However, in practical situations, another part of the variables is needed to describe the power grid, i.e. the topological quantity of the power grid. The topology of the grid refers to the quantities used to describe the topology of the grid, i.e. the discrete changes in the grid, which are referred to as changes in the topology of the grid. The most typical topological quantity is the disconnection condition of each element in the power grid, and the topological quantity of the power grid is described as a variable from 0 to 1.
For the case that a certain element is disconnected, for example, a certain circuit is disconnected, it can be completely equivalent to 0 both active and reactive on the circuit in terms of power flow. However, in dynamic security analysis, it is not equally equivalent. This is because the line disconnection has an influence on the structure and parameters of the system in addition to the initial power flow. That is, the differential algebraic equations describing the system transient state, in addition to changing the initial values of the equations, also change the structure and parameters of the equations. Thus, the topological volume of the system generally has a greater impact on the dynamic security of the system than the continuous volume.
In the existing data driving method for dynamic security analysis, the consideration of system topology changes is mainly divided into the following two ways: (1) the case of topology changes is not considered. For topology changes that occur in practice, a new model is retrained each time a new topology occurs. This approach has the following disadvantages: firstly, the model is rough, because there are often situations where the amount of topology data is small, and it is easy for the model to be overfitting. Secondly, computing resources are consumed comparatively, on one hand, the computation amount in the training process needs to be doubled (related to the topological quantity), on the other hand, the required sample quantity can be greatly increased, and the samples are obtained through off-line simulation, so that the computation burden can be increased, and thirdly, the analysis of the relation between the topological quantity and the dynamic security of the power grid is inconvenient; (2) the way statistics are taken as input. The model uses statistical indexes of certain variables as input quantities, such as the average value and variance of the load in a certain area. Therefore, the influence of the topology change of the power grid is reflected in the change of the statistical index. The method is used for migration learning, for example, a model of a small system migrates to a large system, so that the consideration of the topology is rough for a specific power grid.
Therefore, how to avoid the heavy calculation burden caused by the participation of a large number of sample topological structure parameters in the existing dynamic safety analysis method for the power system, or how to accurately implement safety analysis because the output safety index is also a statistical value due to the fact that the topological structure is converted into the statistic value as the input is still a problem to be solved by the technical staff in the field.
Disclosure of Invention
The invention provides a dynamic analysis method of an electric power system based on integrated learning and power grid topological change, the method is used for solving the problem that the existing power system dynamic security analysis method is heavy in calculation burden caused by the fact that a large number of sample topological structure parameters participate in training, or the output safety index is also a statistical value due to the fact that the topological structure is converted into the statistical quantity to be used as the input, so that the safety analysis cannot be accurately realized, the accuracy of the dynamic safety analysis is ensured through a prediction model trained by a large number of samples, meanwhile, the type of the topological structure is combined in the training process, and the mechanism of the ensemble learning solves the problem that the calculated amount is too large due to excessive parameters of a neural network needing to be adjusted caused by the fact that a large number of sample topological structure parameters need to be considered in model training, so that the calculation burden of dynamic safety analysis of the power system is reduced, and the accuracy can be improved.
The invention provides a dynamic analysis method of an electric power system based on integrated learning and power grid topological change, which comprises the following steps:
collecting current tide parameters and topological structure parameters of a power grid to be tested;
inputting the power flow parameters and the topological structure parameters into a prediction model, and outputting a limit cutting time prediction value of the power grid to be tested;
the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed.
According to the dynamic analysis method of the power system based on the integrated learning and the power grid topological change, the relation between the input sample power flow parameter and the topological structure parameter, the output limit excision time predicted value and the existing training reference set is established based on the Gaussian kernel function constructed by the horse-type distance in the prediction model training process.
According to the dynamic analysis method of the power system based on the integrated learning and the power grid topological change, the Gaussian kernel function constructed based on the Mammai distance establishes the relationship between the input sample power flow parameter and the topological structure parameter, the output limit excision time predicted value and the existing training reference set, and specifically comprises the following steps:
for any combined topological structure type, constructing a relation between a sub-limit excision time predicted value, an input sample power flow parameter, a reference power flow parameter corresponding to any combined topological structure type in the existing training reference set and reference limit excision time under the topological structure type;
constructing a weighting coefficient of the predicted value of any sub-limit cutting time, the combined topological structure type, the input sample power flow parameter, the input sample topological structure and the relation between the reference power flow parameter and the reference limit cutting time in the existing training reference set on the basis of a Gaussian kernel function constructed by the Markov distance;
and determining an output limit excision time predicted value based on all the sub-limit excision time predicted values and the weighting coefficients of all the sub-limit excision time predicted values.
According to the dynamic analysis method of the power system based on the integrated learning and the power grid topological change, for any combined topological structure type, a relation between a predicted value of the sub-limit cutting time, an input sample power flow parameter, a reference power flow parameter corresponding to any combined topological structure type in the existing training reference set and the reference limit cutting time is constructed, and the method specifically comprises the following steps:
determining the predicted value of the sub-limit excision time under any topological structure type s through the following formula
Figure BDA0002828628800000041
Figure BDA0002828628800000042
Figure BDA0002828628800000043
Wherein x is the input sample power flow parameter, x is the vector of D dimension,
Figure BDA0002828628800000044
for the reference power flow parameter matrix X in the existing training reference setTRAINThe (c) th row of (a),
Figure BDA0002828628800000045
vector of dimension D, XTRAINA matrix, X, formed by all the reference tide parameters in the existing training reference setTRAINSize NTRAIN×D，NTRAINFor the number of the reference tidal current parameters in the existing training reference set,
Figure BDA0002828628800000046
for the existing training reference set
Figure BDA0002828628800000047
Corresponding reference limit excision time, ΩsIs XTRAINIn the merged topological structure type, the number of rows of all original topological structures in the S-th topological structure type is collected, γ (S) is a parameter of a scalar to be adjusted during the training of the prediction model, m (S) is a parameter of a matrix to be adjusted during the training of the prediction model, the size of m (S) is DxD, S is the total number of the merged topological structure types, and r is 1,2, … and S;
correspondingly, the constructing of the weighting coefficient of the predicted value of any sub-limit resection time, the combined topological structure type, the input sample power flow parameter, the input sample topological structure, and the relationship between the reference power flow parameter in the existing training reference set and the reference limit resection time by the gaussian kernel function constructed based on the marching distance specifically includes:
determining the predicted value of any one sub-limit excision time by the following formula
Figure BDA0002828628800000051
Is weighted by a weight coefficient Ps(x,t)：
Figure BDA0002828628800000052
Figure BDA0002828628800000053
Figure BDA0002828628800000054
Wherein t is an input sample topological structure parameter corresponding to the input sample power flow parameter x, the elements of the input sample topological structure parameter are variables of 0 to 1, t is a vector of W dimension, and W is the vector of W dimension<D，Ts uniAnd Tr uniRepresenting merged topological junctionsThe s-th topological structure parameter vector and the r-th topological structure parameter vector in the structure type have the dimensionality of W and gamma0And gamma(s) are parameters of the scalar to be adjusted during the training of the prediction model, M0And M(s) are all the parameters of the matrix to be adjusted during the training of the prediction model, M0Is W × W, m (S) is D × D, S is the total number of the merged topology classes, r is 1,2, …, S;
the determining the output limit resection time predicted value based on the all sub-limit resection time predicted values and the weighting coefficients of all sub-limit resection time predicted values specifically comprises:
determining an output limit cut time prediction value by the following formula
Figure BDA0002828628800000057
Figure BDA0002828628800000058
Wherein S is the total number of the combined topological structure types,
Figure BDA0002828628800000059
for sub-limit excision time prediction, P, under any topological structure type ss(x, t) is the predicted value of any one of the sub-limit excision time
Figure BDA00028286288000000510
S is 1,2, …, S.
According to the dynamic analysis method of the power system based on the integrated learning and the power grid topological change, the power flow parameters and the topological structure parameters are input into a prediction model, the predicted value of the ultimate removal time of the power grid to be tested is output, and the prediction model carries the existing training reference set after the training is finished, and the method specifically comprises the following steps:
determining the input power flow parameter and the input topological structure parameter as x respectivelyuAnd tq；
Determining tqBelonging to the merged S topology nodeT of u species in structural speciesu，1≤u≤S；
Calculating and outputting a predicted value of the limit cutting time of the power grid to be tested according to the following formula
Figure BDA0002828628800000061
Figure BDA0002828628800000062
Figure BDA0002828628800000063
Figure BDA0002828628800000064
Figure BDA0002828628800000065
Wherein x isuVector of dimension D, tuIs a vector of dimension W. t is tuIs a variable of 0 to 1, W<D，Ts uniAnd Tr uniRepresents the s-th and r-th topological structure parameter vectors gamma in the combined topological structure types0 done、M0 done、γ(s)doneAnd M(s)doneRespectively corresponding to the parameters gamma to be adjusted during the training of the prediction model0、M0Final adjustment parameters after training of γ(s) and M(s) is completed,
Figure BDA0002828628800000068
for the reference power flow parameter matrix X in the existing training reference setTRAINThe (c) th row of (a),
Figure BDA0002828628800000069
vector of dimension D, XTRAINFormed for all reference tide parameters in the existing training reference setMatrix, XTRAINSize NTRAIN×D，NTRAINFor the number of the reference tidal current parameters in the existing training reference set,
Figure BDA00028286288000000610
for the existing training reference set
Figure BDA00028286288000000611
Corresponding reference limit excision time, ΩsIs XTRAINSet of rows of all original topologies in the s-th category, T, of the merged topology categorys uniAnd Tr uniRepresenting the parameter vectors of the s-th topological structure and the r-th topological structure in the combined topological structure types, wherein the dimensions are W and gamma0 doneAnd gamma(s)doneAre all scalar quantities, M0 doneThe size of (a) is W × W, M(s)doneThe size of (D) is D × D, S is the total number of the combined topology types, and r is 1,2, …, S.
According to the dynamic analysis method of the power system based on the integrated learning and the power grid topological change, provided by the invention, the loss function in the prediction model training process is formed based on the mean square error between the output limit excision time prediction value and the corresponding limit excision time label.
According to the dynamic analysis method of the power system based on the integrated learning and the topological change of the power grid provided by the invention,
elements in the power flow parameters comprise a real part and an imaginary part of any bus voltage in the power grid to be tested, power injected from a delta end on any alternating current line mu from delta to rho, power injected from a delta end on any transformer Q from delta to rho, charging power from a delta end on any alternating current line mu from delta to rho, active power of a generator connected with any node, reactive power of a generator connected with any node, active power of a load connected with any node and reactive power of a load connected with any node;
the elements in the topological structure parameters comprise the switching-in and switching-off condition of any bus in the power grid to be tested, the switching-off condition of any alternating current line mu from delta to rho, the switching-off condition of any transformer Q from delta to rho, the switching-on and switching-off condition of a generator connected with any node and the connection and disconnection condition of any node.
The invention also provides a dynamic analysis device of the power system based on integrated learning and power grid topology change, which comprises the following components:
the acquisition unit is used for acquiring the current tidal current parameter and the topological structure parameter of the power grid to be detected;
the calculation unit is used for inputting the power flow parameters and the topological structure parameters into a prediction model and outputting a predicted value of the limit removal time of the power grid to be tested;
the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed.
The invention also provides an electronic device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the program to realize the steps of the power system dynamic analysis method based on the ensemble learning and the power grid topology change.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the method for integrated learning and grid topology change based dynamic analysis of an electrical power system as described in any of the above.
The dynamic analysis method of the power system based on the integrated learning and the power grid topological change comprises the steps of collecting current tide parameters and topological structure parameters of a power grid to be tested; inputting the power flow parameters and the topological structure parameters into a prediction model, and outputting a limit cutting time prediction value of the power grid to be tested; the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed. Because the prediction model is obtained by training based on the sample power flow parameters, the sample topological structure parameters and the corresponding limit excision time labels on the basis of the existing training reference set, the accuracy of the trained prediction model can be ensured by ensuring that the sample and the existing training reference set have larger data volume, and the problem of too large calculated amount caused by excessive parameters of a neural network needing to be adjusted due to the fact that a large number of sample topological structure parameters need to be considered in the model training process is solved by the integrated learning mechanism, so that the calculated amount of dynamic safety analysis of the power system is reduced. Therefore, the method provided by the invention can reduce the calculation burden of dynamic safety analysis of the power system and improve the accuracy of the safety analysis.
Drawings
In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
FIG. 1 is a schematic flow chart of a dynamic analysis method for an electric power system based on integrated learning and power grid topology change according to the present invention;
FIG. 2 is a flowchart of the predictive model training process provided by the present invention;
FIG. 3 is a schematic structural diagram of an electric power system dynamic analysis device based on integrated learning and power grid topology change according to the present invention;
fig. 4 is a schematic physical structure diagram of an electronic device provided in the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments 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.
The existing dynamic safety analysis method for the power system generally has the problems that the calculation burden is heavy due to the fact that a large number of sample topological structure parameters participate in training, or the safety analysis cannot be accurately realized due to the fact that the topological structure is converted into statistic to serve as an input, and the output safety index is also a statistic value. The following describes a power system dynamic analysis method based on integrated learning and grid topology change according to the present invention with reference to fig. 1 and 2. Fig. 1 is a schematic flow chart of a power system dynamic analysis method based on integrated learning and grid topology change, as shown in fig. 1, the method includes:
and 110, collecting current power flow parameters and topological structure parameters of the power grid to be tested.
Specifically, safety analysis is performed on the power system, that is, a limit cut-off Time (CCT) in the power system is determined, which is also called a fault limit cut-off Time and is a necessary standard for measuring a transient voltage stability margin, and calculating the CCT under each disturbance condition of the system is a key and core of transient voltage stability evaluation of the system. The invention takes all factors influencing the CCT of the power system into consideration, so the acquired data source comprises the tide parameters of the power grid in the power system and the topological structure parameters of the power grid. It should be noted here that the power flow parameter is determined according to the injection power and the output power of each bus or feature node in the power grid, and is a vector composed of continuous numerical values, the dimension of the vector is determined according to the number of the bus and the node to be investigated, and the topological structure parameter is determined according to the switching-in and switching-off condition of each bus or the power-on and power-off condition of the feature node in the power grid, so that the vector is composed of discrete numerical values, and is usually a 0-1 numerical value, where a numerical value of 0 indicates the switching-in and switching-off condition, a numerical value of 1 indicates the switching-in and power-on condition, and the dimension of the vector is determined according to the number of the.
Step 120, inputting the power flow parameters and the topological structure parameters into a prediction model, and outputting a predicted value of the limit removal time of the power grid to be tested;
the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed.
Specifically, the power flow parameters and the topological structure parameters collected in step 110 are input into the prediction model, and the predicted value of the limit cutting time of the power grid to be tested is output. The prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit excision time label on the basis of an existing training reference set, the sample power flow parameter, the sample topological structure parameter and the corresponding limit excision time label are extracted on the basis of historical data relative to the current moment, the existing training reference set comprises a real power flow parameter, a real topological structure parameter and corresponding real limit excision time as the existing training reference set in the training process, a large number of sample power flow parameters, sample topological structure parameters and corresponding limit excision time labels are trained on the basis of the existing training reference set of the test sample set, and the existing training reference set is required for the training of the prediction model because the conventional prediction model training method provided by the invention does not directly use samples to train the prediction model by using a conventional prediction network However, the prediction network calculates the relevance between the input sample and other reference data in the sample training process to calculate the output prediction value, the relevance is determined by calculating the distance, the distance calculation method is not specifically limited here and can be Euclidean distance, Manhattan distance and the like, and then the relevance between the distance calculation method and each reference data is used as a weighting coefficient to obtain the final prediction value output by the prediction network based on the probability statistics idea. After the prediction model is trained, when the prediction model is put into use, an existing training reference set used in the training process of the prediction model also needs to be carried, because when the prediction model is actually used, collected source data is input, the output prediction value is obtained by calculating through a calculation process of the prediction network during training, the correlation between the input source data and other reference data is calculated in the calculation process, then the correlation between the prediction model and each reference data is used as a weighting coefficient to obtain the final prediction value output by the prediction network based on a probability statistics idea, and the only difference is that the parameter to be adjusted in the prediction network is adjusted, optimized and changed into a fixed value. The types of the topological structures are combined in the training process, and the problem of too large calculated amount caused by too many parameters of the neural network needing to be adjusted due to the fact that a large number of sample topological structure parameters need to be considered in model training can be solved by adopting the integrated learning mechanism, so that the calculated amount of dynamic safety analysis of the power system is reduced.
The dynamic analysis method of the power system based on the integrated learning and the power grid topological change comprises the steps of collecting current tide parameters and topological structure parameters of a power grid to be tested; inputting the power flow parameters and the topological structure parameters into a prediction model, and outputting a limit cutting time prediction value of the power grid to be tested; the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed. Because the prediction model is obtained by training based on the sample power flow parameters, the sample topological structure parameters and the corresponding limit excision time labels on the basis of the existing training reference set, the accuracy of the trained prediction model can be ensured by ensuring that the sample and the existing training reference set have larger data volume, and the problem of too large calculated amount caused by excessive parameters of a neural network needing to be adjusted due to the fact that a large number of sample topological structure parameters need to be considered in the model training process is solved by the integrated learning mechanism, so that the calculated amount of dynamic safety analysis of the power system is reduced. Therefore, the method provided by the invention can reduce the calculation burden of dynamic safety analysis of the power system and improve the accuracy of the safety analysis.
On the basis of the above embodiment, the method further includes:
and in the prediction model training process, establishing a relation between an input sample power flow parameter and a topological structure parameter, an output limit excision time prediction value and the existing training reference set based on a Gaussian kernel function constructed by the Mammai distance.
Specifically, in the process of training the prediction model, the corresponding prediction network calculates the relevance between the input sample and other reference data, that is, the relevance between the input sample power flow parameter and the real power flow parameter in the reference set needs to be measured, and a measuring method for specifying the relevance based on a preset standard needs to calculate the relevance between the sub-CCT predicted value corresponding to any sample topological structure parameter and the real CCT in the reference set. In the invention, the Ma-type distance is selected for measurement, the Gaussian kernel constructed based on the Ma-type distance is a variety of the Gaussian kernel function, the distance function is changed from the Euclidean distance in the Gaussian kernel function to the Ma-type distance, so the distance function can be called as the Ma-type kernel function, and the kernel regression constructed by the Ma-type kernel function is called as the Ma-type kernel regression. The reason why the distance between the sample and the real value needs to be calculated by using the mahalanobis kernel function is that the sample and the real value both belong to high-dimensional vectors, a simple euclidean distance is usually used in a two-norm environment to accurately measure the difference between two parameters, and the calculation by using the kernel function for the high-dimensional vectors is more accurate, although the calculation amount is increased due to the high dimensionality of the kernel.
On the basis of the above-described embodiment, in this method,
the method for constructing the relation between the input sample power flow parameter and the topological structure parameter, the output limit excision time predicted value and the existing training reference set based on the Gaussian kernel function constructed based on the Mammai distance specifically comprises the following steps:
for any combined topological structure type, constructing a relation between a sub-limit excision time predicted value, an input sample power flow parameter, a reference power flow parameter corresponding to any combined topological structure type in the existing training reference set and reference limit excision time under the topological structure type;
constructing a weighting coefficient of the predicted value of any sub-limit cutting time, the combined topological structure type, the input sample power flow parameter, the input sample topological structure and the relation between the reference power flow parameter and the reference limit cutting time in the existing training reference set on the basis of a Gaussian kernel function constructed by the Markov distance;
and determining an output limit excision time predicted value based on all the sub-limit excision time predicted values and the weighting coefficients of all the sub-limit excision time predicted values.
Specifically, in the process of processing and calculating an input sample of a prediction network in a prediction model to obtain a CCT prediction value output by the prediction network, because the input parameters are divided into two types, namely, a power flow parameter and a topological structure parameter, and there is no correlation between the two types of parameters, the dimension reduction or dimension increase and then the feature fusion processing cannot be performed together, and only the two types of parameters can be processed separately, therefore, the processing and calculating process is to calculate a calculation method of a sub-limit excision time prediction value under the condition of any combined topological structure type, to obtain a calculation method of a sub-limit excision time prediction value under each topological structure type, and then to determine the importance of any combined topological structure type (i.e. relative to other topological structure types), which is to be the weighting coefficient of the sub-limit excision time prediction value corresponding to any combined topological structure type, namely, the processing and calculating process divides the calculation of two input factor parameters into two layers, wherein the innermost layer is the relation between the predicted sub-CCT under the condition of the limitation of the topological structure type and the real CCT of the topological structure type obtained by calculating the relevance between the sample power flow and the real power flow under the condition of the given topological structure type, the outer layer is the determination of the value of the predicted CCT finally output by the prediction network according to the importance (namely the weighting coefficient) of each flapping structure type after the predicted sub-CCT under the condition of all the flapping structure types is obtained, and the determination method is that the weighted summation is carried out according to the weighting coefficient.
On the basis of the above-described embodiment, in this method,
for any combined topological structure type, constructing a relation between a sub-limit excision time predicted value, an input sample power flow parameter, a reference power flow parameter corresponding to any combined topological structure type in the existing training reference set and a reference limit excision time, and specifically comprising:
determining the predicted value of the sub-limit excision time under any topological structure type s through the following formula
Figure BDA0002828628800000141
Figure BDA0002828628800000142
Figure BDA0002828628800000143
Wherein x is the input sample power flow parameter, x is the vector of D dimension,
Figure BDA0002828628800000144
for the reference power flow parameter matrix X in the existing training reference setTRAINThe (c) th row of (a),
Figure BDA0002828628800000145
vector of dimension D, XTRAINA matrix, X, formed by all the reference tide parameters in the existing training reference setTRAINSize NTRAIN×D，NTRAINFor the number of the reference tidal current parameters in the existing training reference set,
Figure BDA0002828628800000146
for the existing training reference set
Figure BDA0002828628800000147
Corresponding reference limit excision time, ΩsIs XTRAINIn the merged topological structure type, the number of rows of all original topological structures in the S-th topological structure type is collected, γ (S) is a parameter of a scalar to be adjusted during the training of the prediction model, m (S) is a parameter of a matrix to be adjusted during the training of the prediction model, the size of m (S) is DxD, S is the total number of the merged topological structure types, and r is 1,2, … and S;
correspondingly, the constructing of the weighting coefficient of the predicted value of any sub-limit resection time, the combined topological structure type, the input sample power flow parameter, the input sample topological structure, and the relationship between the reference power flow parameter in the existing training reference set and the reference limit resection time by the gaussian kernel function constructed based on the marching distance specifically includes:
determining the predicted value of any one sub-limit excision time by the following formula
Figure BDA0002828628800000148
Is weighted by a weight coefficient Ps(x,t)：
Figure BDA0002828628800000151
Figure BDA0002828628800000152
Figure BDA0002828628800000153
Wherein t is an input sample topological structure parameter corresponding to the input sample power flow parameter x, the elements of the input sample topological structure parameter are variables of 0 to 1, t is a vector of W dimension, and W is the vector of W dimension<D，Ts uniAnd Tr uniRepresenting the parameter vectors of the s-th topological structure and the r-th topological structure in the combined topological structure types, wherein the dimensions are W and gamma0And gamma(s) are parameters of the scalar to be adjusted during the training of the prediction model, M0And M(s) are all the parameters of the matrix to be adjusted during the training of the prediction model, M0Is W × W, m (S) is D × D, S is the total number of the merged topology classes, r is 1,2, …, S;
the determining the output limit resection time predicted value based on the all sub-limit resection time predicted values and the weighting coefficients of all sub-limit resection time predicted values specifically comprises:
determining an output limit cut time prediction value by the following formula
Figure BDA0002828628800000156
Figure BDA0002828628800000157
Wherein S is the total number of the combined topological structure types,
Figure BDA0002828628800000158
for sub-limit excision time prediction, P, under any topological structure type ss(x, t) is the predicted value of any one of the sub-limit excision time
Figure BDA0002828628800000159
S is 1,2, …, S.
Specifically, the above specifically shows a calculation method of each processing flow in the prediction network corresponding to the prediction model during the prediction model training. And x is an input sample power flow parameter, x is a D-dimensional vector, the dimension of the D-dimensional vector is determined by the topological structure of the power grid, and the numerical value of each element in the vector is the injection power or the output power of a corresponding bus or node.
Figure BDA00028286288000001510
For the reference tide parameters in the existing training reference setMatrix XTRAINIs also a vector, XTRAINIs a matrix formed by reference flow parameters (real flow parameters) in an existing training reference set, each row of which represents a real sample (also called a reference sample), NTRAINThe number of the reference tidal current parameters in the existing training reference set is XTRAINIs NTRAINMatrix of x D.
Figure BDA0002828628800000161
And the real limit cutting time (also called reference limit cutting time) CCT corresponding to the w-th real power flow parameter (also called reference power flow parameter). OmegasIs XTRAINThe line number (S) belongs to the S-th topology (S different topology types are shared), and S is 1,2, … and S.
Figure BDA0002828628800000162
Namely, the CCT of (x, t) is predicted by using an existing training reference set under a certain power grid topological structure type s.
t is also a vector with dimension W (W)<D) The specific composition of the elements in the vector is determined by the initial topology of the grid, Ts uniAnd the topological structure vector corresponding to the s-th topological structure type.
Figure BDA0002828628800000164
And kappaT(t，Ts uni) Is the core part of the calculation model constructed by the invention. They are all mahalanobis kernel functions. exp {. is a natural exponential function, γ0γ(s) is a scalar parameter (γ(s) varies with s); m0And m(s) is a matrix parameter (dimensions are W × W, D × D, respectively, and m(s) varies with s), the function is a variation of a gaussian kernel function, and is called a mahalanobis kernel function because a distance function is changed from an euclidean distance in the gaussian kernel function to a mahalanobis distance, and a kernel regression constructed by using the mahalanobis kernel function is called a mahalanobis kernel regression.
In the calculation model, training data under different topologies are utilized to predict CCT under (x, t), and information contained in the data is utilized to the maximum extent. Meanwhile, for training data under different topologies, although the model is constructed by adopting the Markov kernel regression, different model parameters are adopted, so that the model is more accurate. In addition, in order to prevent the overfitting phenomenon caused by too many model parameters, parameters of similar topologies are combined, i.e. γ(s) and m(s) corresponding to different s may be the same.
Each prediction model
Figure BDA0002828628800000166
Referred to as a submodel, or a weak classifier Es, which represents a mahalanobis kernel regression function constructed using historical data for a single topology class.
The above calculation formulas are all explanations of deriving probability meanings from the concept of relevance, and in fact, they can also be explained in a way of "similarity", fig. 2 is a flow architecture diagram in the process of training the prediction model provided by the present invention, as shown in fig. 2, the prediction model can be regarded as S weak learners
Figure BDA0002828628800000171
(ii) a composed strong learner, S ═ 1,2, …, S; the prediction model selects a weak learner in the prediction
Figure BDA0002828628800000172
The magnitude of "tendency" of (A) is PsO of (x, t)sProduct of (a), OsRefers to the data to be predicted and omegasSum of similarity (inner product) of inner data.
In FIG. 2, although the data to be predicted and the topology T are shownr uniIs very similar, however, due to its similarity to ΩrThe overall similarity of the intra-data is too small (because there is too little data in the topology), so the model does not tend to select the weak learner Er, while the data to be predicted is similar to the topology Ts uniIs not as similar as Tr uniHowever, because of the large amount of data in the topology category and the accuracy of the constructed model, the model tends to select the weak learner ES。
On the basis of the above-described embodiment, in this method,
the method comprises the following steps of inputting the power flow parameters and the topological structure parameters into a prediction model, outputting a predicted value of the ultimate removal time of the power grid to be tested, and carrying the existing training reference set by the prediction model after training is completed, wherein the method specifically comprises the following steps:
determining the input power flow parameter and the input topological structure parameter as x respectivelyuAnd tq；
Determining tqT of u type in S topology structure category after combinationu，1≤u≤S；
Calculating and outputting a predicted value of the limit cutting time of the power grid to be tested according to the following formula
Figure BDA0002828628800000176
Figure BDA0002828628800000177
Figure BDA0002828628800000178
Figure BDA0002828628800000179
Figure BDA00028286288000001710
Wherein x isuVector of dimension D, tuIs a vector of dimension W. t is tuIs a variable of 0 to 1, W<D，Ts uniAnd Tr uniRepresents the s-th and r-th topological structure parameter vectors gamma in the combined topological structure types0 done、M0 done、γ(s)doneAnd M(s)doneRespectively waiting for training corresponding to the prediction modelAdjusting parameter gamma0、M0Final adjustment parameters after training of γ(s) and M(s) is completed,
Figure BDA0002828628800000183
for the reference power flow parameter matrix X in the existing training reference setTRAINThe (c) th row of (a),
Figure BDA0002828628800000184
vector of dimension D, XTRAINA matrix, X, formed by all the reference tide parameters in the existing training reference setTRAINSize NTRAIN×D，NTRAINFor the number of the reference tidal current parameters in the existing training reference set,
Figure BDA0002828628800000185
for the existing training reference set
Figure BDA0002828628800000186
Corresponding reference limit excision time, ΩsIs XTRAINSet of rows of all original topologies in the s-th category, T, of the merged topology categorys uniAnd Tr uniRepresenting the parameter vectors of the s-th topological structure and the r-th topological structure in the combined topological structure types, wherein the dimensions are W and gamma0 doneAnd gamma(s)doneAre all scalar quantities, M0 doneThe size of (a) is W × W, M(s)doneThe size of (D) is D × D, S is the total number of the combined topology types, and r is 1,2, …, S.
Specifically, the above calculation formula shows a calculation flow in the model when the prediction model is used after training is completed. After the training of the prediction model is finished, the parameter gamma is to be adjusted0、γ(s)、M0And M(s) gamma which is obtained by finishing adjustment and completing optimization and obtaining a fixed value0 done、γ(s)done、M0 doneAnd M(s)done. When the prediction model is used, the prediction model still carries the existing training reference set used in model training, and the calculation is carried out according to the existing training reference setThe calculation flow given by the formula calculates the predicted CCT value. It should be noted that, as in the training process, the topology structure parameters of each sample need to be classified first, the combined topology structure type to which the sample belongs is found, then the predicted value is calculated, and the parameter to be adjusted is adjusted based on the predicted value calculation error. Because the classification in the training process reduces the types of the topological structure parameters and the types of the input low topological structure parameters in the use process, the calculation amount of model training and use is greatly reduced, and the calculation burden is reduced.
Here, a rule for combining various topology parameters to obtain a combined topology type will be described. Because the number of various buses and various nodes in the power grid structure is very large, the situations that the respective switches or the inputs of the various buses and the various nodes are disconnected are very many, and the types of the topological structure parameters are doubled every time one bus or node is added, so that the topological structure parameters of the same type need to be classified into one class for processing, and the classification method can be a classification method based on a machine learning clustering algorithm, a classification method based on a calculated correlation coefficient between any two topological structure parameter matrixes, or a classification method based on devices with the same property in the topological structure and a connection mode between the devices, and the like, and is not limited specifically herein.
On the basis of the above-described embodiment, in this method,
the loss function in the prediction model training process is formed based on the mean square error between the output limit excision time prediction value and the corresponding limit excision time label.
Specifically, the forward calculation of the prediction network in the prediction model training process is different from the calculation mode of the network output in the common model training process, and the calculation mode needs to depend on the carried existing training reference set, but the feedback parameter adjusting mechanism of the network is the same as that of the common model training, and the most common mean square error is selected to calculate the loss function to obtain the loss functionThe numerical value of the number is used for adjusting the parameter to be adjusted, and the adjusting method may be a commonly used back propagation gradient descent method, or an optimized parameter adjustment based on a particle swarm optimization, and the like, and is not particularly limited herein. Gamma constructed as a result of the invention0、γ(s)、M0And M(s) can be solved, preferably by a stochastic gradient descent algorithm, and a particular solver can be tensorflow.
On the basis of the above-described embodiment, in this method,
elements in the power flow parameters comprise a real part and an imaginary part of any bus voltage in the power grid to be tested, power injected from a delta end on any alternating current line mu from delta to rho, power injected from a delta end on any transformer Q from delta to rho, charging power from a delta end on any alternating current line mu from delta to rho, active power of a generator connected with any node, reactive power of a generator connected with any node, active power of a load connected with any node and reactive power of a load connected with any node;
the elements in the topological structure parameters comprise the switching-in and switching-off condition of any bus in the power grid to be tested, the switching-off condition of any alternating current line mu from delta to rho, the switching-off condition of any transformer Q from delta to rho, the switching-on and switching-off condition of a generator connected with any node and the connection and disconnection condition of any node.
Specifically, the determination of the elements and values in the power flow parameters and the topology parameters is further explained here: table 1 is a continuous variable description table, the power flow parameter is a D-dimensional vector, the meaning of each element of the vector is shown in table 1, the dimension D indicates that the power flow parameter contains D variables in table 1 in total, and the specific composition of the variables is determined by the topological structure of the power grid.
TABLE 1 continuous variable description
Figure BDA0002828628800000201
Table 2 is a discrete variable description table, the topology parameter is a topology vector describing the power grid topology, and its element is a discrete variable, also called a 0-1 variable (because its value is only 0 or 1), and its description is as follows, the topology parameter dimension is W (W < D), it indicates that the topology parameter includes W total variables in table 2, and the specific composition of the variables is determined by the initial topology structure of the power grid.
TABLE 2 description of discrete variables
Figure BDA0002828628800000211
The following describes the power system dynamic analysis apparatus based on ensemble learning and power grid topology change according to the present invention, and the power system dynamic analysis apparatus based on ensemble learning and power grid topology change described below and the first power system dynamic analysis method based on ensemble learning and power grid topology change described above may be referred to correspondingly.
Fig. 3 is a schematic structural diagram of an apparatus for dynamically analyzing a power system based on integrated learning and grid topology change according to the present invention, as shown in fig. 3, the apparatus includes a collecting unit 310 and a calculating unit 320, wherein,
the acquisition unit 310 is configured to acquire current power flow parameters and topology parameters of a power grid to be detected;
the calculating unit 320 is configured to input the power flow parameters and the topological structure parameters into a prediction model, and output a predicted value of the limit removal time of the power grid to be tested;
the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed.
The dynamic analysis device of the power system based on the integrated learning and the power grid topological change provided by the invention collects the current power flow parameters and topological structure parameters of the power grid to be tested; inputting the power flow parameters and the topological structure parameters into a prediction model, and outputting a limit cutting time prediction value of the power grid to be tested; the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed. Because the prediction model is obtained by training based on the sample power flow parameters, the sample topological structure parameters and the corresponding limit excision time labels on the basis of the existing training reference set, the accuracy of the trained prediction model can be ensured by ensuring that the sample and the existing training reference set have larger data volume, and the problem of too large calculated amount caused by excessive parameters of a neural network needing to be adjusted due to the fact that a large number of sample topological structure parameters need to be considered in the model training process is solved by the integrated learning mechanism, so that the calculated amount of dynamic safety analysis of the power system is reduced. Therefore, the device provided by the invention can reduce the calculation burden of dynamic safety analysis of the power system and improve the accuracy of the safety analysis.
On the basis of the above-described embodiment, in the apparatus,
and in the prediction model training process, establishing a relation between an input sample power flow parameter and a topological structure parameter, an output limit excision time prediction value and the existing training reference set based on a Gaussian kernel function constructed by the Mammai distance.
On the basis of the above-described embodiment, in the apparatus,
the method for constructing the relation between the input sample power flow parameter and the topological structure parameter, the output limit excision time predicted value and the existing training reference set based on the Gaussian kernel function constructed based on the Mammai distance specifically comprises the following steps:
for any combined topological structure type, constructing a relation between a sub-limit excision time predicted value, an input sample power flow parameter, a reference power flow parameter corresponding to any combined topological structure type in the existing training reference set and reference limit excision time under the topological structure type;
constructing a weighting coefficient of the predicted value of any sub-limit cutting time, the combined topological structure type, the input sample power flow parameter, the input sample topological structure and the relation between the reference power flow parameter and the reference limit cutting time in the existing training reference set on the basis of a Gaussian kernel function constructed by the Markov distance;
and determining an output limit excision time predicted value based on all the sub-limit excision time predicted values and the weighting coefficients of all the sub-limit excision time predicted values.
On the basis of the above-described embodiment, in the apparatus,
for any combined topological structure type, constructing a relation between a sub-limit excision time predicted value, an input sample power flow parameter, a reference power flow parameter corresponding to any combined topological structure type in the existing training reference set and a reference limit excision time, and specifically comprising:
determining the predicted value of the sub-limit excision time under any topological structure type s through the following formula
Figure BDA0002828628800000231
Figure BDA0002828628800000232
Figure BDA0002828628800000233
Wherein x is the input sample power flow parameter, x is the vector of D dimension,
Figure BDA0002828628800000234
for the reference power flow parameter matrix X in the existing training reference setTRAINThe (c) th row of (a),
Figure BDA0002828628800000235
vector of dimension D, XTRAINA matrix, X, formed by all the reference tide parameters in the existing training reference setTRAINSize NTRAIN×D，NTRAINFor the number of the reference tidal current parameters in the existing training reference set,
Figure BDA0002828628800000236
for the existing training reference set
Figure BDA0002828628800000237
Corresponding reference limit excision time, ΩsIs XTRAINIn the merged topological structure type, the number of rows of all original topological structures in the S-th topological structure type is collected, γ (S) is a parameter of a scalar to be adjusted during the training of the prediction model, m (S) is a parameter of a matrix to be adjusted during the training of the prediction model, the size of m (S) is DxD, S is the total number of the merged topological structure types, and r is 1,2, … and S;
correspondingly, the constructing of the weighting coefficient of the predicted value of any sub-limit resection time, the combined topological structure type, the input sample power flow parameter, the input sample topological structure, and the relationship between the reference power flow parameter in the existing training reference set and the reference limit resection time by the gaussian kernel function constructed based on the marching distance specifically includes:
determining the predicted value of any one sub-limit excision time by the following formula
Figure BDA0002828628800000241
Is weighted by a weight coefficient Ps(x,t)：
Figure BDA0002828628800000242
Figure BDA0002828628800000243
Figure BDA0002828628800000244
Wherein t is an input sample topological structure parameter corresponding to the input sample power flow parameter x, the elements of the input sample topological structure parameter are variables of 0 to 1, t is a vector of W dimension, and W is the vector of W dimension<D，Ts uniAnd Tr uniRepresenting the parameter vectors of the s-th topological structure and the r-th topological structure in the combined topological structure types, wherein the dimensions are W and gamma0And gamma(s) are parameters of the scalar to be adjusted during the training of the prediction model, M0And M(s) are all the parameters of the matrix to be adjusted during the training of the prediction model, M0Is W × W, m (S) is D × D, S is the total number of the merged topology classes, r is 1,2, …, S;
the determining the output limit resection time predicted value based on the all sub-limit resection time predicted values and the weighting coefficients of all sub-limit resection time predicted values specifically comprises:
determining an output limit cut time prediction value by the following formula
Figure BDA0002828628800000247
Figure BDA0002828628800000248
Wherein S is the total number of the combined topological structure types,
Figure BDA0002828628800000249
for sub-limit excision time prediction, P, under any topological structure type ss(x, t) is the predicted value of any one of the sub-limit excision time
Figure BDA00028286288000002410
S is 1,2, …, S.
On the basis of the above-described embodiment, in the apparatus,
the method comprises the following steps of inputting the power flow parameters and the topological structure parameters into a prediction model, outputting a predicted value of the ultimate removal time of the power grid to be tested, and carrying the existing training reference set by the prediction model after training is completed, wherein the method specifically comprises the following steps:
determining the input power flow parameter and the input topological structure parameter as x respectivelyuAnd tq；
Determining tqT of u type in S topology structure category after combinationu，1≤u≤S；
Calculating and outputting a predicted value of the limit cutting time of the power grid to be tested according to the following formula
Figure BDA0002828628800000251
Figure BDA0002828628800000252
Figure BDA0002828628800000253
Figure BDA0002828628800000254
Figure BDA0002828628800000255
Wherein x isuVector of dimension D, tuIs a vector of dimension W. t is tuIs a variable of 0 to 1, W<D，Ts uniAnd Tr uniRepresents the s-th and r-th topological structure parameter vectors gamma in the combined topological structure types0 done、M0 done、γ(s)doneAnd M(s)doneRespectively corresponding to the parameters gamma to be adjusted during the training of the prediction model0、M0Final adjustment parameters after training of γ(s) and M(s) is completed,
Figure BDA0002828628800000256
for the reference power flow parameter matrix X in the existing training reference setTRAINThe (c) th row of (a),
Figure BDA0002828628800000257
vector of dimension D, XTRAINA matrix, X, formed by all the reference tide parameters in the existing training reference setTRAINSize NTRAIN×D，NTRAINFor the number of the reference tidal current parameters in the existing training reference set,
Figure BDA0002828628800000258
for the existing training reference set
Figure BDA0002828628800000259
Corresponding reference limit excision time, ΩsIs XTRAINSet of rows of all original topologies in the s-th category, T, of the merged topology categorys uniAnd Tr uniRepresenting the parameter vectors of the s-th topological structure and the r-th topological structure in the combined topological structure types, wherein the dimensions are W and gamma0 doneAnd gamma(s)doneAre all scalar quantities, M0 doneThe size of (a) is W × W, M(s)doneThe size of (D) is D × D, S is the total number of the combined topology types, and r is 1,2, …, S.
On the basis of the above-described embodiment, in the apparatus,
the loss function in the prediction model training process is formed based on the mean square error between the output limit excision time prediction value and the corresponding limit excision time label.
On the basis of the above-described embodiment, in the apparatus,
elements in the power flow parameters comprise a real part and an imaginary part of any bus voltage in the power grid to be tested, power injected from a delta end on any alternating current line mu from delta to rho, power injected from a delta end on any transformer Q from delta to rho, charging power from a delta end on any alternating current line mu from delta to rho, active power of a generator connected with any node, reactive power of a generator connected with any node, active power of a load connected with any node and reactive power of a load connected with any node;
the elements in the topological structure parameters comprise the switching-in and switching-off condition of any bus in the power grid to be tested, the switching-off condition of any alternating current line mu from delta to rho, the switching-off condition of any transformer Q from delta to rho, the switching-on and switching-off condition of a generator connected with any node and the connection and disconnection condition of any node.
Fig. 4 illustrates a physical structure diagram of an electronic device, which may include, as shown in fig. 4: a processor (processor)410, a communication Interface 420, a memory (memory)430 and a communication bus 440, wherein the processor 410, the communication Interface 420 and the memory 430 are communicated with each other via the communication bus 440. The processor 410 may invoke logic instructions in the memory 430 to perform a method of power system dynamic analysis based on integrated learning and grid topology changes, the method comprising: collecting current tide parameters and topological structure parameters of a power grid to be tested; inputting the power flow parameters and the topological structure parameters into a prediction model, and outputting a limit cutting time prediction value of the power grid to be tested; the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed.
In addition, the logic instructions in the memory 430 may be implemented in the form of software functional units and stored in a computer readable storage medium when the software functional units are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions, which when executed by a computer, enable the computer to perform a method for integrated learning and grid topology change based power system dynamics analysis provided by the above methods, the method comprising: collecting current tide parameters and topological structure parameters of a power grid to be tested; inputting the power flow parameters and the topological structure parameters into a prediction model, and outputting a limit cutting time prediction value of the power grid to be tested; the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed.
In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, the computer program being implemented by a processor to perform the method for integrated learning and grid topology change based power system dynamic analysis provided by the above methods, the method comprising: collecting current tide parameters and topological structure parameters of a power grid to be tested; inputting the power flow parameters and the topological structure parameters into a prediction model, and outputting a limit cutting time prediction value of the power grid to be tested; the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present 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 solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (10)

1. A dynamic analysis method of an electric power system based on integrated learning and power grid topological change is characterized by comprising the following steps:
collecting current tide parameters and topological structure parameters of a power grid to be tested;
inputting the power flow parameters and the topological structure parameters into a prediction model, and outputting a limit cutting time prediction value of the power grid to be tested;
the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed.
2. The integrated learning and power grid topology change-based power system dynamic analysis method according to claim 1, wherein a Gaussian kernel function constructed based on a Mammy distance in the prediction model training process establishes a relationship between input sample power flow parameters and topology structure parameters, output limit excision time prediction values and the existing training reference set.
3. The dynamic power system analysis method based on ensemble learning and power grid topology change according to claim 2, wherein the gaussian kernel function constructed based on the euclidean distance establishes a relationship between the input sample power flow parameter and the topology structure parameter, the output limit excision time prediction value and the existing training reference set, and specifically comprises:
for any combined topological structure type, constructing a relation between a sub-limit excision time predicted value, an input sample power flow parameter, a reference power flow parameter corresponding to any combined topological structure type in the existing training reference set and reference limit excision time under the topological structure type;
constructing a weighting coefficient of the predicted value of any sub-limit cutting time, the combined topological structure type, the input sample power flow parameter, the input sample topological structure and the relation between the reference power flow parameter and the reference limit cutting time in the existing training reference set on the basis of a Gaussian kernel function constructed by the Markov distance;
and determining an output limit excision time predicted value based on all the sub-limit excision time predicted values and the weighting coefficients of all the sub-limit excision time predicted values.
4. The dynamic power system analysis method based on ensemble learning and grid topology change according to claim 3, wherein for any merged topology structure type, a relationship between a predicted value of sub-limit cut-off time, an input sample power flow parameter, a reference power flow parameter corresponding to any merged topology structure type in the existing training reference set, and reference limit cut-off time is constructed, and specifically comprises:
determining the predicted value of the sub-limit excision time under any topological structure type s through the following formula
Figure FDA0002828628790000021
Figure FDA0002828628790000022
Figure FDA0002828628790000023
Wherein x is the input sample power flow parameter, x is the vector of D dimension,
Figure FDA0002828628790000024
for the reference power flow parameter matrix X in the existing training reference setTRAINThe (c) th row of (a),
Figure FDA0002828628790000025
vector of dimension D, XTRAINA matrix, X, formed by all the reference tide parameters in the existing training reference setTRAINSize NTRAIN×D，NTRAINFor the number of the reference tidal current parameters in the existing training reference set,
Figure FDA0002828628790000026
for the existing training reference set
Figure FDA0002828628790000027
Corresponding reference limit excision time, ΩsIs XTRAINIn the merged topological structure type, the number of rows of all original topological structures in the S-th topological structure type is collected, γ (S) is a parameter of a scalar to be adjusted during the training of the prediction model, m (S) is a parameter of a matrix to be adjusted during the training of the prediction model, the size of m (S) is DxD, S is the total number of the merged topological structure types, and r is 1,2, … and S;
correspondingly, the constructing of the weighting coefficient of the predicted value of any sub-limit resection time, the combined topological structure type, the input sample power flow parameter, the input sample topological structure, and the relationship between the reference power flow parameter in the existing training reference set and the reference limit resection time by the gaussian kernel function constructed based on the marching distance specifically includes:
determining the predicted value of any one sub-limit excision time by the following formula
Figure FDA0002828628790000028
Is weighted by a weight coefficient Ps(x,t)：
Figure FDA0002828628790000031
Figure FDA0002828628790000032
Figure FDA0002828628790000033
Wherein t is an input sample topological structure parameter corresponding to the input sample power flow parameter x, the elements of the input sample topological structure parameter are variables of 0 to 1, t is a vector of W dimension, and W is the vector of W dimension<D，Ts uniAnd Tr uniRepresenting the parameter vectors of the s-th topological structure and the r-th topological structure in the combined topological structure types, wherein the dimensions are W and gamma0And gamma(s) are parameters of the scalar to be adjusted during the training of the prediction model, M0And M(s) are all the parameters of the matrix to be adjusted during the training of the prediction model, M0Is W × W, m (S) is D × D, S is the total number of the merged topology classes, r is 1,2, …, S;
the determining the output limit resection time predicted value based on the all sub-limit resection time predicted values and the weighting coefficients of all sub-limit resection time predicted values specifically comprises:
determining an output limit cut time prediction value by the following formula
Figure FDA0002828628790000034
Figure FDA0002828628790000035
Wherein S is the total number of the combined topological structure types,
Figure FDA0002828628790000036
for sub-limit excision time prediction, P, under any topological structure type ss(x, t) is the predicted value of any one of the sub-limit excision time
Figure FDA0002828628790000037
S is 1,2, …, S.
5. The dynamic power system analysis method based on ensemble learning and power grid topology change according to claim 4, wherein the power flow parameters and the topology structure parameters are input into a prediction model, a predicted value of the limit removal time of the power grid to be tested is output, and the prediction model carries the existing training reference set after training is completed, specifically comprising:
determining the input power flow parameter and the input topological structure parameter as x respectivelyuAnd tq；
Determining tqT of u type in S topology structure category after combinationu，1≤u≤S；
Calculating and outputting a predicted value of the limit cutting time of the power grid to be tested according to the following formula
Figure FDA0002828628790000041
Figure FDA0002828628790000042
Figure FDA0002828628790000043
Figure FDA0002828628790000044
Figure FDA0002828628790000045
Wherein x isuVector of dimension D, tuIs a vector of dimension W. t is tuIs a variable of 0 to 1, W<D，Ts uniAnd Tr uniRepresents the s-th and r-th topological structure parameter vectors gamma in the combined topological structure types0 done、M0 done、γ(s)doneAnd M(s)doneRespectively corresponding to the parameters gamma to be adjusted during the training of the prediction model0、M0Final adjustment parameters after training of γ(s) and M(s) is completed,
Figure FDA0002828628790000046
for the reference power flow parameter matrix X in the existing training reference setTRAINThe (c) th row of (a),
Figure FDA0002828628790000047
vector of dimension D, XTRAINA matrix, X, formed by all the reference tide parameters in the existing training reference setTRAINSize NTRAIN×D，NTRAINFor the number of the reference tidal current parameters in the existing training reference set,
Figure FDA0002828628790000048
for the existing training reference set
Figure FDA0002828628790000049
Corresponding reference limit excision time, ΩsIs XTRAINSet of rows of all original topologies in the s-th category, T, of the merged topology categorys uniAnd Tr uniRepresenting the parameter vectors of the s-th topological structure and the r-th topological structure in the combined topological structure types, wherein the dimensions are W and gamma0 doneAnd gamma(s)doneAre all scalar quantities, M0 doneThe size of (a) is W × W, M(s)doneThe size of (D) is D × D, S is the total number of the combined topology types, and r is 1,2, …, S.
6. The integrated learning and power grid topology change based power system dynamic analysis method according to claims 1-5, characterized in that the loss function in the prediction model training process is constructed based on the mean square error between the output limit cut-off time prediction value and the corresponding limit cut-off time label.
7. The integrated learning and grid topology change based power system dynamic analysis method according to claims 1-5,
elements in the power flow parameters comprise a real part and an imaginary part of any bus voltage in the power grid to be tested, power injected from a delta end on any alternating current line mu from delta to rho, power injected from a delta end on any transformer Q from delta to rho, charging power from a delta end on any alternating current line mu from delta to rho, active power of a generator connected with any node, reactive power of a generator connected with any node, active power of a load connected with any node and reactive power of a load connected with any node;
the elements in the topological structure parameters comprise the switching-in and switching-off condition of any bus in the power grid to be tested, the switching-off condition of any alternating current line mu from delta to rho, the switching-off condition of any transformer Q from delta to rho, the switching-on and switching-off condition of a generator connected with any node and the connection and disconnection condition of any node.
8. An electric power system dynamic analysis device based on integrated learning and power grid topology change is characterized by comprising:
the acquisition unit is used for acquiring the current tidal current parameter and the topological structure parameter of the power grid to be detected;
the calculation unit is used for inputting the power flow parameters and the topological structure parameters into a prediction model and outputting a predicted value of the limit removal time of the power grid to be tested;
the prediction model is obtained by training on the basis of a sample power flow parameter, a sample topological structure parameter and a corresponding limit removal time label on the basis of an existing training reference set, the types of topological structures are combined in the training process, and the prediction model carries the existing training reference set after training is completed.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor, when executing the program, implements the steps of the integrated learning and grid topology change based power system dynamic analysis method according to any of claims 1 to 7.
10. A non-transitory computer readable storage medium having stored thereon a computer program, which when executed by a processor, performs the steps of the method of any of claims 1 to 7 for integrated learning and grid topology change based power system dynamics analysis.
CN202011455569.9A 2020-12-10 2020-12-10 Power system dynamic analysis method based on integrated learning and power grid topological change Pending CN112561303A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011455569.9A CN112561303A (en) 2020-12-10 2020-12-10 Power system dynamic analysis method based on integrated learning and power grid topological change

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011455569.9A CN112561303A (en) 2020-12-10 2020-12-10 Power system dynamic analysis method based on integrated learning and power grid topological change

Publications (1)

Publication Number Publication Date
CN112561303A true CN112561303A (en) 2021-03-26

Family

ID=75061880

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011455569.9A Pending CN112561303A (en) 2020-12-10 2020-12-10 Power system dynamic analysis method based on integrated learning and power grid topological change

Country Status (1)

Country Link
CN (1) CN112561303A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113239651A (en) * 2021-07-12 2021-08-10 苏州贝克微电子有限公司 Artificial intelligence implementation method and system for circuit design
CN113783186A (en) * 2021-08-23 2021-12-10 电子科技大学 Voltage prediction method considering topological structure change of power distribution network

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109066687A (en) * 2018-08-22 2018-12-21 深圳先进技术研究院 A kind of electric power system tide calculation method, system and electronic equipment
CN109190672A (en) * 2018-08-02 2019-01-11 清华大学 Operation of Electric Systems operating condition unsupervised clustering and device
CN109412161A (en) * 2018-12-18 2019-03-01 国网重庆市电力公司电力科学研究院 A kind of Probabilistic Load calculation method and system
CN109598386A (en) * 2018-12-12 2019-04-09 国网山东省电力公司临沂供电公司 A kind of accurate analysis method of distribution optimization and system based on deep learning
CN109638826A (en) * 2018-12-24 2019-04-16 国网江苏省电力有限公司苏州供电分公司 A kind of topology identification method and system based on CNN
WO2019218263A1 (en) * 2018-05-16 2019-11-21 深圳大学 Extreme learning machine-based extreme ts fuzzy inference method and system
CN110879917A (en) * 2019-11-08 2020-03-13 北京交通大学 Electric power system transient stability self-adaptive evaluation method based on transfer learning
CN111756034A (en) * 2020-06-12 2020-10-09 清华大学 Transient voltage stability evaluation method for power system based on graph space-time network
CN111914486A (en) * 2020-08-07 2020-11-10 中国南方电网有限责任公司 Power system transient stability evaluation method based on graph attention network
CN111932396A (en) * 2020-06-05 2020-11-13 国网江苏省电力有限公司 Automatic identification method for power distribution network topology network

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019218263A1 (en) * 2018-05-16 2019-11-21 深圳大学 Extreme learning machine-based extreme ts fuzzy inference method and system
CN109190672A (en) * 2018-08-02 2019-01-11 清华大学 Operation of Electric Systems operating condition unsupervised clustering and device
CN109066687A (en) * 2018-08-22 2018-12-21 深圳先进技术研究院 A kind of electric power system tide calculation method, system and electronic equipment
CN109598386A (en) * 2018-12-12 2019-04-09 国网山东省电力公司临沂供电公司 A kind of accurate analysis method of distribution optimization and system based on deep learning
CN109412161A (en) * 2018-12-18 2019-03-01 国网重庆市电力公司电力科学研究院 A kind of Probabilistic Load calculation method and system
CN109638826A (en) * 2018-12-24 2019-04-16 国网江苏省电力有限公司苏州供电分公司 A kind of topology identification method and system based on CNN
CN110879917A (en) * 2019-11-08 2020-03-13 北京交通大学 Electric power system transient stability self-adaptive evaluation method based on transfer learning
CN111932396A (en) * 2020-06-05 2020-11-13 国网江苏省电力有限公司 Automatic identification method for power distribution network topology network
CN111756034A (en) * 2020-06-12 2020-10-09 清华大学 Transient voltage stability evaluation method for power system based on graph space-time network
CN111914486A (en) * 2020-08-07 2020-11-10 中国南方电网有限责任公司 Power system transient stability evaluation method based on graph attention network

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113239651A (en) * 2021-07-12 2021-08-10 苏州贝克微电子有限公司 Artificial intelligence implementation method and system for circuit design
CN113783186A (en) * 2021-08-23 2021-12-10 电子科技大学 Voltage prediction method considering topological structure change of power distribution network
CN113783186B (en) * 2021-08-23 2023-04-18 电子科技大学 Voltage prediction method considering topological structure change of power distribution network

Similar Documents

Publication Publication Date Title
CN110417011B (en) Online dynamic security assessment method based on mutual information and iterative random forest
CN104778173A (en) Determination method, device and equipment of objective user
CN112508442B (en) Transient stability assessment method and system based on automatic and interpretable machine learning
CN111523778A (en) Power grid operation safety assessment method based on particle swarm algorithm and gradient lifting tree
CN108898476A (en) A kind of loan customer credit-graded approach and device
CN108879732B (en) Transient stability evaluation method and device for power system
CN112561303A (en) Power system dynamic analysis method based on integrated learning and power grid topological change
CN108090615B (en) Minimum frequency prediction method after power system fault based on cross entropy integrated learning
CN112217207A (en) Transient voltage stability margin prediction method and device
CN108062566A (en) A kind of intelligent integrated flexible measurement method based on the potential feature extraction of multinuclear
CN115510042A (en) Power system load data filling method and device based on generation countermeasure network
CN113111924A (en) Electric power customer classification method and device
Du et al. Applying deep convolutional neural network for fast security assessment with N-1 contingency
CN113506186B (en) Method and device for screening disturbed track of power system, electronic equipment and storage medium
Li et al. Nonstationary fractionally integrated functional time series
CN110457469A (en) Information classification approach, device based on shot and long term memory network, computer equipment
CN114202174A (en) Electricity price risk grade early warning method and device and storage medium
CN114492742A (en) Neural network structure searching method, model issuing method, electronic device, and storage medium
CN112419098A (en) Power grid safety and stability simulation sample screening and expanding method based on safety information entropy
CN115409317A (en) Transformer area line loss detection method and device based on feature selection and machine learning
CN113627655A (en) Power distribution network pre-disaster fault scene simulation prediction method and device
CN113393023A (en) Mold quality evaluation method, apparatus, device and storage medium
CN113408564A (en) Graph processing method, network training method, device, equipment and storage medium
Sobhanifard et al. An Integrated Neural Networks and MCMC Model to Predicting Bank’s Efficiency
CN111310971B (en) Prospect analysis method, device and equipment for O2O business model

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