CN109829627A - A kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme - Google Patents
A kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme Download PDFInfo
- Publication number
- CN109829627A CN109829627A CN201910011064.4A CN201910011064A CN109829627A CN 109829627 A CN109829627 A CN 109829627A CN 201910011064 A CN201910011064 A CN 201910011064A CN 109829627 A CN109829627 A CN 109829627A
- Authority
- CN
- China
- Prior art keywords
- feature
- elm
- output
- training
- integrated study
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 41
- 239000013598 vector Substances 0.000 claims abstract description 25
- 230000004913 activation Effects 0.000 claims description 21
- 238000005070 sampling Methods 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 13
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 241000218220 Ulmaceae Species 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 description 11
- 238000012502 risk assessment Methods 0.000 description 3
- 102100025093 Zinc fingers and homeoboxes protein 2 Human genes 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000008520 organization Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Abstract
The safe confidence appraisal procedure of a kind of Electrical Power System Dynamic based on integrated study scheme, firstly, in order to the feature selecting of next stage, providing good training set by carrying out feature automation sequence to input feature vector complete or collected works' data for sample training；Individual extreme learning machine (Extreme Learning Machine, ELM) training is carried out to sample set again, integrated study finally is reached by models fitting to trained ELM, to obtain reliable information.The present invention benefits from the peculiar property of ELM, pass through strategic project training and decision rule, make intelligence system (Intelligent System, IS) Fast Learning works, identify potential danger, effectively assessment is provided to dynamic secure estimation (Dynamic Security Assessment, DSA) credible result to have very important significance.
Description
Technical field
The present invention relates to Electrical Power System Dynamic security fields, and in particular to a kind of electric system based on integrated study scheme
Dynamic security confidence appraisal procedure.
Background technique
In recent years, with NETWORK STRUCTURE PRESERVING POWER SYSTEM complicate and for smart grid research and turn countries in the world into
Focus.According to China's fundamental realities of the country, construction strong reliable, economical and efficient, clean and environmental protection, transparent opening, close friend are proposed
The development strategy of the strong smart grid of the unification of interaction.Key link one of of the intelligent distribution network as smart grid, to intelligence
Power grid carries out risk assessment, is the important leverage for realizing smart grid entirety construction object.Intelligence of how making overall planning comprehensively is matched
The key risk assessment index of power grid proposes rationally effective network risk assessment method, to accurately judge network
The risk size faced is the key that the real-time dynamic secure estimation of electric system.
With the continuous renewal progress of data mining technology, such as various intelligence systems (Intelligent System,
IS) technology has been applied to quick dynamic secure estimation (Dynamic Security Assessment, DSA).But these are advanced
Data digging method all has itself some limitation: 1. cumbersome adjustment programme lengthens the training time, can not prevent very quickly
The dangerous propagation of dynamic, weaken the intensity of their online DSA；2. the prior information in training data is not perfect, classification or
Person's prediction error all may cause the DSA of inaccuracy as a result, accuracy is not high；3. robustness and generalization ability be not strong, for being
Series reaction caused by system fluctuates cannot be handled and be adapted to well.
Summary of the invention
The purpose of the present invention is being directed to DSA existing deficiency in terms of rapidity accuracy and robust generalization ability, provide
A kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme.
The purpose of invention is achieved in that
A kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme, the method includes following steps
It is rapid:
Step 1: the various characteristics that electric system is generated in real time are as an input complete or collected works, to input feature vector complete or collected works
Carry out feature automation sequence；
Step 2: the feature complete or collected works of feature automation being selected, optimal feature subset is constructed；
Step 3: stochastical sampling is carried out to optimal sample training collection；
Step 4: single threshold learning machine (Extreme Learning Machine, ELM) instruction is carried out to sample base
Practice；
Step 5: multiple ELM training sets is subjected to integrated study；
Step 6: selecting credible sub- output to form final system by decision rule and confidence level estimation and estimate.
Further, in step 1, feature automation optimal selection strategy step is as follows:
Step 1-1: using Pearson correlation coefficient (Pearson Correlation Coefficient, PCC) to input
Data characteristics carry out the calculating of feature importance；
Step 1-2: one sequence from big to small is carried out according to score to these features；
Step 1-3: corresponding feature vector is sequentially added according to score sequence, it is quasi- to calculate the corresponding classification of whole set of data
It true rate and records.
Further, in step 2, input sample integrates as O (p × f), and feature set is A={ A1, A2..., Af(wherein p is
Total number of samples, f are total characteristic number), output feature weight vector is W, W (Ai) it is AiThe weight of feature, by sampling r (r≤p)
Secondary, setting feature weight threshold value is β '.Using improved ReliefF (feature weight algorithm is suitable for multiclass sample) algorithm structure
Making optimal feature subset, steps are as follows:
Step 2-1: characteristic vector W is set to null vector；
Step 2-2: each sampling process all presses following operation: 1. selecting a random sample R；2. from R similar to sample set
In find out d arest neighbors Hj(j=1,2 ..., k)；3. being focused to find out d arest neighbors M from R inhomogeneity samplej(j=1,2 ...,
k)；4. calculating the weight of each feature for i from 1 to n；
Step 2-3: all feature weight matrix Ws (A) are obtained；
Step 2-4: it filters out feature of the weight greater than β ' and constitutes new character subset.
Further, in step 3, using Bagging (autonomous sampling method has that puts back to extract again) to the feature newly constituted
Subset is randomly choosed, and the sampling set C of N number of training sample is generatedi(i=1,2 ..., N).
Further, in step 4, it to the training of single ELM, is instructed using based on the improved robust ELM of Gauss activation primitive
Practice algorithm, steps are as follows:
Step 4-1: the weights omega of stochastic inputs hidden layer nodeiWith biasing bi；
Step 4-2: it is calculated by obtained robust activation primitive (Robust Activation Function, RAF)
Hidden layer output matrix H in ELM；
Step 4-3: output weight beta is calculated.
Further, in step 4-2, influence in view of noise and outlier to activation primitive, ELM training is swashed using robust
Function living, randomly chooses input weight and the deviation of implicit node to learn, and calculates output weight by matrix.
Further, in steps of 5, it gives E ELM to be integrated, exercise wheel number is from 1 to E.Integrated study mainly includes base
The generation of classifier and the selection of combined strategy, its step are as follows:
Step 5-1: E group training example DB is randomly selected from sampling set；
Step 5-2: the feature in trained example collection is randomly selected；
Step 5-3: this ELMh concealed nodes and a robust activation primitive, h [h in optimum range are randomly assignedmin,
hmax]([hmin, hmax] indicate concealed nodes quantity setting error minimum within this range)；
Step 5-4: the single ELM of training generates base classifier and returns；
Step 5-5: device combination is divided to determine that final output is pre- using Nearest Neighbor with Weighted Voting strategy mode for base in integrated study
Measured value.
Further, in step 6, believable sub- output is selected by adapting to decision rule such as classifying rules, prediction rule
Decision and confidence level estimation are carried out to summarize the final output of IS.
Further, by defining decision boundary in classifying rules, abnormal sub- output is counted, thus credible to classification results
Degree directly predicts that counting is bigger, and confidence level is lower.The decision rule of classification is described as follows:
E ELM is given, defining two decision boundaries is respectively [- 0.6, -2] and [0.6,2].The output category of each ELM
Rule and its reliability estimating are as follows:
Wherein: yiFor the son output of i-th of ELM unit, i=1,2 ..., E.Remember " yi=0 " son output number is α,
“yi=1 " son output number is λ, " yi=-1 " son output number is φ, then α+λ+φ=E.
If α >=η, this classification is worthless；
Otherwise, have
Wherein: η (η≤E) is user-defined threshold value, for assessing the confidence level of final classification result Y.
Further, carry out approximate integrated desired output using intermediate value in forecast and decision, following confidence level, which is estimated, to be proposed to single ELM
Meter and decision rule are as follows:
Wherein: yiSon for i-th of ELM unit exports, i=1,2 ..., E,It is sub- output vector [y1...yi...yE]
Intermediate value.
Given E ELMs set, wherein set has ρ insincere sub- outputs and a believable sub- outputs of δ, then ρ+σ=E.
If ρ >=η, this prediction be it is worthless, then(Y is effectively output prediction, yiIt is the son of i-th of ELM unit
Output).
Wherein: η (η≤E) is user-defined threshold value, for assessing the confidence level of final prediction result Y.
Compared with prior art, the beneficial effects of the present invention are:
It will be dependent on appropriate fitting electric system feature and corresponding dynamic security to traditional Data Mining Tools 1. comparing
Mapping relations between index, the present invention utilize parallel organization, are conducive to model structure and stablize, keep integrated study scheme online
DSA enhanced strength is conducive to improve system structure stability.
2. in terms of DSA result precision of the present invention: integrated study scheme has specific database first, uses reliable spirit
Preparatory failure DSA mechanism living, the prior information for avoiding training data are not perfect；Secondly IS can carry out newest operation information
Online updating, keep On-line funchon, it is ensured that the integrity of information；Finally further protected by means of the efficient tuning mechanism of ELM
Demonstrate,prove the accuracy of DSA result.It compares and other methods, the present invention can be to avoid using insecure DSA result.
3. using unique randoming scheme in the present invention, a series of single ELM are assembled, each individual is not only random
Input weight is selected, but also makes the other parameters randomization in training, enhances system robustness and generalization ability.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples:
Fig. 1 is present system flow chart；
Fig. 2 is extreme learning machine structure chart of the present invention；
Fig. 3 is parallel organization simulation drawing of the present invention；
Fig. 4 is integrated model frame diagram of the present invention；
Fig. 5 is present example confidence level and accuracy rate relational graph.
Specific embodiment
The present invention provides a kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme, the sides
Method includes the following steps, as shown in Figure 1:
Step 1 in Fig. 1: the various characteristics that electric system is generated in real time are special to input as an input complete or collected works
It levies complete or collected works and carries out feature automation sequence, it is as follows that feature automates optimal selection strategy step:
Step 1-1: the calculating of feature importance is carried out using data characteristics of the Pearson correlation coefficient to input；
Step 1-2: one sequence from big to small is carried out according to score to these features；
Step 1-3: corresponding feature vector is sequentially added according to score sequence, it is quasi- to calculate the corresponding classification of whole set of data
It true rate and records.
Step 2 in Fig. 1: input sample integrates as O (p × f), and feature set is A={ A1, A2..., Af(wherein p is total sample
Number, f are total characteristic number), output feature weight vector is W, W (Ai) it is AiThe weight of feature, setting secondary by sampling r (r≤p)
Feature weight threshold value is β '.Using improved ReliefF (feature weight algorithm is suitable for multiclass sample) the optimal spy of algorithm construction
Levying subset, steps are as follows:
Step 2-1: characteristic vector W is set to null vector；
Step 2-2: each sampling process all presses following operation: 1. selecting a random sample R；2. from R similar to sample set
In find out d arest neighbors Hj(j=1,2 ..., k)；3. being focused to find out d arest neighbors M from R inhomogeneity samplej(j=1,2 ...,
k)；4. calculating the weight of each feature for i from 1 to n, then having:
In formula: diff (Ai, S1, S2) indicate two sample S1、S2Distinctiveness ratio difference, i.e. S1、S2In feature AiOn distance.
diff(Ai, R, Hj) smaller, diff (Ai, R, Mj) more big, indicate feature AiThere are preferable classifying quality, respective weights W (Ai) just
It is bigger.
A, discrete type feature
B, continuous type feature
In formula: V (Ai, S1) indicate sample S1In feature AiThe value at place, max (Ai) indicate sample S1In feature AiThe maximum at place
Value, min (Ai) indicate sample S1In feature AiThe minimum value at place.
Step 2-3: all feature weight matrix Ws (A) are obtained；
Step 2-4: it filters out feature of the weight greater than β ' and constitutes new character subset.
Step 3 in Fig. 1: stochastical sampling is carried out to optimal sample training collection, Bagging can be used to the feature newly constituted
Subset is randomly choosed, and the sampling set C of N number of training sample is generatedi(i=1,2 ..., N).
Step 4 in Fig. 1: single ELM training is carried out to sample base, using based on the improved robust of Gauss activation primitive
ELM training algorithm, steps are as follows:
Step 4-1: the weights omega of stochastic inputs hidden layer nodeiWith biasing bi；
Step 4-2: hidden layer output matrix H in ELM is calculated by obtained robust activation primitive；
Step 4-3: output weight beta is calculated.
Wherein, in step 4-2, influence in view of noise and outlier to activation primitive, and the structure of ELM such as Fig. 2 institute
Show, it may be considered that from the activation primitive of hidden layer, ELM training can utilize traditional Gauss activation primitive is transformed to obtain it is new
Robust activation primitive, randomly chooses input weight and the deviation of implicit node to learn, and calculates output weight by matrix.
Given one contains the training dataset of N number of exampleWherein xjFor n ×
1 input vector, RnBelong to input vector set, tjFor the object vector of m × 1, RmBelong to object vector set.By constructing N
A implicit node and one group of function parameter βi、giAnd bi, corresponding hidden layer output function are as follows:
In formula:It is activation primitive, ωiAnd biIt is i-th of section between connection input layer and hidden layer respectively
The weight vectors of point and biasing, βiIndicate the weight vectors of i-th of node between hidden layer and output layer.
Gauss activation primitive is standardized asIf by sampleAnd it is defeated
Enter weightIt is unitization, then obtain RAF activation primitive are as follows: g (ωi, bi, xi)=exp {-bi(1-cosθi), (θi
=< ωi, xi> it is ωiAnd xiInner product).
At this point, due to training dataIt is each in total number of samples N
Input sample (xi, ti), in the network structure containing L hidden layer node, activation primitiveModel can indicate are as follows:
Above-mentioned formula can be write as matrix form: H β=Z
In formula: H is the hidden layer output matrix of the neural network, and β is output weight matrix, and Z is target output matrix.
Step 5 in Fig. 1: given E ELM is integrated, and exercise wheel number is from 1 to E.Integrated study mainly includes base classifier
Generation and combined strategy selection, its step are as follows:
Step 5-1: E group training example DB is randomly selected from sampling set；
Step 5-2: the feature in trained example collection is randomly selected；
Step 5-3: this ELMh concealed nodes and a robust activation primitive, h [h in optimum range are randomly assignedmin,
hmax]([hmin, hmax] indicate concealed nodes quantity setting error minimum within this range)；
Step 5-4: the single ELM of training generates base classifier and returns；
Step 5-5: device combination is divided to determine that final output is pre- using Nearest Neighbor with Weighted Voting strategy mode for base in integrated study
Measured value.
Step 6 in Fig. 1: it selects credible sub- output to form final system by decision rule and confidence level estimation and estimates.
Credible sub- output is such as selected by classifying rules, prediction rule to summarize the final output of IS, carries out decision and confidence level estimation.
(1) by defining decision boundary in classifying rules, abnormal sub- output is counted, thus straight to classification results confidence level
Prediction is connect, counting is bigger, and confidence level is lower.The decision rule of classification is described as follows:
E ELM is given, defining two decision boundaries is respectively [- 0.6, -2] and [0.6,2].The output category of each ELM
Rule and its reliability estimating are as follows:
Wherein: yiFor the son output of i-th of ELM unit, i=1,2 ..., E.Remember " yi=0 " son output number is α,
“yi=1 " son output number is λ, " yi=-1 " son output number is φ, then α+λ+φ=E.
If α >=η, this classification is worthless；
Otherwise, have
Wherein: η (η≤E) is user-defined threshold value, for assessing the confidence level of final classification result Y.
(2) carry out approximate integrated desired output using intermediate value in forecast and decision, to single ELM propose following reliability estimating and
Decision rule is as follows:
Wherein: yiSon for i-th of ELM unit exports, i=1,2 ..., E,It is sub- output vector [y1...yi...yE]
Intermediate value.
Given E ELMs set, wherein set has ρ insincere sub- outputs and a believable sub- outputs of δ, then ρ+σ=E.
If ρ >=η, this prediction be it is worthless, then((Y is effectively output prediction, yiIt is the son of i-th of ELM unit
Output)).
Wherein: η (η≤E) is user-defined threshold value, for assessing the confidence level of final prediction result Y.
Embodiment 1:
It is the system-computed time and behavior pattern table that the present invention tests in IEEE-50 machine system described in table 1.It should
System is made of 50 generators, 145 nodes and 453 branches.In following test, using 200 ELM, and phase is set
The parameter answered, randomly selects 3000 samples and 30 features, sets 40 for Reliability estimation parameter, randomly selects 20%
Database is as test data, remaining is as training data.Single Expeditious Plan, the multiple emergency feelings of last test are considered first
Condition.The average training of single ELM and testing time are only 1.54 seconds and 0.0313 second respectively in verification process.Test IS simultaneously comprehensively
It has been shown that, as shown in table 1: confidence level 93.85%, accuracy 100%, i.e. 1191 in 1269 examples can be reliable by IS
It determines, and their classification is 100% correct, it means that IS can detect potential classification error well.
As shown in figure 5, the confidence level and accuracy rate of system testing can also be sent out when changing training data sample size size
Changing.A line represents accuracy rate above Fig. 5, below a line represent confidence level, when use 200 ELM, 500 training
When data, confidence level and accuracy rate are respectively 98.93% and 98.21%.It is apparent that confidence level is general as training data increases
All over decline, accuracy rate is improved, and keeps 100% when being greater than 2000.This means that IS passes through sacrificial when training data quantity increases
Domestic animal can decline the precise classification that cost is cost to provide 100%.
Table 1
Experimental example 2:
What table 2 was tested is reality of China world power grid dynamic equivalent system-computed time and behavior pattern, have 39 motors,
120 load nodes, 223 alternating current circuits and 4 HVDC transmission lines.In this test, 784 operation samples are chosen
The candidate characteristics of this and 196 randomly select 20% database as test data, remaining has respectively as training data
157 test cases and 627 trained examples, training data is for adjusting single ELM.Finally, comprehensive test result such as 2 institute of table
Show, this demonstrates the rule shown by Fig. 5, IS can be very good identification prediction mistake.
Table 2
Claims (10)
1. a kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme, which is characterized in that including following
Step:
Step 1: the various characteristics that electric system is generated in real time carry out input feature vector complete or collected works as an input complete or collected works
Feature automation sequence；
Step 2: the feature complete or collected works of feature automation being selected, optimal feature subset is constructed；
Step 3: stochastical sampling is carried out to optimal sample training collection；
Step 4: the training of single threshold learning machine is carried out to sample base；
Step 5: multiple ELM training sets is subjected to integrated study；
Step 6: selecting credible sub- output to form final system by decision rule and confidence level estimation and estimate.
2. the safe confidence appraisal procedure of a kind of Electrical Power System Dynamic based on integrated study scheme according to claim 1,
It is characterized by: in step 1, it is as follows to carry out feature automation sequence step to input feature vector complete or collected works:
Step 1-1: the calculating of feature importance is carried out using data characteristics of the Pearson correlation coefficient to input；
Step 1-2: one sequence from big to small is carried out according to score to these features；
Step 1-3: corresponding feature vector is sequentially added according to score sequence, calculates the corresponding classification accuracy of whole set of data
And it records.
3. the safe confidence appraisal procedure of a kind of Electrical Power System Dynamic based on integrated study scheme according to claim 1,
It is characterized by: in step 2, input sample integrates as O (p × f), feature set is A={ A1, A2..., Af(wherein p is gross sample
This number, f are total characteristic number), output feature weight vector is W, W (Ai) it is AiThe weight of feature, it is secondary by sampling r (r≤p), if
Setting feature weight threshold value is β '.
4. the safe confidence appraisal procedure of a kind of Electrical Power System Dynamic based on integrated study scheme according to claim 1,
It is characterized by: in step 2, using improved ReliefF (feature weight algorithm is suitable for multiclass sample), algorithm construction is most
Steps are as follows for excellent character subset:
Step 2-1: characteristic vector W is set to null vector；
Step 2-2: each sampling process all presses following operation: 1. selecting a random sample R；2. from R similar to being looked in sample set
D arest neighbors H outj(j=1,2 ..., k)；3. being focused to find out d arest neighbors M from R inhomogeneity samplej(j=1,2 ..., k)；
4. calculating the weight of each feature for i from 1 to n；
Step 2-3: all feature weight matrix Ws (A) are obtained；
Step 2-4: it filters out feature of the weight greater than β ' and constitutes new character subset.
5. the safe confidence appraisal procedure of a kind of Electrical Power System Dynamic based on integrated study scheme according to claim 1,
It is characterized by: in step 3, using Bagging (autonomous sampling method has that puts back to extract again) to the character subset newly constituted
It is randomly choosed, generates the sampling set C of N number of training samplei(i=1,2 ..., N).
6. the safe confidence appraisal procedure of a kind of Electrical Power System Dynamic based on integrated study scheme according to claim 1,
It is characterized by: in step 4, the training to single ELM is calculated using based on the improved robust ELM training of Gauss activation primitive
Method, steps are as follows:
Step 4-1: the weights omega of stochastic inputs hidden layer nodeiWith biasing bi；
Step 4-2: it is calculated in ELM by obtained robust activation primitive (Robust Activation Function, RAF)
Hidden layer output matrix H；
Step 4-3: output weight beta is calculated.
7. the safe confidence appraisal procedure of a kind of Electrical Power System Dynamic based on integrated study scheme according to claim 6,
It is characterized by: described in step 4-2, the influence in view of noise and outlier to activation primitive, ELM training is swashed using robust
Function living, randomly chooses input weight and the deviation of implicit node to learn, and calculates output weight by matrix.
8. the safe confidence appraisal procedure of a kind of Electrical Power System Dynamic based on integrated study scheme according to claim 1,
It is characterized by: in steps of 5, given E ELM is integrated, for exercise wheel number from 1 to E, integrated study mainly includes base classification
The generation of device and the selection of combined strategy, its step are as follows:
Step 5-1: E group training example DB is randomly selected from sampling set；
Step 5-2: the feature in trained example collection is randomly selected；
Step 5-3: this ELMh concealed nodes and a robust activation primitive, h [h in optimum range are randomly assignedmin, hmax]
([hmin, hmax] indicate concealed nodes quantity setting error minimum within this range)；
Step 5-4: the single ELM of training generates base classifier and returns；
Step 5-5: device combination is divided to determine final output predicted value using Nearest Neighbor with Weighted Voting strategy mode for base in integrated study.
9. the safe confidence appraisal procedure of a kind of Electrical Power System Dynamic based on integrated study scheme according to claim 1,
It is characterized by: in step 6, selecting believable sub- output to converge by adapting to decision rule such as classifying rules, prediction rule
The final output of total intelligence system (Intelligent System, IS) carries out decision and confidence level estimation, leads in classifying rules
Definition decision boundary is crossed, abnormal sub- output is counted, to directly predict classification results confidence level, counts bigger, confidence level
Lower, the decision rule of classification is described as follows:
E ELM is given, defining two decision boundaries is respectively [- 0.6, -2] and [0.6,2], the output category rule of each ELM
And its reliability estimating is as follows:
Wherein: yiFor the son output of i-th of ELM unit, i=1,2 ..., E remember " yi=0 " son output number is α, " yi=
1 " son output number is λ, " yi=-1 " son output number is φ, then α+λ+φ=E,
If α >=η, this classification is worthless；
Otherwise, have
Wherein: η (η≤E) is user-defined threshold value, for assessing the confidence level of final classification result Y.
10. the safe confidence appraisal procedure of a kind of Electrical Power System Dynamic based on integrated study scheme according to claim 9,
It is characterized by: carry out approximate integrated desired output using intermediate value in forecast and decision, to single ELM propose following reliability estimating and
Decision rule is as follows:
Wherein: yiSon for i-th of ELM unit exports, i=1,2 ..., E,It is sub- output vector [y1...yi...yE] in
Value；
Given E ELMs set, wherein set has ρ insincere sub- outputs and a believable sub- outputs of β, then ρ+σ=E, if ρ
>=η, this prediction be it is worthless, then(Y is effectively output prediction, yiIt is the son output of i-th of ELM unit)；
Wherein: η (η≤E) is user-defined threshold value, for assessing the confidence level of final prediction result Y.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910011064.4A CN109829627A (en) | 2019-01-04 | 2019-01-04 | A kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910011064.4A CN109829627A (en) | 2019-01-04 | 2019-01-04 | A kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109829627A true CN109829627A (en) | 2019-05-31 |
Family
ID=66860013
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910011064.4A Pending CN109829627A (en) | 2019-01-04 | 2019-01-04 | A kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109829627A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110264116A (en) * | 2019-07-31 | 2019-09-20 | 三峡大学 | A kind of Electrical Power System Dynamic safety evaluation method explored based on relationship with regression tree |
| CN110428005A (en) * | 2019-07-31 | 2019-11-08 | 三峡大学 | A kind of safe misclassification constrained procedure of Electrical Power System Dynamic based on umbrella-type algorithm |
| CN110765454A (en) * | 2019-10-22 | 2020-02-07 | 深圳大学 | Running state reconstruction method of power system and terminal |
| CN112485394A (en) * | 2020-11-10 | 2021-03-12 | 浙江大学 | Water quality soft measurement method based on sparse self-coding and extreme learning machine |
| CN112926397A (en) * | 2021-01-28 | 2021-06-08 | 中国石油大学(华东) | SAR image sea ice type classification method based on two-round voting strategy integrated learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105512808A (en) * | 2015-11-30 | 2016-04-20 | 武汉大学 | Power system transient stability assessment method based on big data |
| CN108053128A (en) * | 2017-12-25 | 2018-05-18 | 东南大学 | A kind of Power Network Transient Stability fast evaluation method based on ELM and TF |
| CN108595884A (en) * | 2018-05-09 | 2018-09-28 | 清华大学 | Power system transient stability appraisal procedure and device |
| CN109086913A (en) * | 2018-07-11 | 2018-12-25 | 山东大学 | A kind of transient stability evaluation in power system method and system based on deep learning |
-
2019
- 2019-01-04 CN CN201910011064.4A patent/CN109829627A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105512808A (en) * | 2015-11-30 | 2016-04-20 | 武汉大学 | Power system transient stability assessment method based on big data |
| CN108053128A (en) * | 2017-12-25 | 2018-05-18 | 东南大学 | A kind of Power Network Transient Stability fast evaluation method based on ELM and TF |
| CN108595884A (en) * | 2018-05-09 | 2018-09-28 | 清华大学 | Power system transient stability appraisal procedure and device |
| CN109086913A (en) * | 2018-07-11 | 2018-12-25 | 山东大学 | A kind of transient stability evaluation in power system method and system based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 刘颂凯: "基于数据与机理的电力系统安全评估研究", 《中国优秀博士论文期刊》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110264116A (en) * | 2019-07-31 | 2019-09-20 | 三峡大学 | A kind of Electrical Power System Dynamic safety evaluation method explored based on relationship with regression tree |
| CN110428005A (en) * | 2019-07-31 | 2019-11-08 | 三峡大学 | A kind of safe misclassification constrained procedure of Electrical Power System Dynamic based on umbrella-type algorithm |
| CN110428005B (en) * | 2019-07-31 | 2022-11-08 | 三峡大学 | Umbrella algorithm-based dynamic security misclassification constraint method for power system |
| CN110765454A (en) * | 2019-10-22 | 2020-02-07 | 深圳大学 | Running state reconstruction method of power system and terminal |
| CN112485394A (en) * | 2020-11-10 | 2021-03-12 | 浙江大学 | Water quality soft measurement method based on sparse self-coding and extreme learning machine |
| CN112926397A (en) * | 2021-01-28 | 2021-06-08 | 中国石油大学(华东) | SAR image sea ice type classification method based on two-round voting strategy integrated learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829627A (en) | A kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme | |
| Yang et al. | SMAA-based model for decision aiding using regret theory in discrete Z-number context | |
| CN102496069B (en) | Cable multimode safe operation evaluation method based on fuzzy analytic hierarchy process (FAHP) | |
| Liao et al. | Voltage sag estimation in sparsely monitored power systems based on deep learning and system area mapping | |
| CN115270965B (en) | Power distribution network line fault prediction method and device | |
| CN108399248A (en) | A kind of time series data prediction technique, device and equipment | |
| CN107766418A (en) | A kind of credit estimation method based on Fusion Model, electronic equipment and storage medium | |
| CN109242149A (en) | A kind of student performance early warning method and system excavated based on educational data | |
| Yang et al. | Multi-objective short-term hydropower generation operation for cascade reservoirs and stochastic decision making under multiple uncertainties | |
| CN106569030B (en) | Alarm threshold optimization method and device in a kind of electrical energy measurement abnormity diagnosis | |
| CN104077493B (en) | Method for constructing state evaluation index system of electric relaying protection system | |
| CN105740635B (en) | A kind of cloud ideal solution evaluation method of transformer electromagnetic design scheme | |
| Oludolapo et al. | Comparing performance of MLP and RBF neural network models for predicting South Africa's energy consumption | |
| CN109033513A (en) | Method for diagnosing fault of power transformer and diagnosing fault of power transformer device | |
| CN106327357A (en) | Load identification method based on improved probabilistic neural network | |
| Mishra et al. | Rainfall-runoff modeling using clustering and regression analysis for the river brahmaputra basin | |
| CN108549907A (en) | A kind of data verification method based on multi-source transfer learning | |
| CN110378358A (en) | A kind of power distribution network isomeric data integration method and system | |
| CN114266289A (en) | Complex equipment health state assessment method | |
| Li et al. | Analysis and modelling of flood risk assessment using information diffusion and artificial neural network | |
| Ullah et al. | Adaptive data balancing method using stacking ensemble model and its application to non-technical loss detection in smart grids | |
| CN106874676A (en) | A kind of electric power meter state evaluating method | |
| Kim et al. | Anomaly detection using clustered deep one-class classification | |
| CN108830407B (en) | Sensor distribution optimization method in structure health monitoring under multi-working condition | |
| CN110533341A (en) | A kind of Livable City evaluation method based on BP neural network |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190531 |
|
| RJ01 | Rejection of invention patent application after publication |