CN108551167A - A kind of electric power system transient stability method of discrimination based on XGBoost algorithms - Google Patents
A kind of electric power system transient stability method of discrimination based on XGBoost algorithms Download PDFInfo
- Publication number
- CN108551167A CN108551167A CN201810381824.6A CN201810381824A CN108551167A CN 108551167 A CN108551167 A CN 108551167A CN 201810381824 A CN201810381824 A CN 201810381824A CN 108551167 A CN108551167 A CN 108551167A
- Authority
- CN
- China
- Prior art keywords
- tree
- model
- xgboost
- sample
- loss function
- 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
-
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The present invention proposes a kind of electric power system transient stability method of discrimination based on XGBoost algorithms.The present invention simulates the transient process after the various method of operation lower nodes of power grid to be assessed, line fault using grid simulation software first.From the emulation electrical measure feature of extracting data, temporarily steady label is determined using temporary steady criterion.Then sample data is used to train XGBoost models.For the different feature of Type Ⅰ Ⅱ error severity in Transient Stability Prediction, introduces and notice that force coefficient is modified the loss function of algorithm.Using logistic functions by model output probability.The present invention has higher accuracy rate and recall rate, while the difference more between determining prediction and relatively uncertain prediction can be captured with probabilistic manner, so as to avoid a part for model from accidentally exporting.
Description
Technical field
The invention belongs to field of power, specifically a kind of electric power system transient stability method of discrimination.
Background technology
Due to Power System Interconnection level improve, load increasingly increase, new energy access, line transmission ability limit etc. factors,
For Operation of Electric Systems more close to its stability limit, the stable operation of power grid shows the importance of bigger, to transient stability
Evaluation problem (Transient Stability Assessment, TSA) is more exposed to the concern of people.It is traditional based on time domain
The TSA methods of emulation are limited by calculating speed, it is difficult to meet the needs of application on site.In recent years, with artificial intelligence skill
The fast development of art, the Transient Stability Evaluation based on machine learning algorithm become the research hotspot of scholar.
Machine learning algorithm is the modeling method of a kind of data-driven, according to the difference of data source used, is based on engineering
The Transient Stability Evaluation research of learning method can be divided into two major classes.The first kind use simultaneously before failure and after failure feature as model
Data input, the feature before failure is used only as data input in the second class.Feature is that system is in stable state fortune before failure
The feature that can be detected when row, such as Line Flow, node voltage；Feature after failure, such as generator amature acceleration are dynamic
State feature can be just detected only after system really breaks down.The transient process of power grid is quickly grown, once system
It breaks down, the reaction time for leaving dispatcher for is seldom, so, for power grid actual motion, directive significance bigger
It is the modeling method using secondary sources source, various information when being run according to Power System Steady-state differentiate that all kinds of failures may be made
At consequence play the role of trouble-saving so as to adjust the method for operation in time.
In the research in this field, still having problem to be solved mainly has following two aspect.On the one hand it is model
Assessment accuracy still has room for promotion.In recent years, some new machine learning algorithms have in accuracy beyond conventional machines
The performance of learning algorithm carries out Transient Stability Evaluation using such new algorithm and is expected to further promote accuracy.On the other hand it is
Machine learning algorithm is difficult to reach absolutely accuracy rate, often unavoidably has some mistakes, for operation of power networks,
The model output of mistake may bring operations staff misguidance, to cause serious misoperation fault.Existing literature
Analysis surrounds algorithm accuracy mostly, for the unspecial statistical research of the sample classified by mistake, however, for non-mechanism
Property machine learning algorithm, how effectively to avoid model possibility slip up be also a major issue.
Invention content
The technical problem to be solved by the present invention is to overcome defect of the existing technology, including modeling accuracy to need into one
Step is promoted, the research of misprediction proactive problem is insufficient, proposes that a kind of power system transient stability based on XGBoost algorithms is sentenced
Other method.
The technical solution adopted for solving the technical problem of the present invention is：
First, power grid to be assessed is simulated under the various methods of operation using power system simulation software, at each node, circuit
Break down caused consequence.The a large amount of and related initial data of the Power Network Transient Stability is consequently formed.
Secondly, feature is extracted from initial data, and determines temporarily steady label.Whether sent out using generator's power and angle difference after failure
Whether scattered decision-making system occurs unstability situation.For various steady preview roadways, all kinds of electrical measure features are extracted, as follow-up
The feature of XGBoost algorithms inputs.A certain number of samples for establishing transient stability evaluation in power system model are consequently formed
Notebook data.
Then, using XGBoost algorithms and applicability improvement is carried out, the sample data of acquisition is utilized to carry out model training.
In the training process, it for the different feature of Type Ⅰ Ⅱ error severity during Transient Stability Prediction, introduces and pays attention to force coefficient
The loss function of algorithm is modified so that model reduces the prediction case of unstable sample；Use logistic functions
For by model output probability, the degree of reliability for weighing the output of XGBoost models to prevent part misprediction.
It finally, can be according to power grid energy pipe after the Transient Stability Evaluation model training maturation based on XGBoost algorithms
The power grid real-time traffic information that reason system is recorded is configured to the electrical measure feature of reflection Power System Steady-state operating status.Input
XGBoost models, you can to assess transient stability consequence caused by the certain possible failures of electric system in real time.
The present invention is using step in detail below：
Step 1) simulates power grid to be assessed under the various methods of operation using power system simulation software, each node, circuit
Break down caused consequence at place.The a large amount of and related initial data of the Power Network Transient Stability, specific steps are consequently formed
It is as follows：
(1) system for including c generator node and z load bus for one determines a kind of basic method of operation,
Herein under the basic method of operation, the output of each generator is respectively PGbasei, QGbasei(i=1,2 ... c), each load bus
Demand be respectively PLbasej, QLbasej(j=1,2 ... z).
(2)ρj(j=1,2 ... z) and τi(i=1,2 ... c) be the random number independently generated in setting range respectively, is passed through
These random numbers, following two formulas can be used to generate, and different system generators is contributed and workload demand situation, solution stable state are damp
After stream, the different running method of system can be obtained.
(3) during solving steady-state load flow, the uneven situation between total load and gross capability can be by the flat of system
Weighing apparatus node compensates, and under all kinds of methods of operation simulated, can collect the corresponding sample data of all kinds of failures, specifically
Way is that failure is arranged on the node, circuit of concern under the various methods of operation of generation, carries out transient stability emulation,
To obtain the temporary steady consequence of system.
Step 2) extracts feature from initial data, and determines temporarily steady label.Using generator's power and angle difference after failure whether
Whether diverging decision-making system occurs unstability situation.For various steady preview roadways, all kinds of electrical measure features are extracted, as rear
The feature input of continuous XGBoost algorithms.It is consequently formed a certain number of for establishing transient stability evaluation in power system model
Sample data.Steps are as follows for correlation computations：
(1) after grid collapses, stability by the generator rotor angle difference δ between each generator in power grid in a period of time Lai
It weighs, whether is dissipated according to δ, can system failure consequence be divided into transient stability and transient state is unstable.When system maximum generation machine
When generator rotor angle difference is less than 180 degree, system tends not to lose stabilization, when system maximum generation machine generator rotor angle difference is more than 180 degree, often
There is generator rotor angle difference Divergent Phenomenon, system will be unable to continue to keep stable operation.Thus the assessment mark of grid stability label y is given
Standard is shown below.
Wherein, max (δ) refers to after failure the maximum value of generator rotor angle difference between arbitrary two generator of system in a period of time.
(2) extraction can reflect that the feature of Power System Steady-state operating status is as shown in the table, to constitute the stable state electricity of power grid
Tolerance feature：
The electrical measure feature of stable state
Electrical features under steady state
Wherein V, theta indicate that the amplitude and phase angle of node voltage, Δ theta indicate the work(between generator node respectively
Angular difference, PG, QG, PL, QL, PB, QB indicate the active and reactive output of generator node, the active and reactive need of load bus respectively
It sums the active and reactive power of line transmission, the number of corresponding node is designated as on all variables, such as i or j.
Step 3) is using XGBoost algorithms and carries out applicability improvement, and model training is carried out using the sample data of acquisition.
In the training process, it for the different feature of Type Ⅰ Ⅱ error severity during Transient Stability Prediction, introduces and pays attention to force coefficient
The loss function of algorithm is modified so that model reduces the prediction case of unstable sample；Use logistic functions
For by model output probability, the degree of reliability for weighing the output of XGBoost models to prevent part misprediction.Correlation tool
Steps are as follows for body：
(1) XGBoost algorithm principles：For given training sample set the D={ (x with N number of sample and M featurei,
yi) (| D |=N, xi∈RM,yi∈ R), the final training results of XGBoost algorithms is one by K CART decision tree function phase
The integrated model added：
Wherein,It is the output of XGBoost models, F={ f (x)=wq(x)}(q:RM→T,w∈RT) it is CART decision trees
Set, a CART decision tree is made of tree construction q and T leaf node, and there are one successive value is right with it by each leaf node j
It answers, referred to as the weight w of leaf nodej, all weights constitute the weight vectors w ∈ R of the treeT。
Tree construction q is differentiated by attribute to be mapped to the arbitrary sample with M dimensional features on its some leaf node.
Each decision tree function fkA corresponding distinctive tree construction q and corresponding leaf node weight vectors w.For a sample
This, XGBoost models obtain final predicted valueProcess be：The sample is mapped on each decision tree corresponding
On leaf node, then the weight of the corresponding K leaf node of the sample is added.
Machine learning model can define loss function, for weighing the deviation between the predicted value of model and actual value,
In the training process, training objective is so that the value of loss function is small as far as possible.The loss function form of XGBoost models is such as
Shown in lower.
In expression formula, l is training loss function, and logarithm, which can be selected, according to the difference of Machine Learning Problems type loses letter
Number, mean square error loss function etc., for weighing predicted valueWith label value yiBetween deviation, Section 2 Ω is known as canonical
, for the complexity for the model that controlled training goes out, model is made to ensure while the accuracy on training sample, to be unlikely to
Degree is complicated, so as to avoid over-fitting, enhances generalization ability.It is defined as follows.
First item in regular terms is used to control the number of leaf node in tree-model, keeps tree construction q as simple as possible；The
Binomial is used to control the weight distribution of leaf node, and weight vectors w is made to avoid the occurrence of excessive value.γ and λ two parameters are for adjusting just
λ is generally set to 1 by the then ratio in item between two parts, only does necessary adjustment to parameter γ.
According to the loss function of definition, XGBoost models can be trained using training sample.Machine based on tree
The maximum difference in training method is learning model with common machines learning model, and such model parameter includes not only specific
Numerical value such as weight vectors w also include function fk" parameter " of this specific type, it is difficult to straight by way of gradient decline
It connects and optimizes.In XGBoost algorithms, training carries out in such a way that tree-model iteration is increased, i.e., in training process
Each step increases a CART decision tree functions f so that loss function further decreases.It is assumed thatPair the when indicating t steps
The predicted value of i sample needs to increase optimal tree construction f at this point, in order to advanced optimize modeltTo minimize at this time
Object function L(t)。
New tree construction ftSo that prediction output at this time becomesConstant is independently of change
Measure tree construction ftConstant, i.e. the corresponding regular terms of CART tree functions that has obtained before t steps, these regular terms have been calmly
Value.Choose tree construction ftStandard i.e. so that loss function L(t)Reduction amplitude it is maximum.Above formula is launched into following secondary Taylor
The form of series.
Wherein,It is loss function respectively
L is in breaking up pointThe single order and second dervative at place.In expansionIndicate what t steps obtained before
The output of all CART trees functionsWith sample label yiThe loss function of composition and a definite value.Due to losing letter
Several reduction amplitudes is unrelated with constant term, therefore, removes the constant term in above formula, can obtain the target letter simplified when t steps
Number
Define Ij=i | qt(xi)=j } it is all by tree construction qtIt is mapped to the sample number set of j-th of leaf node,
Then above-mentioned simplified object function can be further by abbreviation：
The formula is to wjDerivation can be obtained for a specific tree construction qt, optimal leaf node weight is：
Loss function formula is substituted into, this specific tree structure q is obtainedtCorresponding optimal loss function is：
This optimal loss functionArbitrary tree construction q can be weighedtQuality.It is smaller, illustrate this tree knot
Structure qtThe loss function of model can be made to decline more.
So far, the hands-on process of XGBoost models can be expressed as follows：(a) increase CART in an iterative manner
Set function, when tree-model continue growing so that model accuracy promoted amplitude be less than s when, then stop iteration, do not continue to
The number K for increasing tree-model, obtains final XGBoost models(b) in each round iteration mistake
Cheng Zhong, to obtain a new function ft, since a single leaf node structure, a leaf node is increased by one every time
Set bifurcated, in all possible tree growth scheme (scanning all can crotch and all available features), choose so that
Optimal loss functionThe scheme of minimum, so cycle carry out.The stopping division of tree can be by two state modulators：
When the depth capacity maxdepth of tree reaches specified value, or when the scheme of total division node can not be such that loss function obtains
When obtaining the decline more than γ, tree stops division, calculates this tree construction qtCorresponding optimal weights vector w is new to can be obtained
Set function ft。
(2) following logistic functions are introduced by the output probability of XGBoost models, output is transformed into (0,1) model
Within enclosing.
Selected threshold α=0.5 can obtain final prediction result and be shown below.
Such mode can convert the output of XGBoost models to transient stability and unstable two class of transient state, also, probability
OutputBe sized to reflection model prediction " degree of reliability ", it is believed that whenCloser to 1 when, model is by this sample
Be classified as 1 determination degree it is higher, whenWhen closer 0, model is higher for 0 determination degree by this sample classification.It is subsequent
Sample calculation analysis shows that, for Transient Stability Evaluation problem, the form of such probability output contributes to the reliable of decision model prediction
Degree.
(3) mistake classification and omission classification are the Type Ⅰ Ⅱ errors being likely to occur in temporarily steady assessment.Mistake classification refers to unstable
Sample (yi=1) it is classified as stablize sample, and omits classification and refer to stable sample (yi=0) it is classified as unstable sample.For
Running electric system, mistake classification will cause unstable situation ignored so that operations staff misses the adjustment method of operation
Best Times, for power system security stabilization leave hidden danger, although and omit classification and error situation, can still lead to
It crosses the means such as time-domain-simulation to further confirm that consequence, the influence for Operation of Electric Systems is relatively small.Therefore exist
When model training, mistake classification should obtain more pay attention to than omitting classification.Thus it introduces and notices that force coefficient μ improves loss letter
Number so that model is less susceptible to the problem of mistake classification occur.
Work as μ>When 1, by bigger, the training process of model will be aggravated more the ratio that first item occupies in the composition of loss function
Depending on the sample classified by mistake, so that the problem of model trained is less susceptible to that mistake classification occurs.
Step 4) finally, can be according to power grid after the Transient Stability Evaluation model training maturation based on XGBoost algorithms
The power grid real-time traffic information that Energy Management System is recorded, the electrical quantity for being configured to reflection Power System Steady-state operating status are special
Sign.Input XGBoost models, you can to assess transient stability consequence caused by the certain possible failures of electric system in real time.
Beneficial effects of the present invention：The present invention is above method of discrimination in accuracy rate and recall rate, this is because
XGBoost has preferably taken into account learning ability and generalization ability in algorithm design, although under the frame of loss function, mostly
Number machine learning algorithm can obtain preferable fitting on training set, but design of the XGBoost algorithms on regular terms makes
It has better generalization ability except training set, has higher test set accuracy rate.The present invention can be with probabilistic manner simultaneously
The difference more between determining prediction and relatively uncertain prediction is captured, it is defeated so as to avoid a part for model from missing
Go out.
Description of the drawings
39 node connection figures of Fig. 1 application examples IEEE；
Fig. 2 differences notice that XGBoost models show under force coefficient；
Fig. 3 XGBoost model misclassification sample probabilities export.
Specific implementation mode
The present invention accumulates the sample under a certain number of different running methods for studied power grid, in each operation side
Consequence caused by counting the failure institute possibility for the link that power grid is concerned under formula, while extracting the electrical quantity under each method of operation
Feature forms sample.Modeling is trained to sample using improved XGBoost algorithms later, for following needs
The online differentiation of power system transient stability.
Step 1) simulates power grid to be assessed under the various methods of operation using power system simulation software, each node, circuit
Break down caused consequence at place.The a large amount of and related initial data of the Power Network Transient Stability is consequently formed.
Step 2) extracts feature from initial data, and determines temporarily steady label.Using generator's power and angle difference after failure whether
Whether diverging decision-making system occurs unstability situation.For various steady preview roadways, all kinds of electrical measure features are extracted, as rear
The feature input of continuous XGBoost algorithms.It is consequently formed a certain number of for establishing transient stability evaluation in power system model
Sample data.
Step 3) is using XGBoost algorithms and carries out applicability improvement, and model training is carried out using the sample data of acquisition.
In the training process, it for the different feature of Type Ⅰ Ⅱ error severity during Transient Stability Prediction, introduces and pays attention to force coefficient
The loss function of algorithm is modified so that model reduces the prediction case of unstable sample；Use logistic functions
For by model output probability, the degree of reliability for weighing the output of XGBoost models to prevent part misprediction.
It, can be according to power grid energy after Transient Stability Evaluation model training maturation of the step 4) based on XGBoost algorithms
The power grid real-time traffic information that management system is recorded is configured to the electrical measure feature of reflection Power System Steady-state operating status.It is defeated
Enter XGBoost models, you can to assess transient stability consequence caused by the certain possible failures of electric system in real time.
Application examples
Apply the present invention to 39 node systems of IEEE.The system has 39 nodes, wherein generator node 10, bears
Lotus node 19,34 circuits, 12 transformer branches, system are as shown in Figure 1, wherein with the digital representation circuit in circle
Number, without the digital representation node serial number of circle.Three-phase shortcircuit ground connection event is simulated respectively in node, the circuit of different location
Barrier, to emulate the different types of transient process of the system, as table 1 illustrates four kinds of failures.For each failure, 3000 are established
Kind sample.Sample to establish process as follows：3000 kinds of different running methods for first randomly generating system, use power system mesomeric state
Analysis tool packet Matpower carries out stable state calculating, to extract the corresponding steady state characteristic collection of system under each method of operation.
Later, using the kit PSAT that can carry out transient analysis, under each method of operation, it is imitative to carry out transient state for each failure
Very, the temporary steady consequence of various failures under specific run mode is obtained, constitutes and stablizes sample and unstable sample.In transient emulation
In, generator uses quadravalence model, and when failure betides 1.0s, respective lines are cut off after continuing specified time, remove failure,
Transient process after failure occurs in 7s emulates, and counts each generator's power and angle situation, determines temporarily steady consequence label.So far
Obtain the data set modeled for machine learning.
1 IEEE of table, 39 node systems, four kinds of fault sample descriptions
By taking the failure of node 14 as an example, illustrate the effect that force coefficient is paid attention in temporarily steady loss function.3000 samples are pressed
0.85:0.15 ratio is divided into training set and test set, the training pattern on training set, and accuracy rate index is counted on test set
With recall rate index.For the failure, on the test set got at random, totally 450 samples, wherein unstable sample (label
It it is 256 for number 1), the number for stablizing sample (label 0) is 194.Iteration threshold takes s=0.001, i.e., when tree letter
When several increases can not make test set accuracy rate be lifted beyond 0.001, iteration stopping.Change the value of μ, giving its variation range is
1 to 8, it is divided into 0.5；Meanwhile changing the value of hyper parameter maxdepth and γ in a manner of cross validation, it counts in various super ginsengs
In the case of number, the maximum value and corresponding recall rate of the test set accuracy rate corresponding to each μ, as the μ values are corresponding
The optimum of XGBoost models, as shown in Figure 2.
As it can be seen that with the increase of μ, the proportion temporarily in steady loss function shared by wrong classified part is increasing, model instruction
Attention rate during white silk will be shifted to gradually by the sample of mistake classification, and the model trained such mistake occurs by more difficult
Accidentally, recall rate has the tendency that rising, becomes highest 0.992 (mistake point from 0.969 (8 samples of mistake classification) most started
2 samples of class)；It is noted, however, that force coefficient μ cannot unlimitedly increase, this is because when μ is excessive, temporarily steady loss function will
Almost occupied by its first item, model will be difficult to the study from the sample for be missed classification and learn to useful information
Stabilization and it is unstable between obscure boundary it is clear, the decline of model accuracy rate can be caused instead.According to upper figure as a result, selection μ
=3.5 the case where is best, and recall rate rising at this time reaches highest, while accuracy rate can also reach higher 0.953, still
Do not occur significantly declining.For remaining failure, best attention force coefficient is also searched out using such mode.
For 1 four kinds of failures of table, the XGBoost algorithms and support vector machines (Support after improvement loss function are used
Vector machine, SVM), logistic regression (Logistic Regression, LR), random forest (Random Forest,
RF), k neighbours (k-Nearest Neighbours, KNN) algorithm carries out temporarily steady assessment respectively, obtains accuracy rate and recall rate point
Not as shown in table 2 and table 3.
2 test set accuracy rate of table
3 test set recall rate of table
It can be found that XGBoost algorithms are above other algorithms in accuracy rate and recall rate, this is because XGBoost
Learning ability and generalization ability have preferably been taken into account in algorithm design, although under the frame of loss function, most of machines
Learning algorithm can obtain preferable fitting on training set, but design of the XGBoost algorithms on regular terms makes it instruct
Practicing except collection has better generalization ability, has higher test set accuracy rate.By the improvement to loss function, XGBoost
Algorithm is also significantly greater than other models in recall rate.
It is for statistical analysis to the misclassification situation of four kinds of failures, test set total sample number totally 1800, wherein sample mark
Label have 952 for 1, and sample label is 0 to have 848.Sample label is 1 and totally 938, sample correctly being classified,
The average value of the probability output for these samples that XGBoost models provide is 0.970, wherein the number more than 0.900 has 852
It is a, account for 90.8%；Sample label is 0 and totally 790, sample correctly being classified, these samples that XGBoost models provide
Probability output average value be 0.058, the number less than 0.2 has 704, accounts for 89.1%.This explanation is trained
XGBoost models, for more safe unstable sample, probability output is relatively very big, generally 0.9 or more, for
More safe stable sample, probability output is comparatively small, generally below 0.2.On the other hand, by the sample of misclassification
Sum is 72 (wherein mistake classification 14, omit classification 58), and the probability output situation that XGBoost models provide is as schemed
Shown in 3.
It can be found that between the probability output for the sample being mispredicted has focused largely on 0.200-0.900, compared to by just
The sample really classified, probability output value deviate more from 1 or 0.Therefore, one can be used for assess XGBoost model predictions it is reliable
The strategy of degree is：[0.2,0.9] is indicated it as indeterminacy section when its probability output is between [0.2,0.9]
Hold that degree is relatively low, other modes can be sought at this time and carry out the further differentiation of stability, for example, can directly carry out it is corresponding therefore
The time-domain-simulation of barrier.Any machine learning model is all difficult to reach absolutely accuracy, but based on the electricity of XGBoost algorithms
Force system Transient Stability Evaluation model can be captured with probabilistic manner more between determining prediction and relatively uncertain prediction
Difference, so as to avoid a part for model from accidentally exporting.
Claims (2)
1. a kind of electric power system transient stability method of discrimination based on XGBoost algorithms, which is characterized in that this method includes following
Step：
Step (1) simulates system under evaluation under the various methods of operation, each node, circuit using power system simulation software
Break down caused consequence at place；The a large amount of and related initial data of the Power Network Transient Stability is consequently formed；
Step (2) extracts feature from initial data, and determines temporarily steady label；Whether sent out using generator's power and angle difference after failure
Whether scattered decision-making system occurs unstability situation；For all kinds of steady preview roadways, all kinds of electrical measure features are extracted, as follow-up
The feature of XGBoost algorithms inputs；A certain number of sample datas for modeling electric power system transient stability are consequently formed；
Step (3) using XGBoost algorithms and carries out applicability improvement, and model training is carried out using the sample data of acquisition；
In training process, for the different feature of Type Ⅰ Ⅱ error severity during Transient Stability Prediction, introduces and pay attention to force coefficient pair
The loss function of XGBoost algorithms is modified so that model reduces the prediction case of unstable sample；Use logistic
Function is used for model output probability, and the degree of reliability for weighing the output of XGBoost models prevents part misprediction；
Step (4), after XGBoost model training maturations, the power grid real time execution recorded according to energy management system
Information is configured to the electrical measure feature of reflection Power System Steady-state operating status；Input XGBoost models, you can real time discriminating electricity
Transient stability consequence caused by the certain possible failures of Force system.
2. a kind of electric power system transient stability method of discrimination based on XGBoost algorithms according to claim 1, feature
It is：
The step (1) is specific as follows：
The system that (1-1) includes c generator node and z load bus for one determines a kind of basic method of operation,
Under this basic method of operation, the active power output of each generator is PGbasei, idle contribute is QGbasei, each load bus
Active demand is PLbasej, reactive requirement QLbasej, i=1,2 ... c, j=1,2 ... z；
(1-2)ρjAnd τiIt is the random number independently generated in setting range respectively, by random number, is generated using following two formulas
Different system generators is contributed and workload demand situation, after solving steady-state load flow, obtain the different running method of system；
(1-3) during solving steady-state load flow, uneven situation between total load and gross capability by system balance section
Point compensates, and under all kinds of methods of operation simulated, collects the corresponding sample data of all kinds of failures, is specifically generating
The various methods of operation under, failure is set on the node, circuit of concern, transient stability emulation is carried out, to obtain system
Temporary steady consequence；
The step (2) is specific as follows：
(2-1) after grid collapses, stability is weighed by the generator rotor angle difference δ between each generator in power grid in a period of time
Whether amount, dissipate according to generator rotor angle difference δ, system failure consequence be divided into transient stability and transient state is unstable；When system maximum generation
When machine generator rotor angle difference is less than 180 degree, system will not lose stabilization, when system maximum generation machine generator rotor angle difference is more than 180 degree, work(occur
Angular difference Divergent Phenomenon, system will be unable to continue to keep stable operation；Thus the evaluation criteria for giving grid stability label y is as follows
Shown in formula：
Wherein, max (δ) refers to after failure the maximum value of generator rotor angle difference between arbitrary two generator of system in a period of time；
(2-2) extracts the feature that can reflect Power System Steady-state operating status, including：Vi,thetai, Δ thetai-j, PGi,QGi,
PLi,QLi, PBi-j,QBi-j, the electrical measure feature of stable state to constitute power grid：
Wherein V, theta indicate that the amplitude and phase angle of node voltage, Δ theta indicate the generator rotor angle between generator node respectively
Difference, PG, QG, PL, QL, PB, QB indicate the active and reactive output of generator node, the active and reactive demand of load bus respectively
With the active and reactive power of line transmission；Each subscript respectively represents the node where symbol electric values；
The step (3) is specific as follows：
(1) XGBoost algorithm principles：For given training sample set the D={ (x with N number of sample and M featurei,yi)}
(| D |=N, xi∈RM,yi∈ R), the final training result of XGBoost algorithms is one and is added by K CART decision tree function
The integrated model arrived：
Wherein,It is the output of XGBoost models, F={ f (x)=wq(x)}(q:RM→T,w∈RT) be CART decision trees collection
It closes, a CART decision tree is made of tree construction q and T leaf node, and there are one each leaf node j, and successive value is corresponding with it,
The referred to as weight w of leaf nodej, all weights constitute the weight vectors w ∈ R of the treeT；
Tree construction q is differentiated by attribute to be mapped to the arbitrary sample with M dimensional features on its some leaf node；Each
Decision tree function fkA corresponding distinctive tree construction q and corresponding leaf node weight vectors w；For a sample,
XGBoost models obtain final predicted valueProcess be：The sample is mapped to corresponding leaf on each decision tree
On node, then the weight of the corresponding K leaf node of the sample is added；
Machine learning model can define loss function, for weighing the deviation between the predicted value of model and actual value, instruct
During white silk, training objective is so that the value of loss function is small as far as possible；The following institute of loss function form of XGBoost models
Show；
Wherein, l is training loss function, for weighing predicted valueWith label value yiBetween deviation, be regular terms, for controlling
Make the complexity of the model trained；It is defined as follows；
First item in regular terms is used to control the number of leaf node in tree-model；Section 2 is used to control the weight of leaf node
Distribution；γ and λ two parameters are used to adjust the ratio in regular terms between two parts；
According to the loss function of definition, XGBoost models are trained using training sample；In XGBoost algorithms, training
It is carried out in such a way that tree-model iteration is increased, i.e., each step in training process, increases a CART decision tree function f,
So that loss function further decreases；It is assumed thatThe predicted value of pair i-th of sample when t steps is indicated, at this point, in order into one
Optimized model is walked, optimal tree construction f is increasedtTo minimize object function L at this time(t)；
New tree construction ftSo that prediction output at this time becomesConstant is to be tied independently of variable tree
Structure ftConstant, i.e. the corresponding regular terms of CART tree functions that has obtained before t steps, these regular terms have been definite values；Choosing
Take tree construction ftStandard be so that loss function L(t)Reduction amplitude it is maximum；Above formula is launched into following secondary Taylor series
Form；
Wherein,It is that loss function l exists respectively
Breaking up pointThe single order and second dervative at place；In expansionIndicate that t walks the institute obtained before
There is the output of CART tree functionsWith sample label yiThe loss function of composition and a definite value；Remove in above formula
Constant term obtains the object function simplified when t steps
Define Ij=i | qt(xi)=j } it is all by tree construction qtBe mapped to the sample number set of j-th of leaf node, then it is above-mentioned
Simplify object function is further by abbreviation
The formula is to wjDerivation is obtained for a specific tree construction qt, optimal leaf node weight is：
Loss function formula is substituted into, this tree construction q is obtainedtCorresponding optimal loss function is：
This optimal loss functionFor weighing arbitrary tree construction qtQuality；It is smaller, illustrate this tree construction qt
The loss function of model is set to decline more；
The hands-on process of XGBoost models is expressed as follows：(a) increase CART tree functions in an iterative manner, when tree mould
When continuing growing so that the accuracy of model promotes amplitude less than s of type, then stop iteration, does not continue to for increasing tree-model
Number K, obtains final XGBoost models (b) in each round iterative process, to obtain one
New function ft, since a single leaf node structure, leaf node is increased into a tree bifurcated every time, it is all can
In the tree growth scheme of energy, choose so that optimal loss functionThe scheme of minimum, so cycle carry out；Tree stops
It only divides by two state modulators：When the depth capacity maxdepth of tree reaches specified value or when the side of total division node
When case can not make loss function obtain the decline more than γ, tree stops division, calculates this tree construction qtCorresponding optimal weights
Vectorial w, to which new tree function f can be obtainedt；
(2) following logistic functions are introduced by the output probability of XGBoost models, by output be transformed into (0,1) range it
It is interior；
Selected threshold α=0.5 obtains final prediction result and is shown below；
Such mode converts the output of XGBoost models to transient stability and unstable two class of transient state, also, probability output
Be sized to reflection model prediction the degree of reliability；
(3-3), which is introduced, notices that force coefficient μ improves loss function so that model is less susceptible to the problem of mistake classification occur；
Work as μ>When 1, for the ratio that first item occupies in the composition of loss function by bigger, the training process of model will more pay attention to quilt
The sample of mistake classification, so that the problem of model trained is less susceptible to that mistake classification occurs.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810381824.6A CN108551167B (en) | 2018-04-25 | 2018-04-25 | XGboost algorithm-based power system transient stability discrimination method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810381824.6A CN108551167B (en) | 2018-04-25 | 2018-04-25 | XGboost algorithm-based power system transient stability discrimination method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108551167A true CN108551167A (en) | 2018-09-18 |
| CN108551167B CN108551167B (en) | 2020-04-17 |
Family
ID=63512673
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810381824.6A Active CN108551167B (en) | 2018-04-25 | 2018-04-25 | XGboost algorithm-based power system transient stability discrimination method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108551167B (en) |
Cited By (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109657404A (en) * | 2019-01-08 | 2019-04-19 | 浙江大学 | Coalcutter automatic fault diagnosis system based on chaos amendment colony intelligence optimizing |
| CN109684834A (en) * | 2018-12-21 | 2019-04-26 | 福州大学 | A kind of gate leve hardware Trojan horse recognition method based on XGBoost |
| CN109840541A (en) * | 2018-12-05 | 2019-06-04 | 国网辽宁省电力有限公司信息通信分公司 | A kind of network transformer Fault Classification based on XGBoost |
| CN109873406A (en) * | 2019-03-28 | 2019-06-11 | 华中科技大学 | A kind of electric system weakness route discrimination method |
| CN110163540A (en) * | 2019-06-28 | 2019-08-23 | 清华大学 | Electric power system transient stability prevention and control method and system |
| CN110223193A (en) * | 2019-03-27 | 2019-09-10 | 东北电力大学 | The method of discrimination and system of operation of power networks state are used for based on fuzzy clustering and RS-KNN model |
| CN110245846A (en) * | 2019-05-28 | 2019-09-17 | 国网浙江省电力有限公司绍兴供电公司 | Current break reason method of discrimination based on XGBoost algorithm |
| CN110262465A (en) * | 2019-07-11 | 2019-09-20 | 电子科技大学 | A kind of winged control method for diagnosing faults based on error code classification |
| CN110298085A (en) * | 2019-06-11 | 2019-10-01 | 东南大学 | Analog-circuit fault diagnosis method based on XGBoost and random forests algorithm |
| CN110321651A (en) * | 2019-07-11 | 2019-10-11 | 福州大学 | A kind of transient stability method of discrimination based on regularization SVAE |
| CN110349048A (en) * | 2019-05-28 | 2019-10-18 | 国网浙江省电力有限公司绍兴供电公司 | Substation's multidimensional data runs interactive controlling platform and failure method of disposal |
| CN110535159A (en) * | 2019-07-25 | 2019-12-03 | 中国电力科学研究院有限公司 | A kind of method and system of scale energy-accumulating power station running unit fault pre-alarming |
| CN110689272A (en) * | 2019-10-09 | 2020-01-14 | 华北电力大学 | Sensitive equipment tolerance characteristic real-time online evaluation method based on stream computing |
| CN110718910A (en) * | 2019-10-29 | 2020-01-21 | 国网四川省电力公司经济技术研究院 | Transient stability evaluation method for Bayesian optimization LightGBM |
| CN110829417A (en) * | 2019-11-14 | 2020-02-21 | 电子科技大学 | Electric power system transient stability prediction method based on LSTM double-structure model |
| CN110889255A (en) * | 2019-10-31 | 2020-03-17 | 国网湖北省电力有限公司 | Power system transient stability evaluation method based on cascaded deep forest |
| CN110956010A (en) * | 2019-11-01 | 2020-04-03 | 国网辽宁省电力有限公司阜新供电公司 | Large-scale new energy access power grid stability identification method based on gradient lifting tree |
| CN110987439A (en) * | 2019-12-05 | 2020-04-10 | 山东超越数控电子股份有限公司 | Aeroengine fault prediction method based on Logitics regression and Xgboost model |
| CN111008661A (en) * | 2019-12-04 | 2020-04-14 | 哈尔滨工业大学 | Croston-XGboost prediction method for reserve demand of aircraft engine |
| US20200113505A1 (en) * | 2018-10-11 | 2020-04-16 | Seno Medical Instruments, Inc. | Optoacoustic image analysis method and system for automatically estimating lesion traits |
| CN111222255A (en) * | 2020-02-25 | 2020-06-02 | 湖南世优电力科技股份有限公司 | Big data simulation system and method for electrical fault identification |
| CN111401792A (en) * | 2020-04-16 | 2020-07-10 | 三峡大学 | Dynamic safety assessment method based on extreme gradient lifting decision tree |
| CN111404150A (en) * | 2020-03-30 | 2020-07-10 | 广西电网有限责任公司电力科学研究院 | Transient stability assessment method and system suitable for large power grid operation |
| CN111417124A (en) * | 2019-06-28 | 2020-07-14 | 西南交通大学 | Method for sensing frequency spectrum in cognitive wireless network environment |
| CN111414972A (en) * | 2020-03-30 | 2020-07-14 | 王雁 | XGboost-based automatic eye refractive correction multi-source data classification method |
| CN111445106A (en) * | 2020-03-02 | 2020-07-24 | 国网辽宁省电力有限公司电力科学研究院 | Power utilization acquisition equipment fault processing operation site safety control method and system |
| CN111585276A (en) * | 2020-05-18 | 2020-08-25 | 国网重庆市电力公司电力科学研究院 | PSS parameter online setting method and device and readable storage medium |
| CN111626508A (en) * | 2020-05-27 | 2020-09-04 | 华东师范大学 | Rail transit vehicle-mounted data prediction method based on xgboost model |
| CN111639463A (en) * | 2020-05-31 | 2020-09-08 | 西安交通大学 | XGboost algorithm-based frequency characteristic prediction method for power system after disturbance |
| CN111738477A (en) * | 2019-08-01 | 2020-10-02 | 北方工业大学 | Deep feature combination-based power grid new energy consumption capability prediction method |
| CN112308351A (en) * | 2019-07-26 | 2021-02-02 | 上海漕泾热电有限责任公司 | Heat supply balance prediction system and method for cogeneration power plant |
| CN113469457A (en) * | 2021-07-22 | 2021-10-01 | 中国电力科学研究院有限公司 | Power transmission line fault probability prediction method fused with attention mechanism |
| CN114235740A (en) * | 2021-11-12 | 2022-03-25 | 华南理工大学 | XGboost model-based waste plastic spectrum identification method |
| CN116702629A (en) * | 2023-08-04 | 2023-09-05 | 国网山西省电力公司电力科学研究院 | Power system transient stability evaluation method with migration capability |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101794998A (en) * | 2010-02-05 | 2010-08-04 | 湖南大学 | Online transient stability analysis method based on concise expression form of electromagnetic power of single generator in multi-machine power system |
| CN102074955A (en) * | 2011-01-20 | 2011-05-25 | 中国电力科学研究院 | Method based on knowledge discovery technology for stability assessment and control of electric system |
| CN106229976A (en) * | 2016-08-31 | 2016-12-14 | 山东大学 | Transient rotor angle stability situation predictor method based on data-driven |
| CN107368853A (en) * | 2017-07-14 | 2017-11-21 | 上海博辕信息技术服务有限公司 | Power network classification of the items based on machine learning determines method and device |
| CN107492043A (en) * | 2017-09-04 | 2017-12-19 | 国网冀北电力有限公司电力科学研究院 | stealing analysis method and device |
-
2018
- 2018-04-25 CN CN201810381824.6A patent/CN108551167B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101794998A (en) * | 2010-02-05 | 2010-08-04 | 湖南大学 | Online transient stability analysis method based on concise expression form of electromagnetic power of single generator in multi-machine power system |
| CN102074955A (en) * | 2011-01-20 | 2011-05-25 | 中国电力科学研究院 | Method based on knowledge discovery technology for stability assessment and control of electric system |
| CN106229976A (en) * | 2016-08-31 | 2016-12-14 | 山东大学 | Transient rotor angle stability situation predictor method based on data-driven |
| CN107368853A (en) * | 2017-07-14 | 2017-11-21 | 上海博辕信息技术服务有限公司 | Power network classification of the items based on machine learning determines method and device |
| CN107492043A (en) * | 2017-09-04 | 2017-12-19 | 国网冀北电力有限公司电力科学研究院 | stealing analysis method and device |
Non-Patent Citations (1)
| Title |
|---|
| 王同文等: "人工智能技术在电网稳定性评估中的应用综述", 《电网技术》 * |
Cited By (51)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200113505A1 (en) * | 2018-10-11 | 2020-04-16 | Seno Medical Instruments, Inc. | Optoacoustic image analysis method and system for automatically estimating lesion traits |
| CN109840541A (en) * | 2018-12-05 | 2019-06-04 | 国网辽宁省电力有限公司信息通信分公司 | A kind of network transformer Fault Classification based on XGBoost |
| CN109684834B (en) * | 2018-12-21 | 2022-10-25 | 福州大学 | XGboost-based gate-level hardware Trojan horse identification method |
| CN109684834A (en) * | 2018-12-21 | 2019-04-26 | 福州大学 | A kind of gate leve hardware Trojan horse recognition method based on XGBoost |
| CN109657404B (en) * | 2019-01-08 | 2022-09-23 | 浙江大学 | Automatic fault diagnosis system for coal mining machine based on chaos correction group intelligent optimization |
| CN109657404A (en) * | 2019-01-08 | 2019-04-19 | 浙江大学 | Coalcutter automatic fault diagnosis system based on chaos amendment colony intelligence optimizing |
| CN110223193A (en) * | 2019-03-27 | 2019-09-10 | 东北电力大学 | The method of discrimination and system of operation of power networks state are used for based on fuzzy clustering and RS-KNN model |
| CN109873406A (en) * | 2019-03-28 | 2019-06-11 | 华中科技大学 | A kind of electric system weakness route discrimination method |
| CN109873406B (en) * | 2019-03-28 | 2019-11-22 | 华中科技大学 | A kind of electric system weakness route discrimination method |
| CN110245846A (en) * | 2019-05-28 | 2019-09-17 | 国网浙江省电力有限公司绍兴供电公司 | Current break reason method of discrimination based on XGBoost algorithm |
| CN110349048A (en) * | 2019-05-28 | 2019-10-18 | 国网浙江省电力有限公司绍兴供电公司 | Substation's multidimensional data runs interactive controlling platform and failure method of disposal |
| CN110298085A (en) * | 2019-06-11 | 2019-10-01 | 东南大学 | Analog-circuit fault diagnosis method based on XGBoost and random forests algorithm |
| CN110163540B (en) * | 2019-06-28 | 2021-06-15 | 清华大学 | Power system transient stability prevention control method and system |
| CN110163540A (en) * | 2019-06-28 | 2019-08-23 | 清华大学 | Electric power system transient stability prevention and control method and system |
| CN111417124A (en) * | 2019-06-28 | 2020-07-14 | 西南交通大学 | Method for sensing frequency spectrum in cognitive wireless network environment |
| CN110262465A (en) * | 2019-07-11 | 2019-09-20 | 电子科技大学 | A kind of winged control method for diagnosing faults based on error code classification |
| CN110262465B (en) * | 2019-07-11 | 2021-05-14 | 电子科技大学 | Flight control fault diagnosis method based on fault code classification |
| CN110321651A (en) * | 2019-07-11 | 2019-10-11 | 福州大学 | A kind of transient stability method of discrimination based on regularization SVAE |
| CN110535159A (en) * | 2019-07-25 | 2019-12-03 | 中国电力科学研究院有限公司 | A kind of method and system of scale energy-accumulating power station running unit fault pre-alarming |
| CN112308351A (en) * | 2019-07-26 | 2021-02-02 | 上海漕泾热电有限责任公司 | Heat supply balance prediction system and method for cogeneration power plant |
| CN112308351B (en) * | 2019-07-26 | 2022-09-09 | 上海漕泾热电有限责任公司 | Heat supply balance prediction system and method for cogeneration power plant |
| CN111738477A (en) * | 2019-08-01 | 2020-10-02 | 北方工业大学 | Deep feature combination-based power grid new energy consumption capability prediction method |
| CN111738477B (en) * | 2019-08-01 | 2023-11-07 | 北方工业大学 | Power grid new energy consumption capability prediction method based on deep feature combination |
| CN110689272A (en) * | 2019-10-09 | 2020-01-14 | 华北电力大学 | Sensitive equipment tolerance characteristic real-time online evaluation method based on stream computing |
| CN110718910A (en) * | 2019-10-29 | 2020-01-21 | 国网四川省电力公司经济技术研究院 | Transient stability evaluation method for Bayesian optimization LightGBM |
| CN110889255B (en) * | 2019-10-31 | 2022-09-13 | 国网湖北省电力有限公司 | Power system transient stability evaluation method based on cascaded deep forest |
| CN110889255A (en) * | 2019-10-31 | 2020-03-17 | 国网湖北省电力有限公司 | Power system transient stability evaluation method based on cascaded deep forest |
| CN110956010B (en) * | 2019-11-01 | 2023-04-18 | 国网辽宁省电力有限公司阜新供电公司 | Large-scale new energy access power grid stability identification method based on gradient lifting tree |
| CN110956010A (en) * | 2019-11-01 | 2020-04-03 | 国网辽宁省电力有限公司阜新供电公司 | Large-scale new energy access power grid stability identification method based on gradient lifting tree |
| CN110829417A (en) * | 2019-11-14 | 2020-02-21 | 电子科技大学 | Electric power system transient stability prediction method based on LSTM double-structure model |
| CN111008661A (en) * | 2019-12-04 | 2020-04-14 | 哈尔滨工业大学 | Croston-XGboost prediction method for reserve demand of aircraft engine |
| CN110987439B (en) * | 2019-12-05 | 2022-03-22 | 超越科技股份有限公司 | Aeroengine fault prediction method based on Logitics regression and Xgboost model |
| CN110987439A (en) * | 2019-12-05 | 2020-04-10 | 山东超越数控电子股份有限公司 | Aeroengine fault prediction method based on Logitics regression and Xgboost model |
| CN111222255A (en) * | 2020-02-25 | 2020-06-02 | 湖南世优电力科技股份有限公司 | Big data simulation system and method for electrical fault identification |
| CN111445106A (en) * | 2020-03-02 | 2020-07-24 | 国网辽宁省电力有限公司电力科学研究院 | Power utilization acquisition equipment fault processing operation site safety control method and system |
| CN111445106B (en) * | 2020-03-02 | 2023-12-01 | 国网辽宁省电力有限公司电力科学研究院 | Safety control method and system for fault processing operation site of electricity acquisition equipment |
| CN111404150A (en) * | 2020-03-30 | 2020-07-10 | 广西电网有限责任公司电力科学研究院 | Transient stability assessment method and system suitable for large power grid operation |
| CN111414972B (en) * | 2020-03-30 | 2023-09-05 | 王雁 | XGBoost-based eye refraction correction multisource data automatic classification method |
| CN111414972A (en) * | 2020-03-30 | 2020-07-14 | 王雁 | XGboost-based automatic eye refractive correction multi-source data classification method |
| CN111401792A (en) * | 2020-04-16 | 2020-07-10 | 三峡大学 | Dynamic safety assessment method based on extreme gradient lifting decision tree |
| CN111585276B (en) * | 2020-05-18 | 2021-08-17 | 国网重庆市电力公司电力科学研究院 | PSS parameter online setting method and device and readable storage medium |
| CN111585276A (en) * | 2020-05-18 | 2020-08-25 | 国网重庆市电力公司电力科学研究院 | PSS parameter online setting method and device and readable storage medium |
| CN111626508B (en) * | 2020-05-27 | 2023-12-22 | 华东师范大学 | Track traffic vehicle-mounted data prediction method based on xgboost model |
| CN111626508A (en) * | 2020-05-27 | 2020-09-04 | 华东师范大学 | Rail transit vehicle-mounted data prediction method based on xgboost model |
| CN111639463A (en) * | 2020-05-31 | 2020-09-08 | 西安交通大学 | XGboost algorithm-based frequency characteristic prediction method for power system after disturbance |
| CN111639463B (en) * | 2020-05-31 | 2023-04-07 | 西安交通大学 | XGboost algorithm-based frequency characteristic prediction method for power system after disturbance |
| CN113469457A (en) * | 2021-07-22 | 2021-10-01 | 中国电力科学研究院有限公司 | Power transmission line fault probability prediction method fused with attention mechanism |
| CN113469457B (en) * | 2021-07-22 | 2024-04-19 | 中国电力科学研究院有限公司 | Power transmission line fault probability prediction method integrating attention mechanism |
| CN114235740A (en) * | 2021-11-12 | 2022-03-25 | 华南理工大学 | XGboost model-based waste plastic spectrum identification method |
| CN116702629A (en) * | 2023-08-04 | 2023-09-05 | 国网山西省电力公司电力科学研究院 | Power system transient stability evaluation method with migration capability |
| CN116702629B (en) * | 2023-08-04 | 2023-10-10 | 国网山西省电力公司电力科学研究院 | Power system transient stability evaluation method with migration capability |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108551167B (en) | 2020-04-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108551167A (en) | A kind of electric power system transient stability method of discrimination based on XGBoost algorithms | |
| CN102074955B (en) | Method based on knowledge discovery technology for stability assessment and control of electric system | |
| Bai et al. | A novel parameter identification approach via hybrid learning for aggregate load modeling | |
| CN108053128A (en) | A kind of Power Network Transient Stability fast evaluation method based on ELM and TF | |
| CN111368449B (en) | Cascading failure evolution path online identification method considering alternating current and direct current influences | |
| CN108667005B (en) | Power grid static and dynamic combination vulnerability assessment method considering new energy influence | |
| CN107274105A (en) | Multiple attribute decision making (MADM) tree stabilization of power grids nargin appraisal procedure based on linear discriminant analysis | |
| CN106532688A (en) | Method and system for evaluating operation reliability of micro-grid | |
| CN107153845A (en) | A kind of isolated island detection method of the adaptive grid-connected photovoltaic system based on machine learning | |
| CN109960860A (en) | Transient stability evaluation in power system method based on differential evolution extreme learning machine | |
| Sun et al. | Automatic learning of fine operating rules for online power system security control | |
| CN116973694B (en) | Power distribution network fault diagnosis optimization method and system | |
| CN110224427A (en) | A kind of information physical system modeling method based on microgrid energy control strategy | |
| CN107480432A (en) | A kind of load decomposition method based on cloud platform | |
| CN109993665A (en) | Electrical power system on-line safety and stability evaluation method, apparatus and system | |
| CN109378835A (en) | Based on the large-scale electrical power system Transient Stability Evaluation system that mutual information redundancy is optimal | |
| CN112821424A (en) | Power system frequency response analysis method based on data-model fusion drive | |
| CN109378834A (en) | Large scale electric network voltage stability margin assessment system based on information maximal correlation | |
| Zhang et al. | A data-driven method for power system transient instability mode identification based on knowledge discovery and XGBoost algorithm | |
| Wang et al. | MVMO-based identification of key input variables and design of decision trees for transient stability assessment in power systems with high penetration levels of wind power | |
| Liu et al. | Transient rotor angle stability prediction based on deep belief network and long short-term memory network | |
| CN108183481A (en) | One kind quickly sentences steady method and system based on deep learning power grid | |
| Wang et al. | Ensemble Learning Model of Power System Transient Stability Assessment Based on Bayesian Model Averaging Method | |
| He | Machine learning based method for transient stability assessment of power systems | |
| CN111814394A (en) | Power system safety assessment method based on correlation and redundancy detection |
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 |