CN114492150A - Power distribution network typical service scene early warning method based on digital twin - Google Patents

Power distribution network typical service scene early warning method based on digital twin Download PDF

Info

Publication number
CN114492150A
CN114492150A CN202011148753.9A CN202011148753A CN114492150A CN 114492150 A CN114492150 A CN 114492150A CN 202011148753 A CN202011148753 A CN 202011148753A CN 114492150 A CN114492150 A CN 114492150A
Authority
CN
China
Prior art keywords
data
power distribution
distribution network
early warning
digital twin
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202011148753.9A
Other languages
Chinese (zh)
Other versions
CN114492150B (en
Inventor
周亮
严川
李炜
盛庆博
王晓东
郑炜博
石小满
董伟佳
刘杰
孙东
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.)
China Petroleum and Chemical Corp
Technology Inspection Center of Sinopec Shengli Oilfield Co
Shengli Oilfield Testing and Evaluation Research Co Ltd
Original Assignee
China Petroleum and Chemical Corp
Technology Inspection Center of Sinopec Shengli Oilfield Co
Shengli Oilfield Testing and Evaluation Research Co Ltd
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 China Petroleum and Chemical Corp, Technology Inspection Center of Sinopec Shengli Oilfield Co, Shengli Oilfield Testing and Evaluation Research Co Ltd filed Critical China Petroleum and Chemical Corp
Priority to CN202011148753.9A priority Critical patent/CN114492150B/en
Publication of CN114492150A publication Critical patent/CN114492150A/en
Application granted granted Critical
Publication of CN114492150B publication Critical patent/CN114492150B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/04—Power grid distribution networks

Landscapes

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

Abstract

The invention discloses a power distribution network typical service scene early warning method based on a digital twin body, which relates to the field of electrical engineering, and adopts the technical scheme that multi-source data is generated and fused based on a countermeasure thought; building a digital twin system framework of the power distribution network; building a power failure early warning prediction model; constructing a line loss calculation and early warning model; and inputting the collected power grid data, and carrying out early warning through a digital twin system. The invention has the beneficial effects that: the method not only can repair missing data and fuse multi-source data by adopting an anti-game idea from a data source, retains the measured time sequence characteristics and correlation in the power system, but also can achieve the purpose of analyzing and solving the mechanism of the physical system and finely depicting the information-physical fusion characteristics in modeling and simulation of an actual scene with high integration level, strong mixed type and rich information links.

Description

Power distribution network typical service scene early warning method based on digital twin
Technical Field
The invention relates to the field of electrical engineering, in particular to a power distribution network typical service scene early warning method based on a digital twin body.
Background
The power distribution network is complex in composition structure, different in model form and complex in coupling relation, and forms diversified model components. The traditional power distribution network analysis mode based on physical mechanism modeling is limited by modeling precision and calculation speed, and is very popular in the presence of huge and complex networks. In addition, in the actual operation of the measurement system, all links of data acquisition, measurement, transmission and conversion are likely to break down or be interfered, so that the data is lost and abnormal, and the accuracy and the efficiency of data feature extraction and data mining are seriously influenced.
The correct cognition of the power distribution network is a prerequisite condition for various decisions of power grid operation, management and scheduling, and the complexity of cognition on the power distribution network is aggravated by the characteristics of nonlinearity, high dimensionality, distribution and the like of the conventional power distribution network. The cognitive difficulty of modern power distribution networks lies in: 1) the system is diversified, has high asymmetry degree and strong nonlinearity and uncertainty; 2) the system is complex and in a state of evolution in time.
The typical application scene of the existing power distribution network mostly adopts a physical modeling method, and the mathematical relationship among all variables is constructed by analyzing and explaining the physical and chemical mechanisms of the process. Although the method has the advantages of strict logic, clear process, easy explanation and the like, the method is limited in application in a large and complex system due to the fact that the calculation speed is low, the parameter estimation is complicated, the prediction accuracy is not high enough and the like.
Disclosure of Invention
Aiming at the technical problems, the invention provides a power distribution network typical service scene early warning method based on a digital twin body.
The technical proposal is that the method comprises the following steps,
s1, multi-source data generation and fusion are carried out based on the countermeasure generating idea, the countermeasure generating idea is applied, historical fault data with insufficient sample quantity are simulated and supplemented, real fault data are sampled and trained, a generated sample approaching historical data is obtained, and a specific operation scene is simulated;
s2, building a power distribution network digital twin system framework, and building three-dimensional mapping of a physical entity in a virtual digital world according to the multi-source data supplemented and fused in the S1;
s3, constructing a power failure early warning prediction model; the power distribution network digital twin system built based on the S2 selects the data processed through the countermeasure thought as input, the key characteristic quantity which influences the power failure of the power distribution network is screened according to the sorting result of the distance correlation coefficient and is used as the input quantity of the power failure risk early warning based on the convolutional neural network, the output is the power failure risk level of the specific line of the power distribution network, and when the risk level exceeds a set threshold value, the power failure early warning is triggered;
s4, constructing a line loss calculation and early warning model; the power distribution network digital twin system constructed based on the S2 takes data processed by the countermeasure idea as input, fits a power flow equation and state estimation through a neural network, calculates the current value of a line, calculates the line loss as final output, and triggers line loss abnormity early warning when the line loss value exceeds a set threshold value.
Preferably, in S1, for data loss caused by a fault of the measurement and acquisition device, applying a generation countermeasure concept, filling a missing sample, converting an unbalanced data set into a balanced data set, and performing multi-source data fusion.
Preferably, the generating countermeasure thought comprises two important components, namely a generator and a discriminator;
the generator is responsible for learning the distribution of historical fault data and generating new data, real data and the generated data are simultaneously input into the discriminator, and whether the input data are real data or not is judged;
the two neural networks play against each other in a game mode, the generator tries to generate a generated sample which approximates to the distribution of real data so as to deceive the discriminator, and the discriminator identifies the real sample and the generated sample as far as possible;
after the model training is finished, the generator learns the distribution of the fault data of the power distribution network, and a large number of generated samples according with the data distribution rule can be generated.
The generator is trained to minimize a generator loss function, and the arbiter is trained to minimize an arbiter loss function;
through carrying out alternating iterative optimization training on the generator and the discriminator, the Nash equilibrium point is finally reached between the generator and the discriminator, namely, the generator can synthesize artificial samples which are difficult to discriminate by the discriminator.
Preferably, the loss functions of the generator and the arbiter are,
Figure RE-GDA0002788993990000021
Figure RE-GDA0002788993990000022
wherein, z to Pz(z) represents a noise vector z, x-P sampled from a normal distributionrRepresenting a true sample sampled from a true data distribution; e (-) represents the calculation expectation; g (z) represents the sample generated by the generator, and x or G (z) is input into the discriminator D to obtain an output result D (-).
Preferably, in S1, the simulation and supplementation of the historical fault data with insufficient sample number specifically includes inputting a random noise vector z into a trained generator to obtain generated data corresponding to a missing position in the real data, and the finally filled measurement data is composed of an actual measurement portion in the original sample and a data missing portion in the generated sample.
Preferably, the S2 builds a power distribution network digital twin system framework, and establishes three-dimensional mapping of physical entities in a virtual digital world according to supplemented and fused multi-source data, specifically, maps various index data such as transformers, lines and loads in the power distribution network into a virtual digital space by various digitalization means such as internet of things and virtual reality, so as to complete accurate depiction and simulation of complex physical entities, and divide application models according to typical service scenes.
Preferably, in the step S2, the VGGNet convolutional neural network is adopted to perform regression and classification analysis on the data, analyze and mine the internal rules of the data related to the power consumption system, and perform predictive analysis on the performance index that cannot be measured or directly calculated by the system;
the VGGNet convolutional neural network successfully constructs a convolutional neural network with 16-19 layers of depth through repeated stacking of 3 x 3 small convolutional kernels and 2 x 2 maximum pooling layers;
the VGGNet convolutional neural network divides the network into 5 sections, each section is formed by connecting a plurality of 3 x 3 convolutional networks in series, each section of convolution is followed by a maximum pooling layer, and finally, 3 full-connection layers and a softmax layer are arranged;
the characteristic data of a typical service scene is input into the VGGNet according to the time sequence, data of a future period of time after the time period to which the input data belongs is predicted, and the intermediate layer network can memorize the calculation state of the input data at a certain moment and transmit the calculation state to the calculation process of the data at the next moment so as to form the contact in the time relationship.
Preferably, the distance correlation coefficient in S3 may measure the degree of nonlinear correlation between two variables, and the value of the distance correlation coefficient is in [0,1], where when the coefficient is 0, it indicates that the two variables are independent of each other, and conversely, when the coefficient is 1, it indicates that the two variables are strongly correlated;
according to the definition of the distance correlation coefficient, sequentially calculating the correlation coefficient of the input data to each line during power failure, and sorting and screening out the key characteristic quantity which influences each line during power failure according to the correlation degree;
the distance correlation coefficient is calculated by the following formula:
Figure RE-GDA0002788993990000031
Figure RE-GDA0002788993990000032
Figure RE-GDA0002788993990000033
Figure RE-GDA0002788993990000041
where dCor (X, Y) denotes a distance correlation coefficient of two variables X, Y, dCov (X, Y) denotes a distance covariance, dvar (X) and dvar (Y) denote the distance variances of X and Y, respectively, and n denotes the lengths of the variables X and Y.
Preferably, the outage risk early warning model based on the convolutional neural network divides the power distribution network outage risk level into 4 levels, which are particularly important, larger and general. And classifying the historical data according to the 4 grades, wherein the class labels are the respective power failure risk grades.
Preferably, after the convolutional neural network is trained according to the multi-classification task, the power failure risk level of the specific line of the power distribution network can be obtained from key characteristic quantity data which affect the power failure of the power distribution network.
And when the power failure risk level exceeds a set threshold value, triggering power failure early warning.
Preferably, the neural network in S4 adopts a long and short memory neural network, inputs are line parameters and electrical measurements in the power distribution network, and outputs are current values of the power distribution network line.
Preferably, the line loss calculation formula in S4 is:
Figure RE-GDA0002788993990000042
wherein, Δ WLa,j，ΔWLb,i，ΔWLc,iThe loss of the A, B and C phases of the line i is respectively; ra,i，Rb,i，Rc,iA, B and C three-phase resistors of a circuit i are respectively arranged;
Figure RE-GDA0002788993990000043
the three-phase state currents of the line i in the period j are respectively A, B and C; n is the total time segment number; t is each time interval.
The technical scheme provided by the embodiment of the invention has the following beneficial effects: the modeling method for digital twin body adoption and data driving of the typical service scene of the power distribution network constructed based on the generation countermeasure idea can not only adopt the countermeasure game idea to repair missing data and fuse multi-source data from a data source and retain the measured time sequence characteristics and correlation in a power system, but also can achieve the purpose of combining the analysis and solution of the mechanism of a physical system and the fine description of the information-physical fusion characteristics in the modeling and simulation of the actual scene with high integration level, strong mixing type and rich information links, and can extract valuable information closely related to the core characteristics of the real system by effectively acquiring, exploring and analyzing the multi-source heterogeneous data in the whole life cycle of the system.
Drawings
FIG. 1 is a schematic flow chart of a method according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of the countermeasure idea of the embodiment of the invention.
Fig. 3 is a schematic diagram of a power outage early warning model according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of a line loss calculation and early warning model according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. Of course, the specific embodiments described herein are merely illustrative of the invention and are not intended to be limiting.
It should be noted that the embodiments and features of the embodiments of the present invention may be combined with each other without conflict.
Example 1
The invention provides a power distribution network typical service scene early warning method based on a digital twin body, which comprises the following steps,
s1, multi-source data generation and fusion are carried out based on the countermeasure generating idea, the countermeasure generating idea is applied, historical fault data with insufficient sample quantity are simulated and supplemented, real fault data are sampled and trained, a generated sample approaching historical data is obtained, and a specific operation scene is simulated; for data loss caused by measuring and collecting equipment faults, applying a generation countermeasure thought, filling a missing sample, converting an unbalanced data set into a balanced data set, and performing multi-source data fusion;
s2, building a power distribution network digital twin system framework, and building three-dimensional mapping of a physical entity in a virtual digital world according to the multi-source data supplemented and fused in the S1;
s3, constructing a power failure early warning prediction model; the power distribution network digital twin system built based on the S2 selects the data processed through the countermeasure thought as input, the key characteristic quantity which influences the power failure of the power distribution network is screened according to the sorting result of the distance correlation coefficient and is used as the input quantity of the power failure risk early warning based on the convolutional neural network, the output is the power failure risk level of the specific line of the power distribution network, and when the risk level exceeds a set threshold value, the power failure early warning is triggered;
s4, constructing a line loss calculation and early warning model; the power distribution network digital twin system constructed based on the S2 takes data processed by the countermeasure idea as input, fits a power flow equation and state estimation through a neural network, calculates the current value of a line, calculates the line loss as final output, and triggers line loss abnormity early warning when the line loss value exceeds a set threshold value.
Example 2
The invention provides a power distribution network typical service scene early warning method based on a digital twin body, which comprises the following steps,
s1, multi-source data generation and fusion are carried out based on the countermeasure generating idea, the countermeasure generating idea is applied, historical fault data with insufficient sample quantity are simulated and supplemented, real fault data are sampled and trained, a generated sample approaching historical data is obtained, and a specific operation scene is simulated; for data loss caused by measuring and collecting equipment faults, applying a generation countermeasure thought, filling a missing sample, converting an unbalanced data set into a balanced data set, and performing multi-source data fusion;
the generation countermeasure concept indicated in S1 contains two important components, the generator and the discriminator;
the generator is responsible for learning the distribution of historical fault data and generating new data, real data and the generated data are simultaneously input into the discriminator, and whether the input data are real data or not is judged;
the two neural networks play against each other in a game mode, the generator tries to generate a generated sample which approximates to the distribution of real data so as to deceive the discriminator, and the discriminator identifies the real sample and the generated sample as far as possible;
after the model training is finished, the generator learns the distribution of the fault data of the power distribution network, and a large number of generated samples according with the data distribution rule can be generated.
Training the generator according to the minimum loss function of the generator, and training the arbiter according to the minimum loss function of the arbiter;
through carrying out alternating iterative optimization training on the generator and the discriminator, the Nash equilibrium point is finally reached between the generator and the discriminator, namely, the generator can synthesize artificial samples which are difficult to discriminate by the discriminator.
The loss functions of the generator and the arbiter are,
Figure RE-GDA0002788993990000061
Figure RE-GDA0002788993990000062
wherein, z to Pz(z) represents a noise vector z, x-P sampled from a normal distributionrRepresenting a true sample sampled from a true data distribution; e (-) represents the calculation expectation; g (z) represents the sample generated by the generator, and x or G (z) is input into the discriminator D to obtain an output result D (-).
In S1, the simulation and supplementation of the historical fault data with insufficient sample number is specifically to input the random noise vector z into a trained generator to obtain generated data corresponding to a missing position in the real data, and the finally filled measurement data is composed of an actual measurement portion in the original sample and a data missing portion in the generated sample.
S2, building a power distribution network digital twin system framework, and building three-dimensional mapping of a physical entity in a virtual digital world according to the multi-source data supplemented and fused in the S1;
s2, a power distribution network digital twin system framework is built, three-dimensional mapping of physical entities in a virtual digital world is built according to supplemented and fused multi-source data, specifically, various index data such as transformers, lines and loads in the power distribution network are mapped into a virtual digital space through various digital means such as the Internet of things and virtual reality, accurate depiction and simulation of complex physical entities are completed, and application models are divided according to typical service scenes.
Step S2, a VGGNet convolutional neural network is adopted to conduct regression and classification analysis on data, the internal rules of relevant data of the power utilization system are analyzed and mined, and performance indexes which cannot be measured or directly calculated by the system are subjected to prediction analysis;
the VGGNet convolutional neural network successfully constructs a convolutional neural network with 16-19 layers of depth through repeated stacking of 3 x 3 small convolutional kernels and 2 x 2 maximum pooling layers;
the VGGNet convolutional neural network divides the network into 5 sections, each section connects a plurality of convolutional networks of 3 x 3 in series, each section of convolution is followed by a maximum pooling layer, and finally, 3 full connection layers and a softmax layer are arranged;
the characteristic data of a typical service scene is input into the VGGNet according to the time sequence, data of a future period of time after the time period to which the input data belongs is predicted, and the intermediate layer network can memorize the calculation state of the input data at a certain moment and transmit the calculation state to the calculation process of the data at the next moment so as to form the contact in the time relationship.
S3, constructing a power failure early warning prediction model; the power distribution network digital twin system built based on the S2 selects the data processed through the countermeasure thought as input, the key characteristic quantity which influences the power failure of the power distribution network is screened according to the sorting result of the distance correlation coefficient and is used as the input quantity of the power failure risk early warning based on the convolutional neural network, the output is the power failure risk level of the specific line of the power distribution network, and when the risk level exceeds a set threshold value, the power failure early warning is triggered;
the distance correlation coefficient in the S3 can measure the degree of nonlinear correlation of two variables, the value of the distance correlation coefficient is in [0,1], when the coefficient is 0, the two variables are mutually independent, otherwise, when the coefficient is 1, the two variables are strongly correlated;
according to the definition of the distance correlation coefficient, sequentially calculating the correlation coefficient of the input data to each line during power failure, and sorting and screening out the key characteristic quantity which influences each line during power failure according to the correlation degree;
the distance correlation coefficient is calculated by the following formula:
Figure RE-GDA0002788993990000071
Figure RE-GDA0002788993990000072
Figure RE-GDA0002788993990000073
Figure RE-GDA0002788993990000074
where dCor (X, Y) represents the distance correlation coefficient of two variables X, Y, dCov (X, Y) represents the distance covariance, dvar (X) and dvar (Y) represent the distance variance of X and Y, respectively, and n represents the length of the variables X and Y.
The power failure risk early warning model based on the convolutional neural network divides the power failure risk level of the power distribution network into 4 levels, namely, the power failure risk level is particularly great, larger and general. And classifying the historical data according to the 4 grades, wherein the class labels are the respective power failure risk grades.
After the convolutional neural network is trained according to the multi-classification tasks, the power failure risk level of the specific line of the power distribution network can be obtained from key characteristic quantity data which affect the power failure of the power distribution network.
And when the power failure risk level exceeds a set threshold value, triggering power failure early warning.
S4, constructing a line loss calculation and early warning model; the power distribution network digital twin system constructed based on the S2 takes data processed through the generated countermeasure concept as input, fits a power flow equation and state estimation through a neural network, calculates the current value of a line, calculates line loss as final output, and triggers line loss abnormity early warning when the line loss value exceeds a set threshold value.
The neural network in the S4 adopts a long and short memory neural network, the input is the line parameter and the electrical measurement in the power distribution network, and the output is the current value of the power distribution network line.
The line loss calculation formula in S4 is:
Figure RE-GDA0002788993990000081
wherein, Δ WLa,j，ΔWLb,i，ΔWLc,iThe loss of the A, B and C phases of the line i is respectively; ra,i，Rb,i，Rc,iThe three-phase resistors A, B and C of the line i are respectively;
Figure RE-GDA0002788993990000082
the three-phase state currents of A, B and C of the line i in the period j are respectively; n is the total time segment number; t is each time interval.
Example 3
Referring to fig. 1 to 4, the invention provides a typical service scene early warning method for a power distribution network based on a digital twin body, and a countermeasure idea is generated through a zero sum game between a generator and a discriminator, so that the original data distribution rule can be reproduced, and a good way is provided for complementing missing data.
The digital twin is a product that arises in response to a large data background. The digital twins are also called digital twins, digital mirror images and the like, are holographic mapping from a real space to a virtual space of a complex physical entity, and can construct virtual-real dynamic interactive links through multi-source information, mass data, sensing measurement and the like. The method integrates physical feedback data, assists artificial intelligence, machine learning and software analysis, and establishes a digital simulation in an information platform, so that the physical entity can be known, analyzed and optimized. The simulation automatically makes corresponding changes as the physical entity changes based on the feedback. Ideally, the digital twin can learn itself from the multiple feedback source data, presenting the true status of the physical entity in the digital world in near real time. The digital twin carries out rapid deep learning and accurate simulation according to massive information feedback, thereby supporting the completion of various research applications which are difficult to develop in the real world in a virtual environment.
Firstly, simulating and supplementing data with insufficient sample quantity by generating countermeasure thought to obtain a generated sample approaching historical data, and simulating a specific operation scene; filling missing samples by generating countermeasure thought for the samples of missing part data, converting the unbalanced data set into a balanced data set, and performing multi-source data fusion; taking data processed by the generated countermeasure idea as input, constructing a three-dimensional mapping digital twin body from a physical world to a virtual world, accurately depicting and simulating multi-scene dynamic behaviors of complex physical entities, dividing an application model according to typical service scenes, comprehensively adopting algorithms such as regression analysis, cluster analysis and deep learning, analyzing and mining the internal rules of the related data of the power utilization system, and taking data required by each typical service scene model as output; taking data processed by the generated countermeasure idea as input, constructing a power failure early warning model based on a convolutional neural network, taking a line power failure risk level as output, and triggering power failure early warning when the risk level exceeds a set threshold; and taking the data processed by the generated countermeasure idea as input, constructing a line loss calculation and early warning model, taking a line loss calculation result as output, and triggering line loss abnormity early warning when the line loss value exceeds a set threshold value.
Fig. 2 is a schematic diagram of the countermeasure idea generated in the embodiment of the present invention. A generator and a discriminator are adopted to carry out zero-sum game, the generator learns real measurement samples and then generates simulation data, the simulation data and the real measurement data are used as the input of the discriminator, the discriminator judges the truth and the falseness, the discriminator and the generator continuously progress in mutual confrontation, and finally balance is achieved, and the regular repeated carving of the simulation data on the real data is realized.
During game play confrontation, the generators are trained in accordance with minimizing generator loss functions and the discriminators are trained in accordance with minimizing discriminator loss functions. Through carrying out alternating iterative optimization training on the generator and the discriminator, the Nash equilibrium point is finally reached between the generator and the discriminator, namely, the generator can synthesize artificial samples which are difficult to discriminate by the discriminator.
According to the above, the loss functions of the generator and the discriminator are respectively:
Figure RE-GDA0002788993990000091
Figure RE-GDA0002788993990000092
in the above two formulae, z to Pz(z) represents a noise vector z, x-P sampled from a normal distributionrRepresenting a true sample sampled from a true data distribution; e (-) represents the calculation expectation; g (z) represents the sample generated by the generator, and x or G (z) is input into the discriminator D to obtain an output result D (-).
Fig. 3 is a schematic diagram of a power outage early warning model in an embodiment of the present invention. Based on a digital twin system of the power distribution network, data processed through a generation countermeasure idea is used as input, key characteristic quantities influencing power failure of the power distribution network are screened according to a sorting result of distance correlation coefficients and used as input quantities of power failure risk early warning based on a convolutional neural network, the input quantities are output as power failure risk levels of specific lines of the power distribution network, and when the risk levels exceed a set threshold value, the power failure early warning is triggered.
The distance correlation coefficient can measure the degree of nonlinear correlation of two variables, the value of the distance correlation coefficient is in [0,1], when the coefficient is 0, the two variables are mutually independent, otherwise, when the coefficient is 1, the two variables are strongly correlated. And according to the definition of the distance correlation coefficient, sequentially calculating the correlation coefficient of the input data to each line during power failure, and sorting and screening out the key characteristic quantity which influences each line during power failure according to the correlation degree.
The distance correlation coefficient is calculated by the following formula:
Figure RE-GDA0002788993990000101
Figure RE-GDA0002788993990000102
Figure RE-GDA0002788993990000103
Figure RE-GDA0002788993990000104
in the above four equations, dCor (X, Y) represents a distance correlation coefficient of two variables X, Y, dCov (X, Y) represents a distance covariance, dvar (X) and dvar (Y) represent distance variances of X and Y, respectively, and n represents lengths of the variables X and Y.
The power failure risk early warning model based on the convolutional neural network divides the power failure risk level of the power distribution network into 4 levels, namely, the power failure risk level is particularly great, larger and general. And classifying the historical data according to the 4 grades, wherein the class labels are the respective power failure risk grades.
After the convolutional neural network is trained according to the multi-classification tasks, the power failure risk level of the specific line of the power distribution network can be obtained through key characteristic quantity data which affect the power failure of the power distribution network.
And when the power failure risk level exceeds a set threshold value, triggering power failure early warning.
Fig. 4 is a schematic diagram of a line loss calculation and early warning model according to an embodiment of the present invention. And taking the data processed by the generated countermeasure idea as input, fitting a power flow equation and state estimation in a digital twin body through a neural network, calculating the current value of the line, calculating line loss as final output, and triggering line loss abnormity early warning when the line loss value exceeds a set threshold value.
The neural network adopts a long and short memory neural network, the input is line parameters and electrical measurement in the power distribution network, and the output is the current value of the power distribution network line.
The line loss calculation formula is as follows:
Figure RE-GDA0002788993990000105
in the above formula,. DELTA.WLa,j，ΔWLb,i，ΔWLc,iThe loss of the A, B and C phases of the line i is respectively; ra,i，Rb,i，Rc,iThe three-phase resistors A, B and C of the line i are respectively;
Figure RE-GDA0002788993990000111
the three-phase state currents of A, B and C of the line i in the period j are respectively; n is the total time segment number; t is each time interval.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (10)

1. A power distribution network typical service scene early warning method based on digital twin is characterized by comprising the following steps,
s1, multi-source data generation and fusion are carried out based on the countermeasure generating idea, the countermeasure generating idea is applied, historical fault data with insufficient sample quantity are simulated and supplemented, real fault data are sampled and trained, a generated sample approaching historical data is obtained, and a specific operation scene is simulated;
s2, building a power distribution network digital twin system framework, and building three-dimensional mapping of a physical entity in a virtual digital world according to the multi-source data supplemented and fused in the S1;
s3, constructing a power failure early warning prediction model; the power distribution network digital twin system built on the basis of S2 selects key characteristic quantities which are processed through the generated countermeasure thought as input according to the sorting result of the distance correlation coefficient and used for screening the key characteristic quantities which influence the power failure of the power distribution network, outputs the key characteristic quantities as input quantities of power failure risk early warning based on a convolutional neural network, outputs the power failure risk grades of specific lines of the power distribution network, and triggers the power failure early warning when the risk grades exceed a set threshold;
s4, constructing a line loss calculation and early warning model; the power distribution network digital twin system constructed based on the S2 takes data processed by the countermeasure idea as input, fits a power flow equation and state estimation through a neural network, calculates the current value of a line, calculates the line loss as final output, and triggers line loss abnormity early warning when the line loss value exceeds a set threshold value.
2. The pre-warning method for the typical service scene of the power distribution network based on the digital twin body according to claim 1, wherein in the step S1, for data loss caused by a fault of a measurement and acquisition device, a countermeasure thought is generated, a missing sample is filled, an unbalanced data set is converted into a balanced data set, and multi-source data fusion is performed.
3. The power distribution network typical service scene early warning method based on the digital twin body according to claim 2, wherein the generation countermeasure idea comprises two important components, namely a generator and a discriminator;
the generator is trained to minimize a generator loss function, and the arbiter is trained to minimize an arbiter loss function;
and finally, a Nash equilibrium point is reached between the generator and the discriminator by performing alternating iterative optimization training on the generator and the discriminator.
4. The power distribution network typical service scene early warning method based on the digital twin body according to claim 3, wherein the loss functions of the generator and the arbiter are respectively,
Figure FDA0002740508530000021
Figure FDA0002740508530000022
wherein, z to Pz(z) represents a noise vector z, x-P sampled from a normal distributionrRepresenting data from realitySampling in the distribution to obtain a real sample; e (-) represents the calculation expectation; g (z) represents the sample generated by the generator, and x or G (z) is input into the discriminator D to obtain an output result D (-).
5. The power distribution network typical service scene early warning method based on the digital twin body according to claim 4, wherein in the step S1, the simulation and supplement of the historical fault data with insufficient sample number is specifically to input a random noise vector z into a trained generator to obtain generated data corresponding to a missing position in real data, and the finally filled measured data is composed of an actual measured part in an original sample and a data missing part in a generated sample.
6. The power distribution network typical service scene early warning method based on the digital twin body according to claim 3, wherein S2 builds a power distribution network digital twin system framework, and establishes three-dimensional mapping of a physical entity in a virtual digital world according to supplemented and fused multi-source data, specifically, index data in the power distribution network is mapped into a virtual digital space, and an application model is divided according to a typical service scene.
7. The pre-warning method for the typical service scene of the power distribution network based on the digital twin body according to claim 6, wherein in the step S2, a VGGNet convolutional neural network is adopted to conduct regression and classification analysis on data, the intrinsic law of the data of the power utilization system is analyzed and mined, and prediction analysis is conducted on performance indexes which cannot be measured or directly calculated by the system;
the VGGNet convolutional neural network successfully constructs a convolutional neural network with 16-19 layers of depth through repeated stacking of 3 x 3 small convolutional kernels and 2 x 2 maximum pooling layers;
the VGGNet convolutional neural network divides the network into 5 sections, each section is formed by connecting a plurality of 3 x 3 convolutional networks in series, each section of convolution is followed by a maximum pooling layer, and finally, 3 full-connection layers and a softmax layer are arranged;
and inputting the characteristic data of a typical service scene into the VGGNet according to the time sequence, predicting data of a future period of time after the time period to which the input data belongs, memorizing the calculation state of the input data at a certain moment by the intermediate layer network, and transmitting the calculation state to the calculation process of the data at the next moment so as to form the contact in the time relationship.
8. The power distribution network typical service scene early warning method based on the digital twin body according to claim 7, wherein the distance correlation coefficient in the step S3 is taken within [0,1], and when the coefficient is 0, it indicates that the two variables are independent of each other, otherwise, when the coefficient is 1, it indicates that the two variables are strongly correlated;
according to the definition of the distance correlation coefficient, sequentially calculating the correlation coefficient of the input data to each line during power failure, and sorting and screening out the key characteristic quantity which influences each line during power failure according to the correlation degree;
the distance correlation coefficient is calculated by the following formula:
Figure FDA0002740508530000031
Figure FDA0002740508530000032
Figure FDA0002740508530000033
Figure FDA0002740508530000034
where dCor (X, Y) denotes a distance correlation coefficient of two variables X, Y, dCov (X, Y) denotes a distance covariance, dvar (X) and dvar (Y) denote the distance variances of X and Y, respectively, and n denotes the lengths of the variables X and Y.
9. The power distribution network typical service scene early warning method based on the digital twin body as claimed in claim 8, wherein the neural network in S4 adopts a long and short memory neural network, and the input is line parameters and electrical measurements in the power distribution network and the output is a current value of a power distribution network line.
10. The pre-warning method for the typical service scene of the power distribution network based on the digital twin body as claimed in claim 9, wherein the line loss calculation formula in S4 is as follows:
Figure FDA0002740508530000035
wherein, Δ WLa,j，ΔWLb,i，ΔWLc,iThe loss of the A, B and C phases of the line i is respectively; ra,i，Rb,i，Rc,iThe three-phase resistors A, B and C of the line i are respectively;
Figure FDA0002740508530000036
the three-phase state currents of A, B and C of the line i in the period j are respectively; n is the total time segment number; t is each time interval.
CN202011148753.9A 2020-10-23 2020-10-23 Digital twin-body-based power distribution network typical service scene early warning method Active CN114492150B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011148753.9A CN114492150B (en) 2020-10-23 2020-10-23 Digital twin-body-based power distribution network typical service scene early warning method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011148753.9A CN114492150B (en) 2020-10-23 2020-10-23 Digital twin-body-based power distribution network typical service scene early warning method

Publications (2)

Publication Number Publication Date
CN114492150A true CN114492150A (en) 2022-05-13
CN114492150B CN114492150B (en) 2024-08-13

Family

ID=81471172

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011148753.9A Active CN114492150B (en) 2020-10-23 2020-10-23 Digital twin-body-based power distribution network typical service scene early warning method

Country Status (1)

Country Link
CN (1) CN114492150B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116341396A (en) * 2023-05-30 2023-06-27 青岛理工大学 Complex equipment digital twin modeling method based on multi-source data fusion
CN116881647A (en) * 2023-09-08 2023-10-13 中国长江电力股份有限公司 Time sequence sample management system construction method for hydropower multi-service scene
CN117543564A (en) * 2023-11-24 2024-02-09 国网江苏省电力有限公司南通供电分公司 Power grid regulation and control method and system based on countermeasure type neural network

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111210359A (en) * 2019-12-30 2020-05-29 中国矿业大学(北京) Intelligent mine scene oriented digital twin evolution mechanism and method
CN114444013A (en) * 2020-10-19 2022-05-06 中国石油化工股份有限公司 Power distribution network big data restoration method based on countermeasure game
CN117610934A (en) * 2023-12-08 2024-02-27 国网辽宁省电力有限公司阜新供电公司 Power grid security risk assessment early warning system

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111210359A (en) * 2019-12-30 2020-05-29 中国矿业大学(北京) Intelligent mine scene oriented digital twin evolution mechanism and method
CN114444013A (en) * 2020-10-19 2022-05-06 中国石油化工股份有限公司 Power distribution network big data restoration method based on countermeasure game
CN117610934A (en) * 2023-12-08 2024-02-27 国网辽宁省电力有限公司阜新供电公司 Power grid security risk assessment early warning system

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
JIANPING HUANG 等: "The Application of Digital Twin on Power Industry", IOP CONF. SERIES: EARTH AND ENVIRONMENTAL SCIENCE 647 (2021) 012015, 1 January 2021 (2021-01-01), pages 1 - 10 *
李炜 等: "大数据背景下智能配电网运营管理方法研究", 华电技术, vol. 43, no. 8, 31 August 2021 (2021-08-31), pages 33 - 40 *
王小月;孙路遥;许恒昌;: "5G在智慧电力场景中的应用", 通信企业管理, no. 03, 10 March 2020 (2020-03-10), pages 56 - 59 *
王文礼;刘娜;郝磊;: "山东寿光：构建农业智慧用能生态圈", 中国电力企业管理, no. 14, 15 May 2020 (2020-05-15), pages 18 - 21 *
钱利宏 等: "基于多源数据融合与深度信念网络的 台区停电故障预警方法研究", 广东电力, vol. 35, no. 12, 31 December 2022 (2022-12-31), pages 32 - 40 *
顾永涛 等: "基于生成对抗网络的光伏出力区间预测方法", 哈尔滨理工大学学报, vol. 26, no. 2, 30 April 2021 (2021-04-30), pages 100 - 109 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116341396A (en) * 2023-05-30 2023-06-27 青岛理工大学 Complex equipment digital twin modeling method based on multi-source data fusion
CN116341396B (en) * 2023-05-30 2023-08-11 青岛理工大学 Complex equipment digital twin modeling method based on multi-source data fusion
CN116881647A (en) * 2023-09-08 2023-10-13 中国长江电力股份有限公司 Time sequence sample management system construction method for hydropower multi-service scene
CN117543564A (en) * 2023-11-24 2024-02-09 国网江苏省电力有限公司南通供电分公司 Power grid regulation and control method and system based on countermeasure type neural network
CN117543564B (en) * 2023-11-24 2024-07-30 国网江苏省电力有限公司南通供电分公司 Power grid regulation and control method and system based on countermeasure type neural network

Also Published As

Publication number Publication date
CN114492150B (en) 2024-08-13

Similar Documents

Publication Publication Date Title
CN114492150B (en) Digital twin-body-based power distribution network typical service scene early warning method
CN108399248A (en) A kind of time series data prediction technique, device and equipment
CN110542819B (en) Transformer fault type diagnosis method based on semi-supervised DBNC
CN113688869B (en) Photovoltaic data missing reconstruction method based on generation countermeasure network
CN106656357B (en) Power frequency communication channel state evaluation system and method
CN112668809B (en) Method for establishing autism children rehabilitation effect prediction model
CN113011397B (en) Multi-factor cyanobacterial bloom prediction method based on remote sensing image 4D-Fractalnet
CN110321940A (en) The feature extraction of aircraft telemetry and classification method and device
CN113505477A (en) Process industry soft measurement data supplementing method based on SVAE-WGAN
CN113125960A (en) Vehicle-mounted lithium ion battery charge state prediction method based on random forest model
CN111178585A (en) Fault reporting amount prediction method based on multi-algorithm model fusion
CN109638826B (en) CNN-based power grid topology identification method and system
CN112150304A (en) Power grid running state track stability prejudging method and system and storage medium
CN115392349A (en) Fault diagnosis method and device for cutting part of heading machine and heading machine
CN114297947A (en) Data-driven wind power system twinning method and system based on deep learning network
CN110321493A (en) A kind of abnormality detection of social networks and optimization method, system and computer equipment
CN116186633A (en) Power consumption abnormality diagnosis method and system based on small sample learning
Ardeshiri et al. Gated recurrent unit least-squares generative adversarial network for battery cycle life prediction
CN115438897A (en) Industrial process product quality prediction method based on BLSTM neural network
CN117669384A (en) Intelligent monitoring method and system for temperature sensor production based on Internet of things
CN114548494A (en) Visual cost data prediction intelligent analysis system
CN117473430A (en) Non-invasive load classification method and device
CN116840724A (en) Lithium ion battery health condition estimation and residual service life prediction method
CN108760813B (en) Gas turbine blade health monitoring system and method based on temperature signals
CN115409317A (en) Transformer area line loss detection method and device based on feature selection and machine learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant