CN113435575A - Gate graph neural network transient stability evaluation method based on unbalanced data - Google Patents

Gate graph neural network transient stability evaluation method based on unbalanced data Download PDF

Info

Publication number
CN113435575A
CN113435575A CN202110693635.4A CN202110693635A CN113435575A CN 113435575 A CN113435575 A CN 113435575A CN 202110693635 A CN202110693635 A CN 202110693635A CN 113435575 A CN113435575 A CN 113435575A
Authority
CN
China
Prior art keywords
model
data
transient stability
unstable
sample
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
CN202110693635.4A
Other languages
Chinese (zh)
Other versions
CN113435575B (en
Inventor
孙迪
李俊
关心
周小梅
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.)
Mudanjiang university
State Grid Heilongjiang Electric Power Co Ltd
Heilongjiang University
Original Assignee
Mudanjiang university
State Grid Heilongjiang Electric Power Co Ltd
Heilongjiang 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 Mudanjiang university, State Grid Heilongjiang Electric Power Co Ltd, Heilongjiang University filed Critical Mudanjiang university
Priority to CN202110693635.4A priority Critical patent/CN113435575B/en
Publication of CN113435575A publication Critical patent/CN113435575A/en
Application granted granted Critical
Publication of CN113435575B publication Critical patent/CN113435575B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • 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)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Educational Administration (AREA)
  • Development Economics (AREA)
  • Public Health (AREA)
  • Probability & Statistics with Applications (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

A portal neural network transient stability assessment method based on unbalanced data belongs to the technical field of transient stability analysis of power systems. The invention aims to solve the problem that the current machine learning method has no interpretability. The invention generates unstable samples based on a conditional generation countermeasure network (CGAN), not only can generate unstable samples, but also can be used for generating unstable samples with unbalanced events, so that the samples not only reach stable and unstable balance, but also reach the balance of the events in the unstable samples. After the problem of data imbalance of the samples is solved, the transient stability of the power system is evaluated by using a GGNN algorithm, and the reason causing instability of the power system is judged. The method is mainly used for transient stability evaluation of the power system.

Description

Gate graph neural network transient stability evaluation method based on unbalanced data
Technical Field
The invention relates to a network transient stability evaluation method of a power system, and belongs to the technical field of transient stability analysis of the power system.
Background
Transient stability refers to whether a power system can reach a steady-state operation state or an original operation state through a transient process after being suddenly subjected to large interference in a certain operation state. The power system is an important component in the energy Internet of things, and the transient stability evaluation of the power system is also important. These large disturbances are typically referred to as short-circuit faults, sudden line or generator disconnections, etc. If the power system can not reach a stable or original running state after being greatly disturbed, the power system can be destabilized, the power system can be broken down in a large area, and serious social and economic losses are caused. It is therefore necessary and important that the designed TSA is able to quickly determine whether the system is unstable and analyze the event of instability when the system is subject to a fault.
Methods for analyzing the transient stability of the power system include a time domain simulation method, a direct method and a machine learning method, wherein the most mature method is the time domain simulation method. The time domain simulation method is to build a model for the whole system, namely a high-dimensional nonlinear differential algebraic equation, obtain a curve of each variable in the system changing along with time after solving the equation, and then judge the transient stability of the system. Because the model is established according to the physical characteristics and the network topological relation of each element in the system, the time domain simulation method has interpretability. The time domain simulation method can be used as a test standard of other methods, but with the increasing complexity of the power system and the rapid change of electronic hardware, the model of the time domain simulation method is increasing complex, the calculation is also increasing complex, and the required calculation time is also increasing long, so that the time domain simulation method cannot judge the transient stability of the power system in real time. In addition, the time domain simulation method cannot obtain the transient stability margin of the power system.
The direct method is generally based on an energy function, and there are three methods based on the energy function, namely a potential energy boundary surface method (PEBC), a control unstable equilibrium point method (cup) and an EEAC. Compared with a time domain simulation method, the direct method has higher calculation speed because the motion trail of the whole system does not need to be gradually integrated; and because the direct method is to establish an energy function firstly and then judge the transient stability problem of the system by comparing the energy of the system at the fault removal moment with the energy of the system when the system is in a critical state, the stability margin of the system can be obtained. However, the direct method has poor model adaptability, the judgment result is conservative, and the energy of the system in the critical state is difficult to obtain.
With the development of artificial intelligence, various methods of machine learning are also used for transient stability analysis of power systems, such as SVM, logistic regression, deep learning, and the like. In the Delay Aware Transient Stability Assessment System of J.Q.Yu, A.Y.S.Lam, D.J.Hill and V.O.K.Li et al, the LSTM method is used to solve the problems of Transient Stability Assessment in the power System, the conversion of the Assessment accuracy and response time of the System model and the communication Delay of PMUs. The cascaded CNN based approach addresses fast batch assessment of transient stability of power systems. As graph learning evolves, graph learning is also applied to transient stability assessment of power systems. In Fast Transient Stability Assembly Using masked connected Neural Networks, the TSA model was constructed Using the Cyclic graph convolution network (RGCN). Compared with other neural network methods, the influence of the topological structure of the power grid on the stability of the power system is considered on the basis of the TSA model of the graph attention neural network, and better accuracy can be achieved to a certain extent. Compared with a time domain simulation method and a direct method, the machine learning method has the advantages of low calculation complexity and capability of obtaining the stability margin, has shorter response time, and can also realize real-time judgment of the transient stability of the power system. However, the current machine learning method is not interpretable, and meanwhile, the machine learning method is greatly influenced by data imbalance, so that although the evaluation model has higher accuracy due to data imbalance, wrong judgment can occur to cause larger loss, and meanwhile, the generalization capability of the trained model is limited, so that the robustness and the applicability of the trained model are to be improved.
Disclosure of Invention
The invention aims to solve the problem that the current machine learning method has no interpretability.
The gate map neural network transient stability evaluation method based on the imbalance data comprises the following steps:
s1, collecting stability evaluation data of the power system, wherein the stability evaluation data comprise bus voltage U1The operation parameters of the motor comprise active power P of the motor, reactive power Q of the motor, voltage amplitude U of the motor and current I of the motor;
s2, the stability evaluation of the power system is realized by using the transient stability evaluation model, and the process of realizing the stability evaluation of the power system by using the transient stability evaluation model comprises the following steps:
generating a graph G (V, E) based on the stability evaluation data, V represents the generator set and V belongs to V, and selecting the generator set as a node V, EvwE is an edge between the nodes v and w; the active power P of the motor, the reactive power Q of the motor, the voltage amplitude U of the motor and the current I of the motor are used for forming an annotation vector of a node, and the bus voltage U1For generating edges between nodes;
based on the graph G (V, E), firstly, evaluating the unstable state of transient stability by using a first GGNN model, and entering the next cycle if the output of the model is the transient stability of the power system; otherwise, evaluating and continuously judging the event type causing the system instability by using a second GGNN model;
the first GGNN is a two-classification model, and the output is a stable state and an unstable state; the second GGNN model is a multi-classification model, the output is an unstable type, and the reason for causing the instability of the power system is correspondingly judged;
the transient stability evaluation model is trained in advance, and the training process comprises the following steps:
collecting stability evaluation data of the power system as sample data, wherein the operation parameters of the motor in the sample data further comprise the relative rotor angle delta of each generator, and dividing the steady state of the power system into a steady state and an unstable state by using a transient stability index based on the sample data of the relative rotor angle delta of each generator;
then generating an unstable sample by using the first CGAN according to the data of the unstable state, and generating unstable samples of different events by using the second CGAN;
and generating a graph G (V, E) by using the samples in the stable state and the samples corresponding to the unstable state, and training a first GGNN model for evaluating transient stability and a second CGAN model for evaluating events corresponding to the unstable samples.
Further, the first GGNN uses a gate recursion unit GRU in the propagating step.
Further, the second GGNN uses a gate recursion unit GRU in the propagating step.
Further, a loss function in the process of training the first GGNN model
Figure RE-GDA0003214735820000031
Wherein liLabel representing sample i, hiThe graph level representing sample i represents a vector, i represents sample i, and G represents the sample size.
Further, the propagation recursion process of the first GGNN model is as follows:
Figure RE-GDA0003214735820000032
wherein,
Figure RE-GDA0003214735820000033
expressed as hidden state of node v at time t, xvAnnotating the vector for the node of node v; t represents the transposition of the vector;
Figure RE-GDA0003214735820000034
the node v collects information from the neighbor nodes;
Figure RE-GDA0003214735820000035
the vector represents information collected by the node v at time t about the neighbor node; a. thevIs a sub-matrix of the graph adjacency matrix and represents the connection state of the node v and the neighbor nodes thereof; v represents the number of nodes;
Figure RE-GDA0003214735820000036
representing the hidden state of node 1 at time t-1, V represents the number of nodes,
Figure RE-GDA0003214735820000037
representing a vector
Figure RE-GDA0003214735820000038
B denotes calculation
Figure RE-GDA0003214735820000039
The vector of (a);
Figure RE-GDA00032147358200000310
Figure RE-GDA00032147358200000311
wherein
Figure RE-GDA00032147358200000312
Information representing the update of node v at time t,
Figure RE-GDA00032147358200000313
information representing the node v at time t, matrix WzAnd WrWeight matrix for calculating z and r, matrix UzAnd UrA weight matrix also used to compute z and r; z and r denote update gatesAnd a reset gate, σ representing a sigmoid function;
Figure RE-GDA00032147358200000314
Figure RE-GDA0003214735820000041
where tanh (x) is an activate function, an element-by-element multiplication operation;
when the graph level is output, the representation vector of the graph level is defined as
Figure RE-GDA0003214735820000042
Wherein,
Figure RE-GDA0003214735820000043
acting as a soft attention mechanism that decides which nodes are associated with the current graph-level task, i and j being
Figure RE-GDA0003214735820000044
And xvCascading neural networks as inputs and outputting real values; h isgThe method is used for judging the transient stability of the power system.
Further, h of the first GGNN modelg> 0.5 indicates stabilization.
Further, a loss function in the process of training the second GGNN model
Figure RE-GDA0003214735820000045
Wherein l1Labels expressed as event types, k denotes the number of event types, ajIs the output of the activation function softmax.
Further, before training the first GGNN model and the second CGAN model, the first CGAN and the second CGAN used for generating the unstable samples are trained in advance, and the training process includes the following steps:
constructing data x of the CGAN according to the collected stability evaluation data of the power system as sample data:
x＝[x1,x2,…,xn] (2)
Figure RE-GDA0003214735820000046
ai＝[U1 P Q U I δ]1×89 (4)
wherein n represents the number of input samples, and the input data is n arrays of 500 × 89 according to the formulas (2), (3) and (4);
the input condition of a discriminator D of the CGAN is a label vector of sample data, and the label vector is an n multiplied by 1 array;
generating multi-label data based on sample data, wherein each piece of data has 2 labels:
L＝{l1,l2} (5)
wherein l1Representing events causing the state of data,/2Represents the steady state of the system;
l1＝{0，1，2，3，4} (6)
l1an event represented by 0 is a short-circuit event, l1The event denoted 1 is a tap event, l1The event denoted 2 is a load event, l1The event denoted by 3 is a switching event, l1The event denoted 4 is a synchronous generator event;
l2＝{0，1} (7)
when l is2At 0, the sample is destabilized, when l2When 1, the sample is stable; l is an array with 1 row and 2 columns;
and training to generate a CGAN model corresponding to the unstable sample and a CGAN model corresponding to the unstable sample generating different events based on the sample data and the corresponding label, and respectively recording the CGAN models as a first CGAN model and a second CGAN model.
Further, in the process of collecting the stability evaluation data of the power system, the fault clearing time is recorded as 0 time, and the bus voltage and the motor operation parameters are collected according to the time interval of 0.01 s.
Further, the transient stability index
Figure RE-GDA0003214735820000051
Wherein deltamaxThe maximum relative rotor angle of any two generator sets in the simulation time length is obtained.
Has the advantages that:
the invention generates unstable samples based on a conditional generation countermeasure network (CGAN), not only can generate unstable samples, but also can be used for generating unstable samples with unbalanced events, so that the samples not only reach stable and unstable balance, but also reach the balance of the events in the unstable samples. After the problem of data imbalance of the samples is solved, the transient stability of the power system is evaluated by using a GGNN algorithm, and the reason causing instability of the power system is judged. More importantly, the invention considers the influence of different types of data, so that the invention can have higher accuracy and lower error rate. In addition, the CGAN can solve the problem of data imbalance, gives consideration to the category information when the CGAN is unstable, and is matched with the subsequent GGNN, so that the CGAN has better robustness and applicability to an evaluation model of a power system.
Through experiments, the CGAN is capable of effectively improving the transient stability evaluation performance and solving the problem of data imbalance of the transient stability evaluation of the power system. The GGNN can also quickly and accurately evaluate the transient stability of the power system, judge the cause of the instability of the power system and achieve good performance.
Drawings
Fig. 1 is a structure of a conditional generation network CGAN;
FIG. 2 is a plot of relative rotor angle for various motors of a stable sample after an IEEE-39 bus power system over England has been subjected to a short circuit event;
FIG. 3 is a plot of relative rotor angle for various motors of an unstable sample of an IEEE-39 bus power system over England after being subjected to a short circuit event;
FIG. 4 is a flow chart of the generation of an unstable sample by a CGAN;
FIG. 5 is a framework of a TSA model;
FIG. 6 is a diagram configuration corresponding to the configuration of an IEEE-39 bus system;
FIG. 7 is a structure of a GRU;
FIG. 8 is a flow chart of the model determining transient stability and predicting events that cause system instability;
FIG. 9 shows the structure of a new England 39 bus system
FIG. 10 is an iterative loss plot for the generative model G;
FIG. 11 is a graph of a loss iteration for discriminant model D;
FIG. 12 is a two-dimensional data spatial distribution of CGAN-generated samples;
FIG. 13 is the accuracy of the proposed model at different destabilized sample scales;
fig. 14 shows the results of the reasons for unstable samples predicted by the GGNN model.
Detailed Description
The first embodiment is as follows:
the transient stability evaluation method of the gate map neural network based on the unbalanced data includes the following steps:
firstly, generating unstable samples by using a first CGAN model, and then generating unstable samples of different events by using a second CGAN model:
1. and (3) constructing a condition generation countermeasure network, namely CGAN:
the deep learning model can represent the probability distribution of various data in artificial intelligence application, and the addition of the discrimination model in the deep learning process can enable the probability distribution of various data represented by the model to be more accurate. The generation of the countermeasure network is that a discriminant model is added in deep learning, the existing generation model (G) and the discriminant model (D) in the generation of the countermeasure network are generated, and Nash balance is achieved through continuous countermeasures between the generation model G and the discriminant model D, so that the data distribution generated by the model accords with real data distribution.
Conditional Generative adaptive Nets (Conditional generic adaptive Nets) are an extension of the original GAN, adding extra information y (conditions) to both the Generative model G and the discriminative model D. In the present invention, the condition is the category of sample data, i.e., the tag information.
Fig. 1 shows the structure of a conditional generation network CGAN. As shown in fig. 1, the label y of the exemplar is part of the input layer for the discriminant model D and the generative model G. In the generated model G, the priori input noise z and the sample label y jointly form a joint hidden layer characterization. In the discriminant model D, the input layer is composed of real data x, label information y, and data G (z | y) generated by the generative model. The condition y is added into a condition generating network (CGAN) to enable the generation of the wanted data, the power system has robustness, so that the unstable sample data is less, and in order to solve the problem of data imbalance, the unstable sample data is firstly obtained through the CGAN.
And generating a model G for acquiring the real distribution of the sample data, and generating similar real training data by using the noise z and the condition y which obey a certain random distribution, wherein the pursuit effect is that the closer the real data is, the better the real data is.
The discrimination model D is a two-classifier, which judges the probability that one data comes from real training data, if the data comes from the real training data, D outputs a high probability, otherwise, D outputs a low probability. Models G and D were trained simultaneously: fixing the discriminant model D, adjusting parameters of G such that the expectation of log (1-D (G (z | y))) is minimized; the fixed generative model G adjusts the parameters of D such that the expectation of logD (x | y) + log (1-D (G (z | y))) is maximized. This optimization process can be expressed as:
Figure RE-GDA0003214735820000071
wherein,
Figure RE-GDA0003214735820000072
representing the mathematical expectation of logD (x | y),
Figure RE-GDA0003214735820000073
the mathematical expectation of log (1-D (G (z | y))), V (D, G) is a cost function.
2. Acquiring data description:
the data are observed and transmitted in real time by means of Phase Measurement Units (PMUs). When transient stability of the power system is evaluated on line, input data are measured by PMUs 5 seconds after fault clearing, and then transient stability of the power system is predicted according to the input data. In the present invention, data is generated using time domain simulations, and FIG. 2 is a plot of relative rotor angle for each motor of a stable sample of an IEEE-39 British Power System after being subjected to a short circuit event. FIG. 3 is a plot of relative rotor angle for various motors of an unstable sample of an IEEE-39 bus power system over England after being subjected to a short circuit event.
The stability criterion of the transient stability of the power system is the magnitude of the relative rotor angle or generator rotor angle according to the Transient Stability Index (TSI). . Because a large amount of data is needed to train the model, so that the model can accurately judge the transient stability of the power system, the bus voltage U is selected and used in the invention1The active power P of the motor, the reactive power Q of the motor, the voltage amplitude U of the motor, the current I of the motor and the relative rotor angle delta of each generator are characteristics of input data. The system selected by the invention is the IEEE-39 bus of England, so that
Figure RE-GDA0003214735820000074
P, Q, U, I, δ are the operating parameters of the electrical machines, and the IEEE-39 bus system has 10 electrical machines, so that P ═ P (P)1,p2,…,p10)，Q＝(q1,q2,…,q10)，U＝(u1,u2,…,u10)， I＝(i1,i2,…,i10)，δ＝(δ1,δ2,…,δ10)。
The time of the system suffering the fault is-0.01 s, the time 0 is the fault clearing time, and the time interval is 0.01 s. The input data of CGAN is shown below
x＝[x1,x2,…,xn] (2)
Figure RE-GDA0003214735820000081
ai＝[U1 P Q U I δ]1×89 (4)
Where n represents the number of input samples, and the input data is n arrays of 500 × 89 as can be seen from expressions (2), (3), and (4). The input condition of the discriminator D is a label vector of sample data, which is an n × 1 array.
3. Generation of multi-label data:
the research on the PMUs data shows that a stably running power system may lose stability under the condition of large disturbance, and the large disturbance is an emergency causing system instability. Therefore, each piece of data is set to have 2 tags:
L＝{l1,l2} (5)
wherein l1Representing events causing the state of data,/2Indicating a steady state of the system.
l1＝{0，1，2，3，4} (6)
l1An event represented by 0 is a short-circuit event, l1The event denoted 1 is a tap event, l1The event denoted 2 is a load event, l1The event denoted by 3 is a switching event, l1The event denoted 4 is a synchronous generator event.
l2＝{0，1} (7)
When l is2At 0, the sample is destabilized, when l2At 1, the sample stabilized. From the above equations (5), (6) and (7), L is an array of 1 row and 2 columns.
4. Training a CGAN model corresponding to an unstable sample and CGAN models corresponding to unstable samples generating different events, and respectively recording the models as a first CGAN model and a second CGAN model;
the research on the data finds that the accuracy of the transient stability evaluation model of the power system is related to whether the instability sample and the stable sample in the data are balanced or not, and also finds that the accuracy of the stability evaluation model for judging the instability reason of the power system is related to the balance of each event in the instability sample.
Thus, in balancing samples, the invention is based first on l2Training a CGAN used for generating an unstable sample, and recording as a first CGAN model; based on l1And training the CGAN used for generating unstable samples of different types, and recording the CGAN as a second CGAN model. Fig. 4 is a flow chart illustrating the generation of unstable samples by the CGAN.
When an unstable sample is generated, the condition y is l2The optimization process of the cost function is
Figure RE-GDA0003214735820000091
Wherein l2∈(0,1)；
When the balance of each event in the destabilized sample is taken into consideration, the condition is that y is l1The optimization process of the cost function is
Figure RE-GDA0003214735820000092
Wherein l1∈(0,1,2,3,4)。
In the actual generation of unstable samples, first an unstable sample is generated by the first CGAN, and then an unstable sample of a different event is generated by the second CGAN.
And step two, performing instability evaluation by using the transient stability evaluation model, wherein in the instability evaluation process, firstly, a graph neural network is generated based on the bus voltage and the operation parameters of the motor, then, the instability state of the transient stability is evaluated by using the first GGNN model, and when the instability state is unstable, the type of the instability state is evaluated by using the second GGNN model:
1. constructing a transient stability evaluation model:
the transient stability evaluation model is a TSA model built based on two GGNN models; transient stability evaluation of the power system is to judge that the system cannot return to a stable operation state within 0 to 10 seconds after the fault is removed. Heretofore, there have been many machine learning and data mining based methods for achieving transient stability assessment of power systems. The key to research the transient stability of the power system is to accurately judge the transient stability of the power system in a short time as much as possible. In the present invention, the data 5 seconds after fault clearance is selected for use in evaluating the transient stability of the system. The framework of the TSA model adopts a mode of off-line training model and on-line application model. Online TSA has two modes, periodic and real-time updates. And (3) periodic updating: after training, the trained TSA model can be deployed quickly when real-time data arrives, and the model is updated periodically with the number of newly collected samples. And (3) real-time updating: the off-line training is performed by a large amount of sample data, but the model can be updated on-line, since the model performs transient stability assessment for the power system on-line. The latter can update the model quickly, but has no generality.
In the invention, when the model is applied online, a method for updating the model regularly is selected. As shown in fig. 5, the framework of the proposed TSA model trains the CGAN model and the GGNN model offline, and when online application is performed, the trained GGNN transient stability evaluation model is directly used, and newly evaluated PMU data is used to periodically update the trained TSA model.
(1) Constructing a graph neural network:
the graph neural network is proposed based on the neural network and is used for processing and analyzing data of a graph structure. The invention selects an IEEE-39 bus system to test whether the TSA model is effective or not, and the IEEE-39 bus system is also a system frequently selected for testing a transient stability evaluation algorithm of a power system. The structure and diagram of the IEEE-39 bus system is shown in FIG. 6, which has 39 buses, 10 banks, 19 loads and 34 transmission lines.
Construct graph G (V, E) with IEEE-39 bus System: v represents a genset and V ∈ V, then V is a value from 1 to 10. The generator sets are selected as nodes v, each node v being defined by its characteristics and the associated node. e.g. of the typevwE is an edge between nodes v and w, the edge representing the relationship between the nodes. The generator sets are connected byThe lines are connected to transmission lines and the relationship between the nodes can be represented by the number of buses and transmission lines. If the number of buses and the length of the transmission line are large, the correlation between the nodes is small, and edges between the nodes are generated based on the bus voltage U1. The graph G (V, E) is an undirected graph because the units are interconnected by transmission lines. The goal of the Graphical Neural Network (GNN) is to learn the hidden state
Figure RE-GDA0003214735820000101
Wherein
Figure RE-GDA0003214735820000102
Containing information for each node v neighbor. h isvIs a vector of v, s dimensions of the node that can be used to produce the output ovIn the present invention, ovIs the node score of the node. The graph G (V, E) has a graph-level label, i.e., label l of the sample graph dataG，lG＝l2When the tag is 0,1, the power system is unstable, and when the tag is 1, the power system is stable. The following node characteristics are selected to describe the node, called node annotation xvAnd these annotations are represented using a vector x. The annotation vector for node v is
xv＝[pv,qv,uv,iv]T (10)
Wherein p, q, u, i are respectively expressed as active power, reactive power, voltage and current of the node v.
(2) Building a first GGNN model:
in the transient stability evaluation of the power system, the nodes are in a fixed and static graph structure, the characteristics are time sequence data and dynamic input information, and the stability evaluation of the invention is a classification problem. In the invention, the transient stability evaluation of the power system is judged by using the GGNN.
Ggnn (gated Graph Neural network), using a gate recursion unit GRU in the propagation step, developing a recursion of the number of steps T, and using back propagation to compute the gradient. Fig. 7 shows a structure of a GRU.
The basic recurrences of the propagation model in the transient stability assessment model of the power system are given by the equations (11) to (16):
Figure RE-GDA0003214735820000103
wherein,
Figure RE-GDA0003214735820000104
expressed as hidden state of node v at time t, xvThe vector is annotated for the nodes of node v, T represents the transpose of the vector.
Figure RE-GDA0003214735820000105
Node v collects information from neighboring nodes.
Figure RE-GDA0003214735820000111
The vector represents the information collected by node v at time t about the neighboring nodes. A. thevIs a sub-matrix of the graph adjacency matrix and represents the connection state of the node v and the neighboring nodes thereof. V represents the number of nodes and is,
Figure RE-GDA0003214735820000112
representing the hidden state of node 1 at time t-1, V represents the number of nodes,
Figure RE-GDA0003214735820000113
representing a vector
Figure RE-GDA0003214735820000114
B denotes calculation
Figure RE-GDA0003214735820000115
The vector of (2).
Figure RE-GDA0003214735820000116
Figure RE-GDA0003214735820000117
Wherein
Figure RE-GDA0003214735820000118
Information representing the update of node v at time t,
Figure RE-GDA0003214735820000119
information representing the node v at time t, matrix WzAnd WrWeight matrix for calculating z and r, matrix UzAnd UrAnd is also used to compute the weight matrix for z and r. z and r denote update and reset gates, and σ denotes a sigmoid function, which can be expressed as σ ═ 1/(1+ exp (-x)).
Figure RE-GDA00032147358200001110
Figure RE-GDA00032147358200001111
Where tanh (x) is an activate function, as an element-by-element multiplication operation. The propagation steps of the GGNN neural network are similar to GRUs, have an update function, and can combine information from other nodes and the last time step to update the hidden state of each node. When the graph level is output, the representation vector of the graph level is defined as
Figure RE-GDA00032147358200001112
Wherein,
Figure RE-GDA00032147358200001113
acting as a soft attention mechanism that decides which nodes are associated with the current graph-level task, i and j being
Figure RE-GDA00032147358200001114
And xvA neural network is cascaded as an input and outputs a real value. h isgThe transient stability of the power system can be determined.
Therefore, when the model is trained and the loss function is reduced, the parameters of the model are updated. A loss function of
Figure RE-GDA00032147358200001115
Wherein liLabel representing sample i, hiThe graph level representing sample i represents a vector, i represents sample i, and G represents the sample size.
(3) Building a second GGNN model, evaluating the transient stable instability state by using the first GGNN model, and evaluating the type of the instability state by using the second GGNN model when the transient stable instability state is unstable:
the transient stability evaluation model of the power system based on other machine learning methods can only judge the transient stability of the system, and the transient stability evaluation model based on the GGNN provided by the invention can not only judge the transient stability, but also has interpretability. In the training process, training a transient stability assessment instability state GGNN, and recording as a first GGNN model; training a GGNN for evaluating the type of the instability state, and recording as a second GGNN model; the output of the second GGNN model is of an unstable type, and the reason for causing the instability of the power system is correspondingly judged;
the propagation of the second GGNN model for determining the event type is the same as the first GGNN model, except that the second GGNN model for determining the event type is a multi-classification model. The activation and loss functions of the output layer are also different from the first GGNN model. The activation function for event classification is softmax and the loss function is the class cross entropy. The softmax function is expressed as
Figure RE-GDA0003214735820000121
Wherein j represents the jth neuron, and n is the number of neurons in the layer before the output layer.
Figure RE-GDA0003214735820000122
Wherein l1The label is expressed as event type, k represents the number of event types, and equation (20) is a loss function.
In the actual evaluation process, the first GGNN model is firstly used for evaluating the transient stability instability state, and if the output of the model is the transient stability (namely h) of the power systemg> 0.5), entering the next circulation; otherwise, the second GGNN model is used for evaluating and continuously judging the event type causing the system instability. From this output, it is known what type of event is most likely to cause transient instability in the system. FIG. 8 shows a flow chart of the model determining transient stability and predicting events that cause system instability.
Examples
IEEE-39 bus System:
the validity of the TSA model was verified with the new england 39 bus system. The simulation was performed on a PC with an Intel Core i3 CPU and 8.00GB RAM. The architecture of the new england 39 bus system is shown in fig. 9. As shown in fig. 9, the system has 10 generators, 39 buses, 19 loads, 12 transformers and 34 transmission lines.
B. Data generation
The data set used in the experiment was constructed by time domain simulation of the IEEE-39 bus system using powerfactory. By setting the fault type and the stable state of the predefined system in advance for the power system, the numerical value of each attribute of each node of the power system can be obtained on the premise of starting time domain simulation. The load of the stability analysis of the software has voltage dependence, so the initial operating condition voltage amplitude is set to be 1kpu-2kpu, the frequency is 60 Hz, and the standard voltage is 345 kV. Before the simulation is started, load flow calculation is carried out, and time domain simulation calculation can be started only when the load flow calculation result is converged.
Events considered by the present invention are short circuit events, synchronous generator events, load events, switch events and tap events. The short-circuit event is a three-phase grounding short-circuit event, the fault occurrence position is the random position of all buses and connecting lines, the fault duration time is 0.1s, and the time domain simulation duration time is 5 s. Finally, 1000 samples are obtained, wherein 580 stable samples are obtained, and 420 unstable samples are obtained. The fault durations for the synchronous generator event, the switch event and the tap event are all 0.1s, and the duration of the time domain simulation is also all 5 s. The number of the finally obtained samples is 1000, 1000 and 102 respectively. The number of stable samples of the synchronous generator event is 640, the number of unstable samples is 360, the number of stable samples in the switching event is 570, the number of unstable samples is 430, the number of stable samples in the switching event is 60, and the number of unstable samples is 42. The load variation range for the load event was 75% to 125%, with 5% change in magnitude each time, the duration of the fault was 0.1s, the duration was 5s, and the last to 1000 samples, with 650 stable samples and 350 unstable samples. Therefore, a total of 4102 samples were obtained by time domain simulation, wherein the number of stable samples was 2500 and the number of unstable samples was 1502. The sample dimension obtained was 89 +10+10+10+10+ 39. Because the power system in real life is very stable, and the power system is unstable under only a few conditions, unstable samples are very few in actual data. Therefore, in order to simulate the real power system situation, some unstable samples are randomly discarded, and the unstable samples account for 9% of the total samples.
Another definition of transient stability of a power system is the ability of each generator set to maintain synchronous operation after the power system in steady state operation is subjected to large disturbances (various short circuit faults, large-capacity generator removal, large load increase, etc.). The rotor angle δ is used as a criterion for judging the stability of the power system, so that the stability of the sample is judged by using a Transient Stability Index (TSI) of the disturbed relative rotor angle, and the expression is as follows:
Figure RE-GDA0003214735820000131
in the formula, deltamaxThe maximum relative rotor angle of any two generator sets in the simulation time length is obtained. If TSI is larger than 0, the system keeps a stable operation state; otherwise, the system is unstable.
In order to effectively evaluate the performance of the model, 4 test training sets are selected to test the training model. The proportion of the data set randomly divided into training and test sets was 4: 1,3: 1,7: 3 and 3: 2. the experimental results for the training and test sets for each scale are shown in the following sections.
C. Visualization analysis of synthetic data
Experiments show that the method provided by the invention can effectively solve the problem of high misjudgment rate caused by data imbalance in the transient stability evaluation problem of the power system, and can achieve higher accuracy.
To achieve sample balance, a labeled sample is generated, and the label (l) of the sample is labeled1And l2) If the condition y is set, the input condition y of the discriminator D is the target. And changing the overall label of the sample into the event label of the sample, thereby obtaining unstable data of event balance. Therefore, the problem of unbalanced stability of the sample can be solved, and the problem of unbalanced imbalance of each event in the unstable sample can be solved. Fig. 10 and 11 show loss iteration diagrams of the generative model G and the discriminant model D, respectively, and it is clear that convergence can be achieved and the convergence speed is high in both the generative model G and the discriminant model D.
Fig. 12 shows the result of data balancing of the proposed method, which transforms the balanced data into a two-dimensional spatial distribution. In fig. 12, the green dots represent unstable samples in the original samples, the red dots represent stable samples in the original samples, and the blue dots represent unstable samples generated by CGAN. As can be seen from the graph, the number of the original unstable samples is small, the generated unstable samples and the generated stable samples can reach a balanced state, and the generated unstable samples and the original samples do not overlap, so that the model can clearly judge the type of the samples, and the accuracy of the samples is not influenced.
These results can only show that the CGAN can generate data that conforms to the original data distribution, and can solve the problem of data imbalance. It is necessary to prove that the new balanced data set generated is indeed capable of improving the performance of the transient stability assessment model. Fig. 13 shows the accuracy of the transient stability evaluation model of the power system at each sample scale. As shown in fig. 13, when the ratio of the stable and unstable samples is large, the accuracy of the model is low, the accuracy of the model gradually increases as the sample ratio decreases, and the accuracy of the model reaches the highest when the samples are balanced.
In the invention, a Convolutional Neural Network (CNN) is selected as a comparison experiment, and compared with the result of the transient stability evaluation model of the GGNN, the accuracy and precision of the method provided by the invention are higher. To demonstrate the performance of the TSA models of these two methods, 4 indices were chosen, accuracy, precision, recall, and F1 score, respectively. Tables 1 and 2 show the results of the transient stability evaluation model performance for the two methods at different proportions of the sample. As can be seen from table 1, the minimum model accuracy is 30%, which is the proportion of the unbalanced sample in the actual situation preset in the present invention, and the maximum model accuracy is 92.09%, which is the sample data after balancing, and the accuracy of the trained model. Table 2 shows the transient stability evaluation results of the comparative experiment CNN when the model was trained at different sample ratios. The generation of unstable sample data by the CGAN can solve the problem of data imbalance and does not lose the performance of the model, namely the model provided by the invention or other models. As can be seen from the comparison of Table 1 and Table 2, the accuracy of the GGNN transient stability evaluation model provided by the invention is higher than that of the CNN transient stability evaluation model.
TABLE 1 GGNN transient stability evaluation results of samples at different ratios
Figure RE-GDA0003214735820000141
Figure RE-GDA0003214735820000151
TABLE 2 CNN transient stability evaluation results for different sample ratios
Figure RE-GDA0003214735820000152
In order to prevent overfitting of the model and effectively improve the performance of the model, the invention selects the proportion of 4 different training test sets. As shown in table 4, the results of the proposed transient stability assessment model for four different training test sets a, B, C, D are presented. Table 3 shows four different training test set ratios. In table 4, under the same training test set and under the balanced data, the performance index of each model is larger than that of the unbalanced data as a result of the GGNN transient stability evaluation model. Therefore, the CGAN can solve the problem of data imbalance.
TABLE 3 proportion of different training test sets
Figure RE-GDA0003214735820000153
TABLE 4 GGNN transient stability evaluation results of different test training sets
Figure RE-GDA0003214735820000154
As can be seen from the above-mentioned figures and tables, the CGAN algorithm proposed by the present invention to generate unstable samples to solve the data imbalance problem in the transient stability evaluation of the power system is feasible without losing the generality of the data and the performance of the model. The solution of the data imbalance problem is effective for improving the transient stability evaluation model of the power system, and the performance index of the model is higher when the sample is balanced compared with the unbalanced sample.
Classification performance analysis of GGNN model
To demonstrate the effectiveness of the GGNN model used, a deep convolutional neural network CNN was used as a comparison algorithm. All results were trained using the same training set and test set. Table 5 below shows the results of different training test set ratios and different models in the transient stability assessment of the power system. As can be seen from table 5, under the same training test set ratio, the performance of the transient stability evaluation model of the GGNN power system is always better than that of the CNN model.
Table 5. transient stability evaluation results for different training test set ratios and models.
Figure RE-GDA0003214735820000161
E. Judgment cause performance analysis
The graph neural network is illustrative, and the GGNN is one of graph learning and is also illustrative. In the invention, a reason set is constructed firstly, and the maximum node score is output by using a soft attention mechanism in a GGNN transient stability evaluation model. The node score is maximum, which indicates that the node has the largest relation with the power system instability. And comparing each characteristic of the node with the event set to obtain the probability of each event, and outputting the event with the maximum probability. Fig. 14 shows the results of the above events predicted from samples in which the determination result is unstable among the samples tested. In fig. 14, each blue bar represents the number of unstable samples divided into one of the above events, and the red bar represents the true number of unstable samples actually belonging to the event type. The vertical axis is the number of samples and the horizontal axis is the type of event. As can be seen from fig. 14, the accuracy of the overall cause of the determination was 94.82%, and the accuracy of the unstable sample labeled as a short-circuit event was the highest, 100%. The accuracy of the unstable sample with the label as a tap event was the lowest, 95.91%.
When the real-time data is judged to be unstable, the event with the highest probability is finally output. If the probabilities of two or more causes for each cause are the same, no output is made. As shown in fig. 14, the total number of predictions is less than the number of unstable samples in the real test set, which means that there is a problem of missing judgment for each classification. As shown in table 6, the results of determining the cause of the unstable samples for the GGNN transient stability evaluation model. As can be seen from the above diagrams and tables, the GGNN transient stability evaluation model can determine the cause of the power system instability after evaluating the power system instability, and the overall accuracy is 94.82%.
TABLE 5 results of the GGNN model for reasons
Figure RE-GDA0003214735820000171
In the present invention, a conditional generation countermeasure network (CGAN) is used to generate unstable samples to solve the data imbalance problem. The CGAN can generate not only unstable samples but also unstable samples with unbalanced events, so that the samples can reach not only stable and unstable equilibrium but also equilibrium of events in the unstable samples. After the problem of data imbalance of the samples is solved, the transient stability of the power system is evaluated by using a GGNN algorithm, and the reason causing instability of the power system is judged. Through experiments, the CGAN is capable of effectively improving the transient stability evaluation performance and solving the problem of data imbalance of the transient stability evaluation of the power system. The GGNN can also quickly and accurately evaluate the transient stability of the power system, judge the cause of the instability of the power system and achieve good performance.
The present invention is capable of other embodiments and its several details are capable of modifications in various obvious respects, all without departing from the spirit and scope of the present invention.

Claims (10)

1. The gate diagram neural network transient stability evaluation method based on the imbalance data is characterized by comprising the following steps of:
s1, collecting stability evaluation data of the power system, wherein the stability evaluation data comprisesBus voltage U1The operation parameters of the motor comprise active power P of the motor, reactive power Q of the motor, voltage amplitude U of the motor and current I of the motor;
s2, the stability evaluation of the power system is realized by using the transient stability evaluation model, and the process of realizing the stability evaluation of the power system by using the transient stability evaluation model comprises the following steps:
generating a graph G (V, E) based on the stability evaluation data, V represents the generator set and V belongs to V, and selecting the generator set as a node V, EvwE is an edge between the nodes v and w; the active power P of the motor, the reactive power Q of the motor, the voltage amplitude U of the motor and the current I of the motor are used for forming an annotation vector of a node, and the bus voltage U1For generating edges between nodes;
based on the graph G (V, E), firstly, evaluating the unstable state of transient stability by using a first GGNN model, and entering the next cycle if the output of the model is the transient stability of the power system; otherwise, evaluating and continuously judging the event type causing the system instability by using a second GGNN model;
the first GGNN is a two-classification model, and the output is a stable state and an unstable state; the second GGNN model is a multi-classification model, the output is an unstable type, and the reason for causing the instability of the power system is correspondingly judged;
the transient stability evaluation model is trained in advance, and the training process comprises the following steps:
collecting stability evaluation data of the power system as sample data, wherein the operation parameters of the motor in the sample data further comprise the relative rotor angle delta of each generator, and dividing the steady state of the power system into a steady state and an unstable state by using a transient stability index based on the sample data of the relative rotor angle delta of each generator;
then generating an unstable sample by using the first CGAN according to the data of the unstable state, and generating unstable samples of different events by using the second CGAN;
and generating a graph G (V, E) by using the samples in the stable state and the samples corresponding to the unstable state, and training a first GGNN model for evaluating transient stability and a second CGAN model for evaluating events corresponding to the unstable samples.
2. The imbalance data-based gated neural network transient stability assessment method of claim 1, wherein the first GGNN uses a gated recursion unit GRU in the propagating step.
3. The imbalance data-based gated neural network transient stability assessment method of claim 2, wherein the second GGNN uses a gated recursion unit GRU in the propagating step.
4. The imbalance data-based gate map neural network transient stability assessment method according to claim 1, 2 or 3, wherein a loss function in the process of training the first GGNN model
Figure FDA0003127144050000021
Wherein liLabel representing sample i, hiThe graph level representing sample i represents a vector, i represents sample i, and G represents the sample size.
5. The imbalance data-based gate map neural network transient stability evaluation method of claim 4, wherein the propagation recursion process of the first GGNN model is as follows:
Figure FDA0003127144050000022
wherein,
Figure FDA0003127144050000023
expressed as hidden state of node v at time t, xvAnnotating the vector for the node of node v; t represents the transposition of the vector;
Figure FDA0003127144050000024
the node v collects information from the neighbor nodes;
Figure FDA0003127144050000025
the vector represents information collected by the node v at time t about the neighbor node; a. thevIs a sub-matrix of the graph adjacency matrix and represents the connection state of the node v and the neighbor nodes thereof; v represents the number of nodes;
Figure FDA0003127144050000026
indicating the hidden state of node 1 at time t-1, V indicates the number of nodes,
Figure FDA0003127144050000027
representing a vector
Figure FDA0003127144050000028
B denotes calculation
Figure FDA0003127144050000029
The vector of (a);
Figure FDA00031271440500000210
Figure FDA00031271440500000211
wherein
Figure FDA00031271440500000212
Information representing the update of node v at time t,
Figure FDA00031271440500000213
information representing the node v at time t, matrix WzAnd WrIs used for calculatingWeight matrix of z and r, matrix UzAnd UrA weight matrix also used to compute z and r; z and r denote an update gate and a reset gate, and σ denotes a sigmoid function;
Figure FDA00031271440500000214
Figure FDA00031271440500000215
where tanh (x) is an activate function, an element-by-element multiplication operation;
when the graph level is output, the representation vector of the graph level is defined as
Figure FDA0003127144050000031
Wherein,
Figure FDA0003127144050000032
acting as a soft attention mechanism that decides which nodes are associated with the current graph-level task, i and j being
Figure FDA0003127144050000033
And xvCascading neural networks as inputs and outputting real values; h isgThe method is used for judging the transient stability of the power system.
6. The imbalance data-based gate map neural network transient stability assessment method of claim 5, wherein h of the first GGNN modelg> 0.5 indicates stabilization.
7. The imbalance data-based gate map neural network transient stability assessment method of claim 6, wherein the loss function in the process of training the second GGNN model
Figure FDA0003127144050000034
Wherein l1Labels expressed as event types, k denotes the number of event types, ajIs the output of the activation function softmax.
8. The imbalance data-based gate map neural network transient stability assessment method of claim 7, wherein the first CGAN and the second CGAN for generating unstable samples are pre-trained before training the first GGNN model and the second CGAN model, and the training process comprises the following steps:
constructing data x of the CGAN according to the collected stability evaluation data of the power system as sample data:
x＝[x1,x2,…,xn] (2)
Figure FDA0003127144050000035
ai＝[U1 P Q U I δ]1×89 (4)
wherein n represents the number of input samples, and the input data is n arrays of 500 × 89 according to the formulas (2), (3) and (4);
the input condition of a discriminator D of the CGAN is a label vector of sample data, and the label vector is an n multiplied by 1 array;
generating multi-label data based on sample data, wherein each piece of data has 2 labels:
L＝{l1,l2} (5)
wherein l1Representing events causing the state of data,/2Represents the steady state of the system;
l1＝{0，1，2，3，4} (6)
l1an event represented by 0 is a short-circuit event, l1The event denoted 1 is a tap event, l1The event denoted 2 is a load event, l1The event denoted by 3 is a switching event, l1The event denoted 4 is a synchronous generator event;
l2＝{0，1} (7)
when l is2At 0, the sample is destabilized, when l2When 1, the sample is stable; l is an array with 1 row and 2 columns;
and training to generate a CGAN model corresponding to the unstable sample and a CGAN model corresponding to the unstable sample generating different events based on the sample data and the corresponding label, and respectively recording the CGAN models as a first CGAN model and a second CGAN model.
9. The imbalance data-based transient stability assessment method for a gate map neural network according to claim 8, wherein in the process of assessing the stability of the power system according to the collected imbalance data, the fault clearing time is recorded as 0 time, and the bus voltage and the motor operation parameters are collected according to a time interval of 0.01 s.
10. The imbalance data-based gate map neural network transient stability assessment method of claim 9, wherein the transient stability index is a function of the magnitude of the imbalance data
Figure FDA0003127144050000041
Wherein deltamaxThe maximum relative rotor angle of any two generator sets in the simulation time length is obtained.
CN202110693635.4A 2021-06-22 2021-06-22 Gate graph neural network transient stability evaluation method based on unbalanced data Active CN113435575B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110693635.4A CN113435575B (en) 2021-06-22 2021-06-22 Gate graph neural network transient stability evaluation method based on unbalanced data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110693635.4A CN113435575B (en) 2021-06-22 2021-06-22 Gate graph neural network transient stability evaluation method based on unbalanced data

Publications (2)

Publication Number Publication Date
CN113435575A true CN113435575A (en) 2021-09-24
CN113435575B CN113435575B (en) 2022-07-29

Family

ID=77757057

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110693635.4A Active CN113435575B (en) 2021-06-22 2021-06-22 Gate graph neural network transient stability evaluation method based on unbalanced data

Country Status (1)

Country Link
CN (1) CN113435575B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114006370A (en) * 2021-10-29 2022-02-01 中国电力科学研究院有限公司 Power system transient stability analysis and evaluation method and system

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110532596A (en) * 2019-07-16 2019-12-03 中国电力科学研究院有限公司 A kind of method and system generating grid simulation sample based on LSTM algorithm
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

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110532596A (en) * 2019-07-16 2019-12-03 中国电力科学研究院有限公司 A kind of method and system generating grid simulation sample based on LSTM algorithm
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

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
卢锦玲 等: "基于改进深度残差收缩网络的电力系统", 《电子技术学报》 *
李嘉敏 等: "样本不平衡情况下的电力系统暂态稳定集成评估方法", 《电力系统自动化》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114006370A (en) * 2021-10-29 2022-02-01 中国电力科学研究院有限公司 Power system transient stability analysis and evaluation method and system

Also Published As

Publication number Publication date
CN113435575B (en) 2022-07-29

Similar Documents

Publication Publication Date Title
Tong et al. Detection and classification of transmission line transient faults based on graph convolutional neural network
Belagoune et al. Deep learning through LSTM classification and regression for transmission line fault detection, diagnosis and location in large-scale multi-machine power systems
Huang et al. Recurrent graph convolutional network-based multi-task transient stability assessment framework in power system
James et al. Intelligent time-adaptive transient stability assessment system
Liu et al. A new dynamic security assessment framework based on semi-supervised learning and data editing
CN111523778A (en) Power grid operation safety assessment method based on particle swarm algorithm and gradient lifting tree
Liu et al. Non‐parametric statistics‐based predictor enabling online transient stability assessment
MansourLakouraj et al. A multi-rate sampling PMU-based event classification in active distribution grids with spectral graph neural network
Zhou et al. Transient stability assessment based on gated graph neural network with imbalanced data in Internet of Energy
CN115238980A (en) Micro-grid transient power angle trajectory prediction method based on attention mechanism and space-time diagram convolutional neural network
Huang et al. Research on transformer fault diagnosis method based on GWO optimized hybrid kernel extreme learning machine
CN112149883A (en) Photovoltaic power prediction method based on FWA-BP neural network
Zhang et al. Towards multi-scenario power system stability analysis: An unsupervised transfer learning method combining DGAT and data augmentation
CN117349614A (en) Frequency stability prediction method based on self-attention mechanism and space-time diagram convolution network
CN113435575B (en) Gate graph neural network transient stability evaluation method based on unbalanced data
Yu et al. PIDGeuN: Graph neural network-enabled transient dynamics prediction of networked microgrids through full-field measurement
Xu et al. An improved ELM-WOA–based fault diagnosis for electric power
Gu et al. Power System Transient Stability Assessment Based on Graph Neural Network with Interpretable Attribution Analysis
Abiyev Fuzzy wavelet neural network for prediction of electricity consumption
Shi et al. Integrated advance assessment of power system transient voltage and transient angle stability based on two-stage ensemble spatio-temporal graph neural network
CN118445682A (en) Power system transient stability evaluation method based on condition generation countermeasure network and graph convolution network
Lin et al. Post-disturbance dynamic frequency features prediction based on convolutional neural network
CN116933860A (en) Transient stability evaluation model updating method and device, electronic equipment and storage medium
CN116204771A (en) Power system transient stability key feature selection method, device and product
Zhou et al. Transient stability assessment of large-scale AC/DC hybrid power grid based on separation feature and deep belief networks

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