CN110224401A - In conjunction with the Power system transient stability prediction method of manual features and residual error network - Google Patents
In conjunction with the Power system transient stability prediction method of manual features and residual error network Download PDFInfo
- Publication number
- CN110224401A CN110224401A CN201910529608.6A CN201910529608A CN110224401A CN 110224401 A CN110224401 A CN 110224401A CN 201910529608 A CN201910529608 A CN 201910529608A CN 110224401 A CN110224401 A CN 110224401A
- Authority
- CN
- China
- Prior art keywords
- manual features
- max
- layer
- transient stability
- obtains
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
Abstract
The present invention relates to a kind of Power system transient stability prediction methods of combination manual features and residual error network, belong to Power System Stability Analysis technical field.The present invention acquires each variable of generator after failure removal, constitutes initial characteristics vector；Manual features are extracted from initial characteristics according to being manually set, initial characteristics vector is arranged in three-dimensional data, feature is automatically extracted using the residual unit in depth residual error network, this is automatically extracted into feature and manual features collectively as the input of full articulamentum, Transient Stability Prediction output is obtained after two layers of full articulamentum processing, constitutes the structure of Transient Stability Prediction model；Finally, iterative solution obtains relatively excellent model parameter using training sample set and verifying sample set, to obtain final Transient Stability Prediction model, and it to be used for Transient Stability Prediction.The present invention can be improved the predictablity rate of electric power system transient stability by the combination of residual unit and the optimum selecting of model parameter in manual features and depth residual error network.
Description
Technical field
The present invention relates to a kind of Power system transient stability prediction methods of combination manual features and residual error network, belong to electricity
Force system stability analysis technical field.
Background technique
Transient stability destruction is the major reason that massive blackout accident occurs for electric system, how quick and precisely to be judged
The transient stability of system is power system security prevention and control one of the main problem to be considered.In recent years, smart grid is built
It deepens continuously, the operation data acquired in electric system becomes increasingly abundant and perfect, so that the Transient Stability Prediction side of data-driven
Extensive concern of the method by domestic and foreign scholars.
Currently, the input feature vector for Transient Stability Prediction model includes that artificial extraction feature and intelligent method automatically extract
Feature.Electric system is a highly complex nonlinear system, and artificial feature extracting method inevitably needs to make
Ideal is assumed and is simplified, it is difficult to reflect each variable in incidence relation macroscopically, there are certain limitations.Recently, depth
Practise theoretical, algorithm and in terms of rapidly develop, have some scholars and deep learning divided in electric power system transient stability
The application start in analysis field is studied.China Patent Publication No. CN107391852A proposes steady based on depth confidence network struction transient state
Determine assessment models, improves assessment calculating speed and Evaluation accuracy.China Patent Publication No. CN107846012A proposes one kind
Feature extraction is successively carried out to characteristic variable using autocoder is stacked, then constructs stable class using convolutional neural networks
Model.Application publication number is the Chinese patent of CN108832619A, is proposed a kind of temporary based on the building of depth convolutional neural networks
State Stability Assessment model, the Automatic Feature Extraction based on deep learning can carry out comprehensive analysis to history and present data,
High-order feature is formed, more accurate, objective, effectively expressing is carried out to data, reduction artificially designs not perfect, however the party
Method does not account for the three dimensional characteristic of input data dynamic trajectory, is not bound with existing artificial experience feature yet.It is artificial to extract spy
Sign is not antagonistic with the method that deep learning automatically extracts feature, does not conflict, and has only and sufficiently two alanysis methods is combined to carry out feature
It extracts, realizes and have complementary advantages, could construct and obtain the higher Transient Stability Prediction model of accuracy rate.
Summary of the invention
The purpose of the present invention is to propose to the Power system transient stability prediction sides of a kind of combination manual features and residual error network
Method the shortcomings that for prior art, is extracted artificial feature and is combined with the method that deep learning automatically extracts feature, to construct
The higher Transient Stability Prediction model of accuracy rate is obtained, and is used for Transient Stability Prediction, improves Transient Stability Prediction precision.
The Power system transient stability prediction method of combination manual features proposed by the present invention and residual error network, this method packet
Include following steps:
(1) to the electric system with N platform generator, according to electric system history run and operations staff's experience,
Time-domain-simulation calculating is carried out to transient stability of the s kind operating condition under f kind failure, obtains the feature of s × f kind Run-time scenario
Vector XkWith transient stability yk, wherein subscript k indicates kth kind Run-time scenario, k=1,2 ..., s × f, yk=(0,1) indicates
Electric system is able to maintain transient stability, y after fault clearancek=(1,0) indicate that electric system cannot be protected after fault clearance
Transient stability is held, the fault clearance time sets according to artificial experience, n sampled point after fault clearance in kth kind Run-time scenario
Generator active power PGi k, generator amature angle δi k, generator amature angular velocity omegai k, generator bus voltage magnitude
VGi kWith the voltage phase angle θ of generator busGi kConstitutive characteristic vector Xk:
Xk=[PGi k(t),δi k(t),ωi k(t),VGi k(t),θGi k(t)]
Wherein, subscript i indicates that i-th generator in electric system, i=1,2 ..., N, t indicate t-th of sampled point, t
=1,2 ..., n, n are the sampling number being manually set, and sample frequency is selected as the rated frequency of electric system；
(2) according to the feature vector, X in step (1)k=[PGi k(t),δi k(t),ωi k(t),VGi k(t),θGi k(t)] it, counts
Calculate following 30 × n artificially defined manual features:
Manual featuresWherein 0-Indicate the last one sampled value before failure occurs
Manual features
Manual features
Manual features
Manual features Y5 k(t)=Y3 k(t)-Y4 k(t)
Manual features Y6 k(t)=Y2 k(t)/Y1 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y11 k(t)=Y9 k(t)-Y10 k(t)
Manual features Y12 k(t)=Y8 k(t)/Y7 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y17 k(t)=Y15 k(t)-Y16 k(t)
Manual features Y18 k(t)=Y14 k(t)/Y13 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y23 k(t)=Y21 k(t)-Y22 k(t)
Manual features Y24 k(t)=Y20 k(t)/Y19 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y29 k(t)=Y27 k(t)-Y28 k(t)
Manual features Y30 k(t)=Y26 k(t)/Y25 k(t)
By above-mentioned 30 × n artificial characteristic Ysp k(t) minimax normalization is carried out, wherein subscript p=1 ..., 30 obtain
Manual features after normalizationNormalized formula are as follows:
(3) by the feature vector, X of every kind of scenek=[PGi k(t),δi k(t),ωi k(t),VGi k(t),θGi k(t)] it carries out most
Big minimum normalization, is arranged in three-dimensional data according to generator dimension, variable dimension and time dimensionThen convolution is set
Layer, pond layer, residual unit and full articulamentum, in conjunction with 30 × n manual features in step (2)Obtain transient stability
The structure of prediction model M, specifically includes the following steps:
Feature vector, X under each Run-time scenario that (3-1) obtains step (1)k=[PGi k(t),δi k(t),ωi k
(t),VGi k(t),θGi k(t)] minimax normalization, normalized formula are carried out are as follows:
Then, by the data after normalizationAccording to generator dimension,
Time dimension and variable dimension are arranged in three-dimensional dataThe three-dimensional dataDimension be N × n × 5；
(3-2) combines the residual error network in manual features and deep learning in step (2), and it is pre- that design obtains transient stability
The structure of model M is surveyed, the input data of M is the normalization manual features that step (2) obtainsIt is obtained with step (3-1)
Three-dimensional dataThe output of M is Ok 2, work as Ok 2When=(0,1), indicate that electric system is able to maintain under kth kind Run-time scenario
Transient stability works as Ok 2When=(1,0), indicate that electric system is not able to maintain transient stability under kth kind Run-time scenario, model M by
Following multiple element stacks form:
(3-2-1) convolutional layer
Utilize c convolution kernel wlWith a bias matrix V0To the three-dimensional data of kth kind Run-time scenario in step (3-1)
Convolution operation is carried out, feature vector O is obtainedk, wherein l=1 ..., c, convolution kernel wlWith bias matrix V0It is step (4) wait ask
Parameter, wl∈Ra×d, Ra×dIndicating that a × d ties up matrix, and each element is real number in matrix, the value of a and d are taken as odd number, and
Meet a≤N, d≤n, number c >=5 of convolution kernel；
The pond (3-2-2) layer
To feature vector OkMaximum pond is carried out, the feature A of Chi Huahou is obtainedk 0；
The m residual unit that (3-2-3) is stacked
Using the m residual unit stacked in depth residual error network, to the Chi Huahou feature A of step (3-2-3)k 0It carries out special
Sign is extracted, wherein the output of f-th of residual unit are as follows:
Ak f=σ (σ (Ak f-1*Jl f,1+Vf,1)*Jl f,2+Vf,2+Ak f-1)
Wherein, subscript f=1 ..., m, m are the sum of residual unit, and for the value of m by being manually set, σ () is ReLU activation letter
Number, Ak fIt is the output of f-th of residual unit, Ak f-1It is the output of the f-1 residual unit, Jl f,1It is that f grades of residual units make
First of convolution kernel of first layer convolutional layer, Jl f,2It is first volume of the second layer convolutional layer that f grades of residual units use
Product core, l=1 ..., c, Vf,1It is the bias matrix for the first layer convolutional layer that f grades of residual units use, Vf,2It is f grades of residual errors
The bias matrix for the second layer convolutional layer that unit uses, convolution kernel Jl f,1And Jl f,2, bias matrix Vl f,1And Vl f,2It is step
(4) parameter to be asked；
The pond (3-2-4) layer
To the output A of m grades of residual unitsk mMaximum pond is carried out, Chi Huahou feature Q is obtainedk；
(3-2-5) batch normalization layer
Using batch method for normalizing, to the Chi Huahou feature Q of step (3-2-4)kIt is normalized, is normalized
Feature U afterwardsk；
(3-2-6) tiling layer
Using tiling function, by the normalization characteristic U of step (3-2-5)kTiling is the dimensional feature vector of h × 1 Vk, wherein h
Size is by normalization characteristic UkDimension determine；
The full articulamentum of (3-2-7) first layer
30 × n normalization the manual features that step (2) is obtainedWith the dimensional feature of h × 1 of step (3-2-6) to
Measure VkMerge into Zk, ZkIt is (h+30 × n) × 1 dimensional vector, then by ZkIt is input in the full articulamentum of first layer, obtains first layer
The output of full articulamentum is Ok 1:
Ok 1=σ (GZk+b1)
Wherein, subscript 1 indicates the full articulamentum of first layer, weight matrix G ∈ Rg×h, Rg×hIndicate that g × h ties up matrix, and matrix
In each element be real number, the bias vector b of the full articulamentum of first layer1∈Rg×1, Rg×1Indicate the dimensional vector of g × 1, and vector
In each element be real number, g indicates that the output dimension of full articulamentum, the output dimension of full articulamentum are examined by being manually set
Consider and there was only two layers of full articulamentum in this patent, the input dimension of the full articulamentum of first layer is d1+ 30 × n, the second layer connect entirely
The output dimension of layer is 2 × 1 dimensions to indicate transient stability or Transient Instability, the output dimension value range of the full articulamentum of first layer
It is set as g ∈ (2, h+30 × n), weight matrix G and bias vector b1It is the parameter to be asked of step (4)；
The full articulamentum of (3-2-8) second layer
By the output O of step (3-2-7)k 1It is input in the full articulamentum of the second layer, obtains the output of the full articulamentum of the second layer
For Ok 2:
Ok 2=Softmax (HOk 1+b2)
Wherein, subscript 2 indicates the full articulamentum of the second layer, weight matrix H ∈ R2×g, R2×gIndicate that 2 × g ties up matrix, and matrix
In each element be real number, the bias vector b of the full articulamentum of the second layer2∈R2×1, R2×1Indicate 2 × 1 dimensional vectors, and vector
In each element be real number, Softmax () is Softmax activation primitive, weight matrix H and bias vector b2It is step
(4) parameter to be asked；
(4) s × f sample obtained according to step (1) and the gradient descent algorithm based on adaptive moments estimation, i.e. Adam
Algorithm iterates to calculate the parameter to be asked in M, obtains final Transient Stability Prediction model, specifically comprise the following steps:
(4-1) is randomly selected from the s × f sample that step (1) obtainsA sample is remained as training set
It is remainingA sample collects as verifying, whereinExpression is rounded downwards 0.8 × s × f；
(4-2) sets set S={ emax,Amax,Mmax, wherein AmaxIt is that Transient Stability Prediction model is obtained in iterative process
Highest prediction accuracy rate, emaxIt is to obtain highest prediction accuracy rate AmaxWhen the number of iterations, MmaxIt is emaxWhat secondary iteration obtained
Transient Stability Prediction model, note the number of iterations are r, maximum number of iterations rmax, minimum the number of iterations is rmin, wherein rmaxWith
rminValue by being manually set, and meet rmax> rmin>=10, if the initial value of the number of iterations r is 0, emaxInitial value is 0, AmaxJust
Value is 0, model MmaxIt is set as empty；
(4-3) is by the number of iterations r and maximum number of iterations rmaxIt is compared:
(4-3-1) is if r >=rmax, then the M in set SmaxAs final Power system transient stability prediction model；
(4-3-2) is if r < rmax, then r:=r+1 is enabled, step (4-4) is transferred to；
(4-4) utilizes the training set and Adam algorithm of step (4-1), and model M needs to be asked ginseng in calculating step (3)
Number, including wl、V0、Jl f,1、Jl f,2、Vl f,1、Vl f,2、G、H、b1And b2, obtain the corresponding Transient Stability Prediction model of parameter current
Mr；
(4-5) utilizes MrIt concentrates the transient stability of all samples to predict step (4-1) verifying, it is quasi- to obtain prediction
True rate, is denoted as Ar, by ArValue and AmaxIt is compared:
(4-5-1) is if Ar>Amax, then e is enabledmax=r, Amax=Ar, Mmax=Mr, update and obtain new set S, be then transferred to
Step (4-3)；
(4-5-2) is if Ar≤Amax, then by emaxValue and r and r-rminValue be compared, if meeting r-rmin≤emax
≤ r is then transferred to step (4-3), if being unsatisfactory for r-rmin≤emax≤ r, then stop iteration, takes the M in set SmaxAs final
Power system transient stability prediction model；
(5) obtain failure removal after generator active power, rotor angle, rotor velocity, voltage magnitude and voltage
Phase angle obtains Transient Stability Prediction by calculating and being input in the Power system transient stability prediction model that step (4) obtains
As a result, specifically comprising the following steps:
(5-1) calculates using offline time-domain-simulation or directly acquires the measurement number of electrical power system wide-area metrical information system
According to obtaining the generator active power P of n sampled point after fault clearanceGi(t), generator amature angle δi(t), generator turns
Sub- angular velocity omegai(t), the voltage magnitude V of generator busGi(t) and the voltage phase angle θ of generator busGi(t), it constitutes initial
Input feature vector, wherein t=1 ..., n；
After (5-2) obtains the 30 × n normalization that step (2) defines to the initial input feature calculation of step (5-1)
Manual features；
(5-3) using step (3-1) minimax normalization initial input feature is normalized after side by side
Arrange into the three-dimensional data of N × n × 5；
N × n that manual features and step (5-3) after the 30 × n normalization that (5-4) obtains step (5-2) obtain
× 5 three-dimensional data is input to jointly in the Power system transient stability prediction model that step (4) obtains, and obtains electric system
Transient Stability Prediction result.
The Power system transient stability prediction method of combination manual features proposed by the present invention and residual error network, feature and
Advantage is:
The method of the present invention acquires after failure removal the active power, generator amature angle of generator in a period of time, turns
Sub- angular speed, the voltage magnitude of generator bus, generator bus voltage phase angle, constitute initial input feature；According to artificial
Manual features are extracted in setting from initial input feature, and by initial characteristics vector according to generator dimension, variable dimension and when
Between dimension be arranged in three-dimensional data, utilize convolutional layer in depth residual error network, pond layer, residual unit, pond layer, batch normalizing
It obtains automatically extracting feature after changing the processing of layer peace laying, is incorporated as connecting entirely by manually extracting feature and automatically extracting feature
The input of layer obtains Transient Stability Prediction output after two layers of full articulamentum processing, constitutes in conjunction with manual features and residual error
The Transient Stability Prediction model structure of network；Using training sample set and verifying collection, iterative solution obtains relatively excellent model ginseng
Number, to obtain final Transient Stability Prediction model；Finally, acquiring initial input feature and being input to Transient Stability Prediction mould
In type, Transient Stability Prediction result is obtained.The method of the present invention passes through combination and the model of manual features and depth residual error network
The optimum selecting of parameter improves the accuracy rate of Transient Stability Prediction.
Detailed description of the invention
Fig. 1 is the flow diagram of the method for the present invention.
Fig. 2 is Transient Stability Prediction model structure schematic diagram involved in the method for the present invention.
Fig. 3 is three-dimensional input data schematic diagram involved in the method for the present invention.
The flow chart of building Transient Stability Prediction model in the step of Fig. 4 is the method for the present invention (4).
Specific embodiment
The Power system transient stability prediction method of combination manual features proposed by the present invention and residual error network, flow chart element
Figure as shown in Figure 1, method includes the following steps:
(1) to the electric system with N platform generator, according to electric system history run and operations staff's experience,
Time-domain-simulation calculating is carried out to transient stability of the s kind operating condition under f kind failure, obtains the feature of s × f kind Run-time scenario
Vector XkWith transient stability yk, wherein subscript k indicates kth kind Run-time scenario, k=1,2 ..., s × f, yk=(0,1) indicates
Electric system is able to maintain transient stability, y after fault clearancek=(1,0) indicate that electric system cannot be protected after fault clearance
Transient stability is held, the fault clearance time sets according to artificial experience, n sampled point after fault clearance in kth kind Run-time scenario
Generator active power PGi k, generator amature angle δi k, generator amature angular velocity omegai k, generator bus voltage magnitude
VGi kWith the voltage phase angle θ of generator busGi kConstitutive characteristic vector Xk:
Xk=[PGi k(t),δi k(t),ωi k(t),VGi k(t),θGi k(t)]
Wherein, subscript i indicates that i-th generator in electric system, i=1,2 ..., N, t indicate t-th of sampled point, t
=1,2 ..., n, n are the sampling number being manually set, and sample frequency is selected as the rated frequency of electric system, is to rated frequency
The electric system of 50Hz, sample frequency 50Hz are the electric system of 60Hz to rated frequency if n=5, and sample frequency is
60Hz, if n=6；
(2) according to the feature vector, X in step (1)k=[PGi k(t),δi k(t),ωi k(t),VGi k(t),θGi k(t)] it, counts
Calculate following 30 × n artificially defined manual features:
Manual featuresWherein 0-Indicate the last one sampled value before failure occurs
Manual features
Manual features
Manual features
Manual features Y5 k(t)=Y3 k(t)-Y4 k(t)
Manual features Y6 k(t)=Y2 k(t)/Y1 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y11 k(t)=Y9 k(t)-Y10 k(t)
Manual features Y12 k(t)=Y8 k(t)/Y7 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y17 k(t)=Y15 k(t)-Y16 k(t)
Manual features Y18 k(t)=Y14 k(t)/Y13 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y23 k(t)=Y21 k(t)-Y22 k(t)
Manual features Y24 k(t)=Y20 k(t)/Y19 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y29 k(t)=Y27 k(t)-Y28 k(t)
Manual features Y30 k(t)=Y26 k(t)/Y25 k(t)
By above-mentioned 30 × n artificial characteristic Ysp k(t) minimax normalization is carried out, wherein subscript p=1 ..., 30 obtain
Manual features after normalizationNormalized formula are as follows:
(3) by the feature vector, X of every kind of scenek=[PGi k(t),δi k(t),ωi k(t),VGi k(t),θGi k(t)] it carries out most
Big minimum normalization, is arranged in three-dimensional data according to generator dimension, variable dimension and time dimensionThen convolution is set
Layer, pond layer, residual unit and full articulamentum, in conjunction with 30 × n manual features in step (2)Obtain transient stability
The structure of prediction model M, as shown in Fig. 2, specifically includes the following steps:
Feature vector, X under each Run-time scenario that (3-1) obtains step (1)k=[PGi k(t),δi k(t),ωi k
(t),VGi k(t),θGi k(t)] minimax normalization, normalized formula are carried out are as follows:
Then, by the data after normalizationAccording to generator dimension,
Time dimension and variable dimension are arranged in three-dimensional dataThe three-dimensional dataDimension be N × n × 5, as shown in Figure 3；
(3-2) combines the residual error network in manual features and deep learning in step (2), and it is pre- that design obtains transient stability
The structure of model M is surveyed, the input data of M is the normalization manual features that step (2) obtainsIt is obtained with step (3-1)
Three-dimensional dataThe output of M is Ok 2, work as Ok 2When=(0,1), indicate that electric system is able to maintain under kth kind Run-time scenario
Transient stability works as Ok 2When=(1,0), indicate that electric system is not able to maintain transient stability under kth kind Run-time scenario, model M by
Following multiple element stacks form:
(3-2-1) convolutional layer
Utilize c convolution kernel wlWith a bias matrix V0To the three-dimensional data of kth kind Run-time scenario in step (3-1)
Convolution operation is carried out, feature vector O is obtainedk, wherein l=1 ..., c, convolution kernel wlWith bias matrix V0It is step (4) wait ask
Parameter, wl∈Ra×d, Ra×dIndicating that a × d ties up matrix, and each element is real number in matrix, the value of a and d are taken as odd number, and
Meet a≤N, d≤n, a=3, d=3, c=32 are selected as in number c >=5 of convolution kernel in one embodiment of the invention；
The pond (3-2-2) layer
To feature vector OkMaximum pond is carried out, the feature A of Chi Huahou is obtainedk 0, in one embodiment of the invention, pond
The step-length of change is selected as 2 × 2, and the size of pond filter is selected as 2 × 2；
The m residual unit that (3-2-3) is stacked
Using the m residual unit stacked in depth residual error network, to the Chi Huahou feature A of step (3-2-3)k 0It carries out special
Sign is extracted, wherein the output of f-th of residual unit are as follows:
Ak f=σ (σ (Ak f-1*Jl f,1+Vf,1)*Jl f,2+Vf,2+Ak f-1)
Wherein, subscript f=1 ..., m, m are the sum of residual unit, and for the value of m by being manually set, σ () is ReLU activation letter
Number, Ak fIt is the output of f-th of residual unit, Ak f-1It is the output of the f-1 residual unit, Jl f,1It is that f grades of residual units make
First of convolution kernel of first layer convolutional layer, Jl f,2It is first volume of the second layer convolutional layer that f grades of residual units use
Product core, l=1 ..., c, Vf,1It is the bias matrix for the first layer convolutional layer that f grades of residual units use, Vf,2It is f grades of residual errors
The bias matrix for the second layer convolutional layer that unit uses, convolution kernel Jl f,1And Jl f,2, bias matrix Vl f,1And Vl f,2It is step
(4) parameter to be asked；
The pond (3-2-4) layer
To the output A of m grades of residual unitsk mMaximum pond is carried out, Chi Huahou feature Q is obtainedk, in a reality of the invention
It applies in example, the step-length in pond is selected as 2 × 2, and the size of pond filter is selected as 2 × 2；
(3-2-5) batch normalization layer
Using batch method for normalizing, to the Chi Huahou feature Q of step (3-2-4)kIt is normalized, is normalized
Feature U afterwardsk；
(3-2-6) tiling layer
Using tiling function, by the normalization characteristic U of step (3-2-5)kTiling is the dimensional feature vector of h × 1 Vk, wherein h
Size is by normalization characteristic UkDimension determine；
The full articulamentum of (3-2-7) first layer
30 × n normalization the manual features that step (2) is obtainedWith the dimensional feature of h × 1 of step (3-2-6) to
Measure VkMerge into Zk, ZkIt is (h+30 × n) × 1 dimensional vector, then by ZkIt is input in the full articulamentum of first layer, obtains first layer
The output of full articulamentum is Ok 1:
Ok 1=σ (GZk+b1)
Wherein, subscript 1 indicates the full articulamentum of first layer, weight matrix G ∈ Rg×h, Rg×hIndicate that g × h ties up matrix, and matrix
In each element be real number, the bias vector b of the full articulamentum of first layer1∈Rg×1, Rg×1Indicate the dimensional vector of g × 1, and vector
In each element be real number, g indicates that the output dimension of full articulamentum, the output dimension of full articulamentum are examined by being manually set
Consider and there was only two layers of full articulamentum in this patent, the input dimension of the full articulamentum of first layer is d1+ 30 × n, the second layer connect entirely
The output dimension of layer is 2 × 1 dimensions to indicate transient stability or Transient Instability, the output dimension value range of the full articulamentum of first layer
It is set as g ∈ (2, h+30 × n), weight matrix G and bias vector b1It is the parameter to be asked of step (4), at of the invention one
In embodiment, if g=50；
The full articulamentum of (3-2-8) second layer
By the output O of step (3-2-7)k 1It is input in the full articulamentum of the second layer, obtains the output of the full articulamentum of the second layer
For Ok 2:
Ok 2=Softmax (HOk 1+b2)
Wherein, subscript 2 indicates the full articulamentum of the second layer, weight matrix H ∈ R2×g, R2×gIndicate that 2 × g ties up matrix, and matrix
In each element be real number, the bias vector b of the full articulamentum of the second layer2∈R2×1, R2×1Indicate 2 × 1 dimensional vectors, and vector
In each element be real number, Softmax () is Softmax activation primitive, weight matrix H and bias vector b2It is step
(4) parameter to be asked；
(4) s × f sample obtained according to step (1) and the gradient descent algorithm based on adaptive moments estimation, i.e. Adam
Algorithm iterates to calculate the parameter to be asked in M, obtains final Transient Stability Prediction model, and flow chart is as shown in figure 4, specific
Include the following steps:
(4-1) is randomly selected from the s × f sample that step (1) obtainsA sample is remained as training set
It is remainingA sample collects as verifying, whereinExpression is rounded downwards 0.8 × s × f；
(4-2) sets set S={ emax,Amax,Mmax, wherein AmaxIt is that Transient Stability Prediction model is obtained in iterative process
Highest prediction accuracy rate, emaxIt is to obtain highest prediction accuracy rate AmaxWhen the number of iterations, MmaxIt is emaxWhat secondary iteration obtained
Transient Stability Prediction model, note the number of iterations are r, maximum number of iterations rmax, minimum the number of iterations is rmin, wherein rmaxWith
rminValue by being manually set, and meet rmax> rmin>=10, if the initial value of the number of iterations r is 0, emaxInitial value is 0, AmaxJust
Value is 0, model MmaxIt is set as empty, in one embodiment of the invention, rmax1000 times are set as, rminIt is 50 times；
(4-3) is by the number of iterations r and maximum number of iterations rmaxIt is compared:
(4-3-1) is if r >=rmax, then the M in set SmaxAs final Power system transient stability prediction model；
(4-3-2) is if r < rmax, then r:=r+1 is enabled, step (4-4) is transferred to；
(4-4) utilizes the training set and Adam algorithm of step (4-1), and model M needs to be asked ginseng in calculating step (3)
Number, including wl、V0、Jl f,1、Jl f,2、Vl f,1、Vl f,2、G、H、b1And b2, obtain the corresponding Transient Stability Prediction model of parameter current
Mr；
(4-5) utilizes MrIt concentrates the transient stability of all samples to predict step (4-1) verifying, it is quasi- to obtain prediction
True rate, is denoted as Ar, by ArValue and AmaxIt is compared:
(4-5-1) is if Ar>Amax, then e is enabledmax=r, Amax=Ar, Mmax=Mr, update and obtain new set S, be then transferred to
Step (4-3)；
(4-5-2) is if Ar≤Amax, then by emaxValue and r and r-rminValue be compared, if meeting r-rmin≤emax
≤ r is then transferred to step (4-3), if being unsatisfactory for r-rmin≤emax≤ r, then stop iteration, takes the M in set SmaxAs final
Power system transient stability prediction model；
(5) obtain failure removal after generator active power, rotor angle, rotor velocity, voltage magnitude and voltage
Phase angle calculates and is input in the Power system transient stability prediction model that step (4) obtains, obtains Transient Stability Prediction knot
Fruit specifically comprises the following steps:
(5-1) calculates using offline time-domain-simulation or directly acquires the measurement number of electrical power system wide-area metrical information system
According to obtaining the generator active power P of n sampled point after fault clearanceGi(t), generator amature angle δi(t), generator turns
Sub- angular velocity omegai(t), the voltage magnitude V of generator busGi(t) and the voltage phase angle θ of generator busGi(t), it constitutes initial
Input feature vector, wherein t=1 ..., n；
After (5-2) obtains the 30 × n normalization that step (2) defines to the initial input feature calculation of step (5-1)
Manual features；
(5-3) using step (3-1) minimax normalization initial input feature is normalized after side by side
Arrange into the three-dimensional data of N × n × 5；
N × n that manual features and step (5-3) after the 30 × n normalization that (5-4) obtains step (5-2) obtain
× 5 three-dimensional data is input in the Power system transient stability prediction model that step (4) obtains, and obtains the transient state of electric system
Stability forecast result.
Claims (1)
1. a kind of Power system transient stability prediction method of combination manual features and residual error network, which is characterized in that this method
The following steps are included:
(1) to the electric system with N platform generator, according to electric system history run and operations staff's experience, to s kind
Transient stability of the operating condition under f kind failure carries out time-domain-simulation calculating, obtains the feature vector of s × f kind Run-time scenario
XkWith transient stability yk, wherein subscript k indicates kth kind Run-time scenario, k=1,2 ..., s × f, yk=(0,1) indicates electric power
System is able to maintain transient stability, y after fault clearancek=(1,0) indicate that electric system is not able to maintain temporarily after fault clearance
State is stablized, and the fault clearance time sets according to artificial experience, in kth kind Run-time scenario after fault clearance n sampled point power generation
Machine active-power PGi k, generator amature angle δi k, generator amature angular velocity omegai k, generator bus voltage magnitude VGi kWith
The voltage phase angle θ of generator busGi kConstitutive characteristic vector Xk:
Xk=[PGi k(t),δi k(t),ωi k(t),VGi k(t),θGi k(t)]
Wherein, i-th generator in subscript i expression electric system, i=1,2 ..., N, t t-th of sampled point of expression, t=1,
2 ..., n, n are the sampling number being manually set, and sample frequency is selected as the rated frequency of electric system；
(2) according to the feature vector, X in step (1)k=[PGi k(t),δi k(t),ωi k(t),VGi k(t),θGi k(t)] it, calculates such as
Lower 30 × n artificially defined manual features:
Manual featuresWherein 0-Indicate the last one sampled value before failure occurs
Manual features
Manual features
Manual features
Manual features Y5 k(t)=Y3 k(t)-Y4 k(t)
Manual features Y6 k(t)=Y2 k(t)/Y1 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y11 k(t)=Y9 k(t)-Y10 k(t)
Manual features Y12 k(t)=Y8 k(t)/Y7 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y17 k(t)=Y15 k(t)-Y16 k(t)
Manual features Y18 k(t)=Y14 k(t)/Y13 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y23 k(t)=Y21 k(t)-Y22 k(t)
Manual features Y24 k(t)=Y20 k(t)/Y19 k(t)
Manual features
Manual features
Manual features
Manual features
Manual features Y29 k(t)=Y27 k(t)-Y28 k(t)
Manual features Y30 k(t)=Y26 k(t)/Y25 k(t)
By above-mentioned 30 × n artificial characteristic Ysp k(t) minimax normalization is carried out, wherein subscript p=1 ..., 30 obtain normalizing
Manual features after changeNormalized formula are as follows:
(3) by the feature vector, X of every kind of scenek=[PGi k(t),δi k(t),ωi k(t),VGi k(t),θGi k(t)] maximum is carried out most
Small normalization is arranged in three-dimensional data according to generator dimension, variable dimension and time dimensionThen convolutional layer, pond are set
Change layer, residual unit and full articulamentum, in conjunction with 30 × n manual features in step (2)Obtain Transient Stability Prediction
The structure of model M, specifically includes the following steps:
Feature vector, X under each Run-time scenario that (3-1) obtains step (1)k=[PGi k(t),δi k(t),ωi k(t),
VGi k(t),θGi k(t)] minimax normalization, normalized formula are carried out are as follows:
Then, by the data after normalizationAccording to generator dimension, time
Dimension and variable dimension are arranged in three-dimensional dataThe three-dimensional dataDimension be N × n × 5；
(3-2) combines the residual error network in manual features and deep learning in step (2), and design obtains Transient Stability Prediction mould
The structure of type M, the input data of M are the normalization manual features that step (2) obtainsThe three-dimensional obtained with step (3-1)
DataThe output of M is Ok 2, work as Ok 2When=(0,1), indicate that electric system is able to maintain transient state under kth kind Run-time scenario
Stablize, works as Ok 2When=(1,0), indicate that electric system is not able to maintain transient stability under kth kind Run-time scenario, model M is by as follows
Multiple element stacks form:
(3-2-1) convolutional layer:
Utilize c convolution kernel wlWith a bias matrix V0To the three-dimensional data of kth kind Run-time scenario in step (3-1)It carries out
Convolution operation obtains feature vector Ok, wherein l=1 ..., c, convolution kernel wlWith bias matrix V0It is step (4) wait seek ginseng
Number, wl∈Ra×d, Ra×dIndicate that a × d ties up matrix, and each element is real number in matrix, the value of a and d are taken as odd number, and full
Sufficient a≤N, d≤n, number c >=5 of convolution kernel；
The pond (3-2-2) layer:
To feature vector OkMaximum pond is carried out, the feature A of Chi Huahou is obtainedk 0；
The m residual unit that (3-2-3) is stacked:
Using the m residual unit stacked in depth residual error network, to the Chi Huahou feature A of step (3-2-3)k 0Feature is carried out to mention
It takes, wherein the output of f-th of residual unit are as follows:
Ak f=σ (σ (Ak f-1*Jl f,1+Vf,1)*Jl f,2+Vf,2+Ak f-1)
Wherein, subscript f=1 ..., m, m are the sum of residual unit, and for the value of m by being manually set, σ () is ReLU activation primitive, Ak f
It is the output of f-th of residual unit, Ak f-1It is the output of the f-1 residual unit, Jl f,1Be f grades of residual units use
First of convolution kernel of one layer of convolutional layer, Jl f,2It is first of convolution kernel of the second layer convolutional layer that f grades of residual units use, l
=1 ..., c, Vf,1It is the bias matrix for the first layer convolutional layer that f grades of residual units use, Vf,2It is that f grades of residual units make
The bias matrix of second layer convolutional layer, convolution kernel Jl f,1And Jl f,2, bias matrix Vl f,1And Vl f,2It is step (4) wait ask
Parameter；
The pond (3-2-4) layer:
To the output A of m grades of residual unitsk mMaximum pond is carried out, Chi Huahou feature Q is obtainedk；
(3-2-5) batch normalization layer:
Using batch method for normalizing, to the Chi Huahou feature Q of step (3-2-4)kIt is normalized, after being normalized
Feature Uk；
(3-2-6) tiling layer:
Using tiling function, by the normalization characteristic U of step (3-2-5)kTiling is the dimensional feature vector of h × 1 Vk, the wherein size of h
By normalization characteristic UkDimension determine；
The full articulamentum of (3-2-7) first layer:
30 × n normalization the manual features that step (2) is obtainedWith the dimensional feature vector of h × 1 V of step (3-2-6)kIt closes
It and is Zk, ZkIt is (h+30 × n) × 1 dimensional vector, then by ZkIt is input in the full articulamentum of first layer, obtains first layer and connect entirely
The output of layer is Ok 1:
Ok 1=σ (GZk+b1)
Wherein, subscript 1 indicates the full articulamentum of first layer, weight matrix G ∈ Rg×h, Rg×hIndicate that g × h ties up matrix, and every in matrix
One element is all real number, the bias vector b of the full articulamentum of first layer1∈Rg×1, Rg×1Indicate the dimensional vector of g × 1, and every in vector
One element is all real number, and g indicates the output dimension of full articulamentum, the output dimension of the full articulamentum of first layer by being manually set,
Value range is set as g ∈ (2, h+30 × n), weight matrix G and bias vector b1It is the parameter to be asked of step (4)；
The full articulamentum of (3-2-8) second layer:
By the output O of step (3-2-7)k 1It is input in the full articulamentum of the second layer, the output for obtaining the full articulamentum of the second layer is Ok 2:
Ok 2=Softmax (HOk 1+b2)
Wherein, subscript 2 indicates the full articulamentum of the second layer, weight matrix H ∈ R2×g, R2×gIndicate that 2 × g ties up matrix, and every in matrix
One element is all real number, the bias vector b of the full articulamentum of the second layer2∈R2×1, R2×1Indicate 2 × 1 dimensional vectors, and every in vector
One element is all real number, and Softmax () is Softmax activation primitive, weight matrix H and bias vector b2It is step (4)
Parameter to be asked；
(4) s × f sample obtained according to step (1) and the gradient descent algorithm based on adaptive moments estimation, i.e. Adam are calculated
Method iterates to calculate the parameter to be asked in M, obtains final Transient Stability Prediction model, specifically comprise the following steps:
(4-1) is randomly selected from the s × f sample that step (1) obtainsA sample is remaining as training setA sample collects as verifying, whereinExpression is rounded downwards 0.8 × s × f；
(4-2) sets a set S={ emax,Amax,Mmax, wherein AmaxIt is that Transient Stability Prediction model is obtained in iterative process
Highest prediction accuracy rate, emaxIt is to obtain highest prediction accuracy rate AmaxWhen the number of iterations, MmaxIt is emaxSecondary iteration obtains
Transient Stability Prediction model, note the number of iterations be r, maximum number of iterations rmax, minimum the number of iterations is rmin, wherein rmax
And rminValue by being manually set, and meet rmax> rmin>=10, if the initial value of the number of iterations r is 0, emaxInitial value is 0, Amax's
Initial value is -1, model MmaxIt is set as empty；
(4-3) is by the number of iterations r and maximum number of iterations rmaxIt is compared:
(4-3-1) is if r >=rmax, then the M in set SmaxAs final Power system transient stability prediction model；
(4-3-2) is if r < rmax, then r:=r+1 is enabled, step (4-4) is transferred to；
(4-4) utilizes the training set and Adam algorithm of step (4-1), and model M needs to be asked parameter, packet in calculating step (3)
Include wl、V0、Jl f,1、Jl f,2、Vl f,1、Vl f,2、G、H、b1And b2, obtain the corresponding Transient Stability Prediction mould of the r times iteration parameter current
Type Mr；
(4-5) utilizes MrIt concentrates the transient stability of all samples to predict step (4-1) verifying, obtains predictablity rate,
It is denoted as Ar, by ArValue and AmaxIt is compared:
(4-5-1) is if Ar>Amax, then e is enabledmax=r, Amax=Ar, Mmax=Mr, update and obtain new set S, be then transferred to step
(4-3)；
(4-5-2) is if Ar≤Amax, then by emaxValue and r and r-rminValue be compared, if meeting r-rmin≤emax≤ r,
It is then transferred to step (4-3), if being unsatisfactory for r-rmin≤emax≤ r, then stop iteration, takes the M in set SmaxAs final electricity
Force system Transient Stability Prediction model；
(5) obtain failure removal after generator active power, rotor angle, rotor velocity, voltage magnitude and voltage phase angle,
By calculating and be input in the Power system transient stability prediction model that step (4) obtains, obtain Transient Stability Prediction as a result,
Specifically comprise the following steps:
(5-1) calculates using offline time-domain-simulation or directly acquires the metric data of electrical power system wide-area metrical information system, obtains
The generator active power P of n sampled point after to fault clearanceGi(t), generator amature angle δi(t), generator amature angle speed
Spend ωi(t), the voltage magnitude V of generator busGi(t) and the voltage phase angle θ of generator busGi(t), it is special to constitute initial input
It levies, wherein t=1 ..., n；
(5-2) obtains the initial input feature calculation of step (5-1) artificial after the 30 × n normalization that step (2) defines
Feature；
After (5-3) is normalized initial input feature using the minimax normalization of step (3-1) and it is arranged in N
The three-dimensional data of × n × 5；
N × n × 5 that manual features and step (5-3) after the 30 × n normalization that (5-4) obtains step (5-2) obtain
Three-dimensional data is input to jointly in the Power system transient stability prediction model that step (4) obtains, and obtains the transient state of electric system
Stability forecast result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910529608.6A CN110224401B (en) | 2019-06-19 | 2019-06-19 | Power system transient stability prediction method combining artificial features and residual error network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910529608.6A CN110224401B (en) | 2019-06-19 | 2019-06-19 | Power system transient stability prediction method combining artificial features and residual error network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110224401A true CN110224401A (en) | 2019-09-10 |
| CN110224401B CN110224401B (en) | 2020-09-01 |
Family
ID=67817691
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910529608.6A Active CN110224401B (en) | 2019-06-19 | 2019-06-19 | Power system transient stability prediction method combining artificial features and residual error network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110224401B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111193260A (en) * | 2020-01-16 | 2020-05-22 | 清华大学 | Power system transient stability automatic evaluation method of self-adaptive expansion data |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103400046A (en) * | 2013-08-19 | 2013-11-20 | 武汉大学 | Data modeling method suitable for power grid WAMS (wide area measurement system) and application |
| CN106229976A (en) * | 2016-08-31 | 2016-12-14 | 山东大学 | Transient rotor angle stability situation predictor method based on data-driven |
| CN109116203A (en) * | 2018-10-31 | 2019-01-01 | 红相股份有限公司 | Power equipment partial discharges fault diagnostic method based on convolutional neural networks |
| CN109840350A (en) * | 2018-12-21 | 2019-06-04 | 中国电力科学研究院有限公司 | A kind of Power System Dynamic Simulation method and system |
-
2019
- 2019-06-19 CN CN201910529608.6A patent/CN110224401B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103400046A (en) * | 2013-08-19 | 2013-11-20 | 武汉大学 | Data modeling method suitable for power grid WAMS (wide area measurement system) and application |
| CN106229976A (en) * | 2016-08-31 | 2016-12-14 | 山东大学 | Transient rotor angle stability situation predictor method based on data-driven |
| CN109116203A (en) * | 2018-10-31 | 2019-01-01 | 红相股份有限公司 | Power equipment partial discharges fault diagnostic method based on convolutional neural networks |
| CN109840350A (en) * | 2018-12-21 | 2019-06-04 | 中国电力科学研究院有限公司 | A kind of Power System Dynamic Simulation method and system |
Non-Patent Citations (1)
| Title |
|---|
| 周艳真等: ""基于两阶段支持向量机的电力系统暂态稳定预测及预防控制"", 《中国电机工程学报》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111193260A (en) * | 2020-01-16 | 2020-05-22 | 清华大学 | Power system transient stability automatic evaluation method of self-adaptive expansion data |
| CN111193260B (en) * | 2020-01-16 | 2021-01-05 | 清华大学 | Power system transient stability automatic evaluation method of self-adaptive expansion data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110224401B (en) | 2020-09-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Amraee et al. | Transient instability prediction using decision tree technique | |
| CN101882167B (en) | Wind power station equivalent modeling method of large-scale wind power concentration access power grid | |
| CN110223195A (en) | Distribution network failure detection method based on convolutional neural networks | |
| CN106383947B (en) | The fast acquiring method of wind power plant current collection network dynamic equivalent parameters | |
| CN104199302B (en) | Molding system and method of pump storage group speed regulating system | |
| CN109635928A (en) | A kind of voltage sag reason recognition methods based on deep learning Model Fusion | |
| CN102593865B (en) | Dynamic simulation system and simulation method for accessing wind power into power grids | |
| CN103700036B (en) | A kind of transient stability projecting integral method being suitable to power system Multiple Time Scales | |
| CN104951834A (en) | LSSVM (least squares support vector machine) wind speed forecasting method based on integration of GA (genetic algorithm) and PSO (particle swarm optimization) | |
| CN110059348A (en) | A kind of axial phase magnetically levitated flywheel motor suspending power numerical modeling method | |
| CN103559347A (en) | Method for establishing electromagnetic transient simulation model of large-scale AC-DC (Alternating Current - Direct Current) power system | |
| CN109766586A (en) | A kind of method and system automatically generating large scale electric network electromagnetic transient simulation model | |
| CN105182245A (en) | High-capacity battery system charge state estimation method based on unscented Kalman filter | |
| CN108008641A (en) | Generator-transformer protection device performance detecting system and method | |
| CN106778041B (en) | A kind of simplified calculation method of double feedback electric engine three short circuit current maximum value | |
| CN104716646B (en) | A kind of node Coupling Degrees method based on Injection Current | |
| CN108667069A (en) | A kind of short-term wind power forecast method returned based on Partial Least Squares | |
| CN103678798B (en) | It is a kind of for the electro-magnetic transient real-time emulation method containing distributed power distribution network | |
| CN105989206B (en) | Wind power plant and photovoltaic plant model verification method based on fast reaction generator | |
| CN110224401A (en) | In conjunction with the Power system transient stability prediction method of manual features and residual error network | |
| CN111680823A (en) | Wind direction information prediction method and system | |
| CN102510072A (en) | Power grid system transient destabilization differentiation method | |
| CN108054768B (en) | Power system transient stability evaluation method based on principal component analysis | |
| CN104701839A (en) | Air conditioner load modeling method based on least squares parameter identification | |
| CN106570229A (en) | Three-dimensional power frequency electric field analysis method and system of AC filter filed of converter station |
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 |