CN106786560A - A kind of power system stability characteristic automatic extraction method and device - Google Patents
A kind of power system stability characteristic automatic extraction method and device Download PDFInfo
- Publication number
- CN106786560A CN106786560A CN201710078550.9A CN201710078550A CN106786560A CN 106786560 A CN106786560 A CN 106786560A CN 201710078550 A CN201710078550 A CN 201710078550A CN 106786560 A CN106786560 A CN 106786560A
- Authority
- CN
- China
- Prior art keywords
- logistic regression
- cost function
- parameter matrix
- input quantity
- classification
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- 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]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Power Engineering (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
A kind of power system stability characteristic automatic extraction method of present invention proposition and device, its sample input quantity for including determining logistic regression training pattern and its classification quantity, and define hyper parameter set；Cost function of the definition comprising LASSO penalty terms, the logistic regression parameter matrix θ in the cost function with LASSO penalty terms is updated using iterative method；And logistic regression training pattern that error rate deviation in the hyper parameter set generated less than the maximum hyper parameter of threshold value is chosen as optimal models；Using the maximum of the logistic regression parameter matrix θ row vector absolute value sums of the optimal models as stabilization of power grids feature.The program is fitted by Logic Regression Models to the degree of stability (CCT) of power network forecast failure, and the penalty function of 1 norm of parameter matrix θ is added in cost function, is highlighted principal character；It is achieved thereby that invariant feature is automatically extracted.
Description
Technical field
The invention belongs to bulk power grid stabilization and control field, and in particular to be a kind of power system stability Automatic signature extraction
Method and device.
Background technology
With the lifting at full speed of economic level, Chinese society is also increasingly strong to the demand of electric power.In order to ensure that electric energy is pacified
Complete reliable transmission, has carried out the Important Projects, alternating current-direct current such as transferring electricity from the west to the east, on national network and UHV transmission in China Power Grids
The short-term load of series-parallel connection has been basically formed.With the expansion of power network scale, power grid security and stability is difficult to control further.The world
On the multiple electric network fault that has occurred and that show, the raising of transmission voltage grade, Interconnection Scale expand and transmission capacity increasing
Plus, can all increase the harm that electric network fault brings, failure cause and process are also increasingly complex.It is comprehensively careful to operation power network to carry out
In-service monitoring, analysis and control, ensure that power generation, the safety transmitting and use are the active demands of various countries' power industry.
Carry out the work of powernet security and stability analysis, calculating speed is one of the core index that must ensure, if lost
Go calculating speed, then on-line analysis also just loses ageing, and becomes nonsensical.Existing on-line analysis system is mainly adopted
It is analyzed with time-domain simulation method, amount of calculation is larger, it is difficult to further lift speed；And use and quickly sentence steady method, although
Speed, but it is highly dependent on the selection of stabilization of power grids feature, and inaccurate feature will cause the very big mistake for predicting the outcome
Difference.On the other hand, on-line analysis system have accumulated substantial amounts of historical data, wherein contained the operation of power networks rule of preciousness, together
When again closing to reality ruuning situation, can as invariant feature recognize foundation.Target of the invention be exactly using historical data come
Stabilization of power grids feature under different modes and forecast failure is recognized automatically.
With the expansion of power network scale, influence the factor of the stabilization of power grids to become increasingly complex, and therefrom effectively excavate crucial
Invariant feature also just turn into control operation of power networks important topic.The existing certain research at present of invariant feature discrimination method,
Existing method often excessively relies on artificial experience, and selected feature compares limitation, it is impossible to widely choose, cause exist leakage choosing can
Energy.
The content of the invention
In order to overcome drawbacks described above, a kind of power system stability characteristic automatic extraction method and device are proposed, using history
Data are recognized to the stabilization of power grids feature under different modes and forecast failure automatically.
The purpose of the present invention is realized using following technical proposals：
A kind of power system stability characteristic automatic extraction method, methods described includes：
Determine the sample input quantity and its classification quantity of logistic regression training pattern, and define hyper parameter set；
Cost function of the definition comprising LASSO penalty terms, the cost comprising LASSO penalty terms is updated using iterative method
Logistic regression parameter matrix θ in function；
The value in proportion under index coordinate system, chooses maximum of the error rate deviation less than threshold value in the hyper parameter set
The logistic regression training pattern that hyper parameter is generated is used as optimal models；
It is special using the maximum of the logistic regression parameter matrix θ row vector absolute value sums of the optimal models as the stabilization of power grids
Levy.
Preferably, the sample input quantity for determining logistic regression training pattern includes：Network system failure is carried out from
Line is emulated, and is always generated electricity with the power of the assembling unit, line power, plant stand and plant stand total load is to be input into, during with the critical excision of discretization
Between for output, using critical clearing time as stabilization of power grids level index, so that it is determined that the sample of logistic regression training pattern is defeated
Enter amount.
Preferably, carrying out classification to the sample input quantity includes：By the logistic regression parameter of logistic regression training pattern
Matrix θ is mapped in [0,1] interval range, self-defined line of demarcation, be will be above respectively or less than the logistic regression in the line of demarcation
Parameter is classified；
The classification quantity of the logistic regression training pattern is determined by following formula：
In formula, hθX () represents the probability that 1 is taken for training sample x classification results, x is sample input quantity, and θ is returned for logic
Return parameter matrix.
Preferably, it is described to define hyper parameter set λ={ λ1, λ2..., λn, λ ∈ [0,1], n are the number of hyper parameter.
Preferably, the cost function of the definition comprising LASSO penalty terms includes：Increase a parameter in cost function
The penalty function of the norm of θ matrixes 1, obtains cost function of the following formula comprising LASSO penalty terms：
In formula, J (x) is the cost function of x, and x is sample input quantity, and y is sample classification label, λiIt is surpassing for adjustment ratio
Parameter, for obtaining balance between the penalty value of cost function and logistic regression parameter matrix θ, m is total sample number, and k is to divide
Class quantity.
Preferably, the utilization iterative method updates the logistic regression ginseng in the cost function comprising LASSO penalty terms
Matrix number θ includes：
A. initialization logic regression parameter matrix θ；
B. θ is substituted into formula (2) and solves cost function, and gradient of the cost function to θ is obtained using gradient descent method；
C. θ is updated according to Grad and default learning rate iteration；
D. judge that cost function is not reduced within the unit interval or iterations reaches the upper limit, if so, then exporting current
θ, if otherwise return to step b.
Further, the initialization logic regression parameter matrix θ includes：By m sample input quantity x={ x1,x2,…,
xm, and its corresponding tag along sort y={ y1,y2,…,ym, a characteristic vector for k dimensions is constituted, by m k dimensional feature vector
Constitute a logistic regression parameter matrix θ of m*k.
A kind of power system stability Automatic signature extraction device, including：
Setup module, for determining the sample input quantity and its classification quantity of logistic regression training pattern, and defines super ginseng
Manifold is closed；
Update module, for defining the cost function comprising LASSO penalty terms, includes using described in iterative method renewal
Logistic regression parameter matrix θ in the cost function of LASSO penalty terms；
Acquisition module, for the value in proportion under index coordinate system, chooses error rate deviation in the hyper parameter set
The logistic regression training pattern generated less than the maximum hyper parameter of threshold value is used as optimal models；
Definition module, for by the maximum of the logistic regression parameter matrix θ row vector absolute value sums of the optimal models
As stabilization of power grids feature.
Compared with immediate prior art, beneficial effects of the present invention are：
The present invention program proposes a kind of active distribution network uncertainty Power flow simulation method and device, determines that logistic regression is instructed
Practice the sample input quantity and its classification quantity of model, and define hyper parameter set；The critical clearing time of forecast failure will be made
It is stabilization of power grids level index, using the historical data produced in electrical power system on-line safety and stability analysis system, determines logic
The sample input quantity and its classification quantity of regression training model, and define hyper parameter set；It is special from data own characteristic and distribution
Levy and set out, expand range of choice, the static amount of armamentarium in power system online data can be covered；
Cost function of the definition comprising LASSO penalty terms, the cost comprising LASSO penalty terms is updated using iterative method
Logistic regression parameter matrix θ in function；Outer loop is asked the Logic Regression Models that hyper parameter λ takes different numerical value one by one
Solution；Inner loop is the iterative process of solution logic regression model；After the completion of circulation, optimal models is obtained by contrast, finally
It is automatic to pick out the principal character maximum for stabilization of power grids influence.Example of calculation demonstrates the validity of this method, selected spy
The expection for meeting stabilization of power grids analysis theories is levied, can be further used for quickly sentencing the advanced analysis functions such as steady, method of operation contrast.
Brief description of the drawings
Fig. 1 is a kind of power system stability characteristic automatic extraction method flow chart of offer in the embodiment of the present invention；
Specific embodiment
A kind of power system stability characteristic automatic extraction method of the present invention and device, its key point are logistic regression algorithm,
Existing historical data and its stability indicator are fitted by Logic Regression Models, and are added in cost function part
LASSO penalty terms.When the model for obtaining has preferably performance in test data, it is believed that can the model reflect power network
Stability characteristic (quality).And then according to the characteristics of logistic regression, the larger input quantity of absolute value is used as power network in preference pattern parameter matrix
Feature, algorithm performs are finished.
This algorithm can be divided into following two big steps：
1st, model learning
It is identical with common logistic regression, the method that this algorithm is also adopted by iterative：First, input quantity and classification are determined
Quantity, and then determine the dimension of parameter matrix θ and initialize；Secondly, define shown in cost function such as formula (4)；Finally carry out generation
The iterative process of valency function minimization, each iteration asks for cost function pair all according to θ currency calculation cost functions
The gradient of θ, the learning rate that sets according to Grad and in advance updates θ；If meet iterations requirement, or cost function
Long period does not decline, then exit iteration.After obtaining Logic Regression Models, in addition it is also necessary to tested by test set, if test
The error of collection is excessive, then it is assumed that the model is unavailable.
Because cost function is determined jointly by the penalty term of actual cost and parameter matrix θ, it is therefore desirable to by super ginseng
Number λ is balanced between two.The larger effects for being to emphasize θ penalty terms of λ, may cause model actual error bigger than normal, it is impossible to
Reflection stabilization of power grids characteristic；And the smaller effects for being to emphasize actual cost of λ, the compression of parameter matrix θ may be caused inadequate, it is impossible to
Therefrom select principal character.It can be seen that, the selection of λ is one of key issue of this algorithm.
This algorithm is used under same training set and test set data, is repeatedly learnt based on different λ values, and by right
The method of ratio obtains optimal λ value.Do not consider penalty term, i.e., (λ=0) when, due to the limitation in the absence of penalty term, gained model
Error rate it is necessarily minimum, can be with this benchmark as a comparison.λ value can be under index coordinate system according to actual conditions, by certain
Ratio value, for example select 1,0.1, the model obtained by 0.01... each different λ be compared with benchmark model, selection mistake
The model that the maximum λ that rate deviation is less than threshold value is generated is used as optimal models.
2nd, feature selecting
According to logistic regression feature, parameter matrix θ represents the weight of each input quantity, and its absolute value is bigger to represent the input quantity
It is more important, it is bigger on final classification results influence.θ matrixes are m*k matrixes, i.e., an often capable corresponding input quantity, each column correspondence
One kind classification.Therefore, this algorithm sues for peace the absolute value that θ often goes, and several of selection maximum are special as the stabilization of power network
Levy and exported.
As shown in figure 1, the specific steps of methods described include：
(1) determine the sample input quantity and its classification quantity of logistic regression training pattern, and define hyper parameter set；
The sample input quantity for determining logistic regression training pattern includes：Off-line simulation is carried out to network system failure,
Always generated electricity with the power of the assembling unit, line power, plant stand and plant stand total load is input, with the critical clearing time of discretization as defeated
Go out, as stabilization of power grids level index, so that it is determined that the sample input quantity of logistic regression training pattern.
Critical clearing time：Three phase short circuit fault is most typical failure mode in power system, and three-phase shortcircuit is critical
Mute time (CCT, critical clearing time) refers to after power network occurs three phase short circuit fault, it is ensured that system stabilization
Maximum fault clearing time.Critical clearing time represents system stabilization and unstable border, can be used to characterize power train
There is the degree of stability for specifying failure in system, critical clearing time is bigger, represents that the short trouble is smaller to systematic influence, and system is just
It is more stable.If critical clearing time is less than normal operating time of protection, illustrate that the failure can cause system unstability, that is, be
There is potential safety hazard in system.
The method for solving critical clearing time mainly includes time-domain-simulation method and direct method, and the former is calculated using time-domain-simulation
Critical clearing time is accurately solved, it is as a result the most accurate and reliable, but calculate time-consuming more long, equivalent to transient state several times
Stability Calculation, it is difficult to adapt to the requirement of on-line analysis；The advantage of the latter is fast calculating speed, using the teaching of the invention it is possible to provide stability index, but essence
Degree is relatively low.
Because critical clearing time is a floating number index, and logistic regression method is for solving many classification problems
, it is therefore desirable to critical clearing time is carried out discretization.Due to temporarily steady calculating online at present all using 0.01 second as emulation
Step-length, therefore critical clearing time can be carried out by the method for retaining two-decimal naturally discrete.It is important through emulation testing
500kV alternating current circuits critical clearing time all within 0.10-1.00, be then considered highly stable if greater than 1.0 seconds
State, the gear of 1.0 seconds can be included into.Therefore, the prediction of critical clearing time can be exchanged into many points no more than 100 types
Class problem.
Carrying out classification to the sample input quantity includes：Logistic regression is a kind of supervised learning side for solving classification problem
Method, it is that a sigmoid function is increased on the basis of linear regression, linear regression result is mapped to the area of [0,1]
Interior, definition predetermined threshold is line of demarcation, and it is classified, will be Numerical regression problem by the Sigmoid functions of following formula
Be converted to classification problem.In the way of cost function optimizes, many classification problems of CCT suitable for on-line analysis are solved：
The classification quantity of the logistic regression training pattern is determined by following formula：
In formula, hθX () represents the probability that 1 is taken for training sample x classification results, θ is logistic regression parameter matrix.
Simple two classification problems definable cost function J (θ) is shown below.Asked by minimizing cost function
Optimal θ parameters are taken, complete Logic Regression Models are obtained.
In formula, x is sample input quantity；Y is sample classification label；M is total sample number.
For many classification problems, transformed using softmax function pairs formula (2), obtained new cost function：
In formula, x is sample input quantity；Y is sample classification label；M is total sample number；K is classification quantity.
Define hyper parameter set λ={ λ1, λ2..., λn, λ ∈ [0,1], n are the number of hyper parameter.
(2) cost function of the definition comprising LASSO penalty terms, LASSO penalty terms are included using described in iterative method renewal
Logistic regression parameter matrix θ in cost function；
Lasso (Least Absolute Shrinkage and Selection Operator) method is in cost function
It is middle increase by one penalty function of the norm of parameter θ matrix 1 so that the coefficient of the secondary input quantity in part be compressed to zero or close to
Zero, so as to obtain a model for more refining.By refining, minor parameter is greatly compressed model, and major parameter is more
It is prominent, so also it is achieved that the automatic identifying function of principal character.
Cost function of the definition comprising LASSO penalty terms；
In formula, J (x) is the cost function of x, and x is sample input quantity, and y is sample classification label, λiIt is surpassing for adjustment ratio
Parameter, for obtaining balance between the penalty value of actual cost function and logistic regression parameter matrix θ, m is total sample number, k
It is classification quantity.
Updating the logistic regression parameter matrix θ in the cost function comprising LASSO penalty terms using iterative method includes：
A. initialization logic regression parameter matrix θ；
B. θ is substituted into formula (4) and solves cost function, and utilize the gradient descent method of formula (5) to obtain ladder of the cost function to θ
Degree；
C. θ is updated according to Grad and default learning rate iteration；
D. judge that cost function is not reduced within the unit interval or iterations reaches the upper limit, if so, then exporting current
θ, if otherwise return to step b.
Initialization logic regression parameter matrix θ includes：By m sample input quantity x={ x1,x2,…,xm, and its it is corresponding
Tag along sort y={ y1,y2,…,ym, the characteristic vector of k dimension is constituted, by patrolling for m k dimensional feature vectors one m*k of composition
Collect regression parameter matrix θ.
(3) value in proportion under index coordinate system, error rate deviation is less than threshold value in choosing the hyper parameter set
The logistic regression training pattern that maximum hyper parameter is generated is used as optimal models；
(4) it is special using the maximum of the logistic regression parameter matrix θ row vector absolute value sums of the model as the stabilization of power grids
Levy.
Based on same inventive concept, the present invention also proposes a kind of power system stability Automatic signature extraction device, the device
Including：
Setup module, for determining the sample input quantity and its classification quantity of logistic regression training pattern, and defines super ginseng
Manifold is closed；
Update module, for defining the cost function comprising LASSO penalty terms, includes using described in iterative method renewal
Logistic regression parameter matrix θ in the cost function of LASSO penalty terms；
Acquisition module, for the value in proportion under index coordinate system, chooses error rate deviation in the hyper parameter set
The logistic regression training pattern generated less than the maximum hyper parameter of threshold value is used as optimal models；
Definition module, for by the maximum of the logistic regression parameter matrix θ row vector absolute value sums of the optimal models
As stabilization of power grids feature.
Embodiment：
By the State Grid Corporation of China year 1-10 months in line computation data based on, verify the validity of this problem method.When
Month North China-Central China is in networking operation state, therefore state's straightening tune and North China, all 220kV in Central China are included in online data
Grid equipment above.Electric network state amount and statistic amount to 28201, as shown in the table；Effective sample number (section number) is
29254.
The electric network state amount of table 1 and statistic list
| Device type | Quantity of state | Quantity |
| More than 220kV AC lines | Put into operation state, active power | 6644×2 |
| Whole units | Put into operation state, active power, reactive power, set end voltage | 1435×4 |
| AC line | Dc power | 14 |
| Range statistics amount | Region always generates electricity, total load, average voltage | 15×3 |
| Plant stand | Plant stand gross capability, put into operation unit number, total load, ceiling voltage | 3038×3 |
Investigating failure includes that the line of Sichuan mountain peach one, North China Huangs shore one line, state adjust gorges Pueraria lobota I line, Central China Fan Bai II line, state
Fishing suitable line, state is adjusted to adjust Ge Gang lines, Central China boards I loop lines long, Central China interwined dragon I line, Central China Ai He I loop line, the gorgeous board I of Central China to return
Line, altogether 10.
Receive the limitation of sample number, it is impossible to using whole grid variables as input quantity, be otherwise easy to over-fitting occur
Situation, one by one can only learn every class variable as input quantity.
A) model learning
This 7 different hyper parameters of λ=0,0.1,0.01,0.001,0.0001,0.00001,0.000001 are selected respectively
To carry out model learning, by taking Ge Gang lines as an example：Input quantity is the power of the assembling unit, totally 1325 variables；Output quantity is Ge Gang line CCT,
In 0.19-0.36,18 kinds of possibility, therefore parameter matrix altogether are 1325*18, altogether 23850 parameters in interval.Learning outcome is such as
Shown in following table.
The Ge Gang line model learning outcomes of table 2
| λ | Error rate | Quantity of the absolute value less than 0.01 in θ matrixes | Compression ratio |
| 0 (benchmark) | 13.79% | 810 | 3.40% |
| 0.000001 | 14.00% | 6349 | 26.62% |
| 0.00001 | 14.15% | 15740 | 66.00% |
| 0.0001 | 18.06% | 20330 | 85.24% |
| 0.01 | 33.06% | 20847 | 87.41% |
| 0.1 | 57.47% | 16523 | 69.28% |
| 1 | 74.41% | 2264 | 9.49% |
From result, the error rate of λ=0.1 and λ=1 correspondence model is substantially excessive, illustrates that model can not reflect electricity
Net stability characteristic (quality), it is therefore desirable to ignore；Remaining result, error rate is raised with the increase of λ, θ matrix small parameter quantity and compression
Than also becoming big with the increase of λ, as a result meet the expection of algorithm design.Wherein, λ is worked as>After 0.0001, the lifting of error rate is obvious
Accelerate, and compression ratio change tends towards stability, therefore Logic Regression Models and parameter when should select λ=0.0001 are used as optimal mould
Type.
It is respectively adopted that the power of the assembling unit, line power, plant stand always generate electricity and plant stand total load carries out model as input quantity
Study, model learning result during λ=0 is as shown in the table.
The error rate contrast of the different input quantities of table 4
| Forecast failure | The power of the assembling unit | Line power | Plant stand always generates electricity | Plant stand total load |
| Ge Gang lines | 13.79% | 19.35% | 15.35% | 15.21% |
| The line of yellow shore one | 61.97% | 79.29% | 69.03% | 68.97% |
| Gorge Pueraria lobota I lines | 13.74% | 18.89% | 16.24% | 15.59% |
| The suitable line of fishing | 12.88% | 18.38% | 15.03% | 16.32% |
From result, the error rate of the lines of Huang Bin mono- is very high all the time, reason be the lines of Huang Bin mono- CCT results span compared with
Greatly, it is 0.20-0.71, about having more than 50 kinds of classification may.Contrast other 3 failures：18 kinds of Ge Gang lines (0.19-0.36), gorge
11 kinds of Pueraria lobota I lines (0.19-0.29), 31 kinds of suitable line of fishing (0.17-0.47), the decentralization of the line CCT results of Huang Bin mono- is substantially much larger.
Such case causes that every kind of classification samples number is smaller, and model learning is not abundant enough, therefore error rate is larger.Remaining 3 failure
Result is substantially similar, and optimal as the result of input quantity using the power of the assembling unit, and line power is worst as the result of input quantity,
Plant stand always generates electricity and total load falls between.
B) feature selecting
As a example by the line of Reng Yige hilllocks, using the power of the assembling unit as input quantity, optimal models during with λ=0.0001 selects stabilization
Feature, the absolute value sum often gone using parameter matrix θ is as a result as shown in the table as foundation is selected.
The Ge Gang line stabilization feature identification results of table 4
| Sequence number | Name variable (unit is active) | θ matrixes correspondence row absolute value sum |
| 1. | State adjusts Gorges Right Banks factory/20kV.19# units | 40.48 |
| 2. | State adjusts Gorges Right Banks factory/20kV.16# units | 39.61 |
| 3. | State adjusts Gorges Right Banks factory/20kV.20# units | 22.34 |
| 4. | State adjusts Gorges Right Banks factory/20kV.15# units | 22.09 |
| 5. | State adjusts Gorges Right Banks factory/20kV.24# units | 19.81 |
| 6. | State adjusts Gorges Right Banks factory/20kV.26# units | 19.19 |
| 7. | State adjusts Gorges Right Banks factory/20kV.25# units | 18.99 |
| 8. | State adjusts Three Gorges left banks factory/20kV.7# units | 18.68 |
| 9. | State adjusts Gorges Right Banks factory/20kV.23# units | 19.33 |
| 10. | The weeks gulf/20kV.#2 machine in Central China | 17.35 |
| 11. | State adjusts Gorges Right Banks factory/20kV.21# units | 17.22 |
| 12. | State adjusts Three Gorges left banks factory/20kV.2# units | 17.09 |
| 13. | State adjusts Gorges Right Banks factory/20kV.22# units | 16.70 |
| 14. | State adjusts Three Gorges left banks factory/20kV.10# units | 16.57 |
| 15. | Sichuan Ludings factory/15.75kV.1# units | 16.33 |
| 16. | Central China Baoqing/27kV.#2 machines | 15.99 |
| 17. | State adjusts Gorges Right Banks factory/20kV.18# units | 15.51 |
| 18. | State adjusts Gorges Right Banks factory/20kV.17# units | 15.49 |
| 19. | Hubei E Zhouchang/22kV.#4 machines | 15.25 |
| 20. | State adjusts Three Gorges left banks factory/20kV.13# units | 14.30 |
Because Ge Gang lines are near Three Gorges Hydropower Plant, and in Ge Gang line CCT calculating process, unstability form is more with Three Gorges unit
Based on the generator rotor angle unstability in North China unit (such as Shandong, Hebei), therefore unit Zhong Yuge hilllock line stabilization degree correlation maximums
Necessarily Three Gorges unit, this with upper table invariant feature choose result it is identical, be also consistent with the universal law of stability analysis.
It should be understood by those skilled in the art that, embodiments herein can be provided as method, system or computer program
Product.Therefore, the application can be using the reality in terms of complete hardware embodiment, complete software embodiment or combination software and hardware
Apply the form of example.And, the application can be used and wherein include the computer of computer usable program code at one or more
The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) is produced
The form of product.
The application is the flow with reference to method, equipment (system) and computer program product according to the embodiment of the present application
Figure and/or block diagram are described.It should be understood that every first-class during flow chart and/or block diagram can be realized by computer program instructions
The combination of flow and/or square frame in journey and/or square frame and flow chart and/or block diagram.These computer programs can be provided
The processor of all-purpose computer, special-purpose computer, Embedded Processor or other programmable data processing devices is instructed to produce
A raw machine so that produced for reality by the instruction of computer or the computing device of other programmable data processing devices
The device of the function of being specified in present one flow of flow chart or multiple one square frame of flow and/or block diagram or multiple square frames.
These computer program instructions may be alternatively stored in can guide computer or other programmable data processing devices with spy
In determining the computer-readable memory that mode works so that instruction of the storage in the computer-readable memory is produced and include finger
Make the manufacture of device, the command device realize in one flow of flow chart or multiple one square frame of flow and/or block diagram or
The function of being specified in multiple square frames.
These computer program instructions can be also loaded into computer or other programmable data processing devices so that in meter
Series of operation steps is performed on calculation machine or other programmable devices to produce computer implemented treatment, so as in computer or
The instruction performed on other programmable devices is provided for realizing in one flow of flow chart or multiple flows and/or block diagram one
The step of function of being specified in individual square frame or multiple square frames.
Finally it should be noted that：Above example is only used to illustrate the technical scheme of the application rather than to its protection domain
Limitation, although being described in detail to the application with reference to above-described embodiment, those of ordinary skill in the art should
Understand：Those skilled in the art read still can be to applying after the application specific embodiment carry out a variety of changes, modification or
Person's equivalent, these changes, modification or equivalent, it is within the pending right of its application.
Claims (8)
1. a kind of power system stability characteristic automatic extraction method, it is characterised in that methods described includes：
Determine the sample input quantity and its classification quantity of logistic regression training pattern, and define hyper parameter set；
Cost function of the definition comprising LASSO penalty terms, the cost function comprising LASSO penalty terms is updated using iterative method
In logistic regression parameter matrix θ；
The value in proportion under index coordinate system, error rate deviation is less than the super ginseng of the maximum of threshold value in choosing the hyper parameter set
Several generated logistic regression training patterns are used as optimal models；
Using the maximum of the logistic regression parameter matrix θ row vector absolute value sums of the optimal models as stabilization of power grids feature.
2. the method for claim 1, it is characterised in that the sample input quantity bag of the determination logistic regression training pattern
Include：Off-line simulation is carried out to network system failure, is always generated electricity with the power of the assembling unit, line power, plant stand and plant stand total load is as defeated
Enter, be output with the critical clearing time of discretization, using critical clearing time as stabilization of power grids level index, so that it is determined that patrolling
Collect the sample input quantity of regression training model.
3. the method for claim 1, it is characterised in that carrying out classification to the sample input quantity includes：Logic is returned
Return the logistic regression parameter matrix θ of training pattern to be mapped in [0,1] interval range, self-defined line of demarcation, will be above respectively or
Logistic regression parameter less than the line of demarcation is classified；
The classification quantity of the logistic regression training pattern is determined by following formula：
In formula, hθX () represents the probability that 1 is taken for training sample x classification results, x is sample input quantity, and θ is logistic regression parameter
Matrix.
4. the method for claim 1, it is characterised in that the definition hyper parameter set step is to define the super ginseng
Manifold closes λ={ λ1, λ2..., λn, λ ∈ [0,1], n are the number of hyper parameter.
5. the method for claim 1, it is characterised in that cost function of the definition comprising LASSO penalty terms includes：
Increase a penalty function for the norm of parameter θ matrix 1 in cost function, obtain cost function of the following formula comprising LASSO penalty terms：
In formula, J (x) is the cost function of x, and x is sample input quantity, and y is sample classification label, λiTo adjust the hyper parameter of ratio,
For obtaining balance between the penalty value of cost function and logistic regression parameter matrix θ, m is total sample number, and k is classification number
Amount.
6. the method for claim 1, it is characterised in that the utilization iterative method updates and described includes LASSO penalty terms
Cost function in logistic regression parameter matrix θ include：
A. initialization logic regression parameter matrix θ；
B. θ is substituted into formula (2) and solves cost function, and gradient of the cost function to θ is obtained using gradient descent method；
C. θ is updated according to Grad and default learning rate iteration；
D. judge that cost function is not reduced within the unit interval or iterations reaches the upper limit, if so, current θ is then exported, if
Otherwise return to step b.
7. method as claimed in claim 6, it is characterised in that the initialization logic regression parameter matrix θ includes：By m
Sample input quantity x={ x1,x2,…,xm, and its corresponding tag along sort y={ y1,y2,…,ym, constitute a feature for k dimensions
Vector, a logistic regression parameter matrix θ of m*k is constituted by m k dimensional feature vector.
8. a kind of power system stability Automatic signature extraction device, described device includes：
Setup module, for determining the sample input quantity and its classification quantity of logistic regression training pattern, and defines hyper parameter collection
Close；
Update module, for defining the cost function comprising LASSO penalty terms, punishes using described in iterative method renewal comprising LASSO
The logistic regression parameter matrix θ penalized in the cost function of item；
Acquisition module, for the value in proportion under index coordinate system, error rate deviation is less than in choosing the hyper parameter set
The logistic regression training pattern that the maximum hyper parameter of threshold value is generated is used as optimal models；
Definition module, for using the maximum of the logistic regression parameter matrix θ row vector absolute value sums of the optimal models as
Stabilization of power grids feature.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710078550.9A CN106786560B (en) | 2017-02-14 | 2017-02-14 | Method and device for automatically extracting stability characteristics of power system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710078550.9A CN106786560B (en) | 2017-02-14 | 2017-02-14 | Method and device for automatically extracting stability characteristics of power system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106786560A true CN106786560A (en) | 2017-05-31 |
| CN106786560B CN106786560B (en) | 2021-04-30 |
Family
ID=58955543
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710078550.9A Active CN106786560B (en) | 2017-02-14 | 2017-02-14 | Method and device for automatically extracting stability characteristics of power system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106786560B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107544907A (en) * | 2017-09-07 | 2018-01-05 | 郑州云海信息技术有限公司 | A kind of automated testing method and device |
| CN107947156A (en) * | 2017-11-24 | 2018-04-20 | 国网辽宁省电力有限公司 | Based on the electric network fault critical clearing time method of discrimination for improving Softmax recurrence |
| CN108564423A (en) * | 2017-12-28 | 2018-09-21 | 携程旅游网络技术（上海）有限公司 | Malice occupy-place recognition methods, system, equipment and the storage medium of ticketing service order |
| CN109003254A (en) * | 2018-05-28 | 2018-12-14 | 南方电网科学研究院有限责任公司 | Logic-based returns method for detecting ice coating, device, equipment, system and medium |
| CN109190713A (en) * | 2018-09-29 | 2019-01-11 | 王海燕 | The minimally invasive fast inspection technology of oophoroma based on serum mass spectrum adaptive sparse feature selecting |
| CN109765883A (en) * | 2019-03-04 | 2019-05-17 | 积成电子股份有限公司 | Distribution power automation terminal evaluation of running status and method for diagnosing faults |
| CN111448734A (en) * | 2017-12-21 | 2020-07-24 | 维斯塔斯风力***集团公司 | Improvements relating to current injection in wind power plants |
| CN111860940A (en) * | 2020-06-02 | 2020-10-30 | 山西潞安环保能源开发股份有限公司 | Coal mine distribution network power failure accident prediction method based on logistic regression algorithm |
| CN112183805A (en) * | 2019-12-23 | 2021-01-05 | 成都思晗科技股份有限公司 | Method for predicting state of online inspection result of power transmission line |
| CN112580252A (en) * | 2020-11-26 | 2021-03-30 | 北京航天智造科技发展有限公司 | Rotor drop-out fault diagnosis and analysis method and device |
| CN112613157A (en) * | 2020-11-26 | 2021-04-06 | 北京航天智造科技发展有限公司 | Rotor fault analysis method and device |
| CN113658676A (en) * | 2021-08-26 | 2021-11-16 | 北京践行者教育科技研究院 | Internet-based cognitive behavior mental health management method and system |
| CN113660210A (en) * | 2021-07-20 | 2021-11-16 | 北京天融信网络安全技术有限公司 | Malicious TLS encrypted traffic detection model training method, detection method and terminal |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103544546A (en) * | 2013-10-31 | 2014-01-29 | 浙江大学 | Method for online prediction of quiescent voltage stability limit of electric system |
| CN103559556A (en) * | 2013-10-31 | 2014-02-05 | 浙江大学 | Method for forecasting load capacity limit of electric power system online |
| CN105406476A (en) * | 2015-12-31 | 2016-03-16 | 国家电网公司 | Historical data-based power system stability fast judging method |
| CN105512799A (en) * | 2015-11-26 | 2016-04-20 | 中国电力科学研究院 | Mass online historical data-based power system transient stability evaluation method |
| WO2016143021A1 (en) * | 2015-03-09 | 2016-09-15 | 株式会社日立製作所 | Power system stabilization system |
| CN105975443A (en) * | 2016-05-04 | 2016-09-28 | 西南大学 | Lasso-based anomaly detection method and system |
-
2017
- 2017-02-14 CN CN201710078550.9A patent/CN106786560B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103544546A (en) * | 2013-10-31 | 2014-01-29 | 浙江大学 | Method for online prediction of quiescent voltage stability limit of electric system |
| CN103559556A (en) * | 2013-10-31 | 2014-02-05 | 浙江大学 | Method for forecasting load capacity limit of electric power system online |
| WO2016143021A1 (en) * | 2015-03-09 | 2016-09-15 | 株式会社日立製作所 | Power system stabilization system |
| CN105512799A (en) * | 2015-11-26 | 2016-04-20 | 中国电力科学研究院 | Mass online historical data-based power system transient stability evaluation method |
| CN105406476A (en) * | 2015-12-31 | 2016-03-16 | 国家电网公司 | Historical data-based power system stability fast judging method |
| CN105975443A (en) * | 2016-05-04 | 2016-09-28 | 西南大学 | Lasso-based anomaly detection method and system |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107544907A (en) * | 2017-09-07 | 2018-01-05 | 郑州云海信息技术有限公司 | A kind of automated testing method and device |
| CN107947156B (en) * | 2017-11-24 | 2021-02-05 | 国网辽宁省电力有限公司 | Power grid fault critical clearing time discrimination method based on improved Softmax regression |
| CN107947156A (en) * | 2017-11-24 | 2018-04-20 | 国网辽宁省电力有限公司 | Based on the electric network fault critical clearing time method of discrimination for improving Softmax recurrence |
| CN111448734B (en) * | 2017-12-21 | 2023-05-26 | 维斯塔斯风力***集团公司 | Method for controlling a renewable energy generator and renewable energy power plant |
| CN111448734A (en) * | 2017-12-21 | 2020-07-24 | 维斯塔斯风力***集团公司 | Improvements relating to current injection in wind power plants |
| US11855458B2 (en) | 2017-12-21 | 2023-12-26 | Vestas Wind Systems A/S | Current injection in wind power plants |
| CN108564423A (en) * | 2017-12-28 | 2018-09-21 | 携程旅游网络技术（上海）有限公司 | Malice occupy-place recognition methods, system, equipment and the storage medium of ticketing service order |
| CN109003254A (en) * | 2018-05-28 | 2018-12-14 | 南方电网科学研究院有限责任公司 | Logic-based returns method for detecting ice coating, device, equipment, system and medium |
| CN109190713A (en) * | 2018-09-29 | 2019-01-11 | 王海燕 | The minimally invasive fast inspection technology of oophoroma based on serum mass spectrum adaptive sparse feature selecting |
| CN109765883A (en) * | 2019-03-04 | 2019-05-17 | 积成电子股份有限公司 | Distribution power automation terminal evaluation of running status and method for diagnosing faults |
| CN109765883B (en) * | 2019-03-04 | 2020-09-22 | 积成电子股份有限公司 | Power distribution automation terminal operation state evaluation and fault diagnosis method |
| CN112183805A (en) * | 2019-12-23 | 2021-01-05 | 成都思晗科技股份有限公司 | Method for predicting state of online inspection result of power transmission line |
| CN112183805B (en) * | 2019-12-23 | 2023-10-24 | 成都思晗科技股份有限公司 | Prediction method for online inspection result state of power transmission line |
| CN111860940A (en) * | 2020-06-02 | 2020-10-30 | 山西潞安环保能源开发股份有限公司 | Coal mine distribution network power failure accident prediction method based on logistic regression algorithm |
| CN111860940B (en) * | 2020-06-02 | 2024-03-19 | 湖北凌云电力建设有限公司 | Coal mine power distribution network power failure accident prediction method based on logistic regression algorithm |
| CN112613157A (en) * | 2020-11-26 | 2021-04-06 | 北京航天智造科技发展有限公司 | Rotor fault analysis method and device |
| CN112580252A (en) * | 2020-11-26 | 2021-03-30 | 北京航天智造科技发展有限公司 | Rotor drop-out fault diagnosis and analysis method and device |
| CN113660210A (en) * | 2021-07-20 | 2021-11-16 | 北京天融信网络安全技术有限公司 | Malicious TLS encrypted traffic detection model training method, detection method and terminal |
| CN113658676B (en) * | 2021-08-26 | 2023-09-01 | 北京践行者教育科技研究院 | Internet-based cognitive behavioral and psychological health management method and system |
| CN113658676A (en) * | 2021-08-26 | 2021-11-16 | 北京践行者教育科技研究院 | Internet-based cognitive behavior mental health management method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106786560B (en) | 2021-04-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106786560A (en) | A kind of power system stability characteristic automatic extraction method and device | |
| CN111179121B (en) | Power grid emergency control method based on expert system and deep reverse reinforcement learning | |
| CN102074955B (en) | Method based on knowledge discovery technology for stability assessment and control of electric system | |
| CN110417011B (en) | Online dynamic security assessment method based on mutual information and iterative random forest | |
| CN103488869A (en) | Wind power generation short-term load forecast method of least squares support vector machine | |
| CN106951695A (en) | Plant equipment remaining life computational methods and system under multi-state | |
| CN108053128A (en) | A kind of Power Network Transient Stability fast evaluation method based on ELM and TF | |
| CN108776855A (en) | A kind of smart machine health status evaluation method and system | |
| CN111523778A (en) | Power grid operation safety assessment method based on particle swarm algorithm and gradient lifting tree | |
| CN114006370B (en) | Power system transient stability analysis and evaluation method and system | |
| CN109214708A (en) | Study of Risk Evaluation Analysis for Power System method based on cross entropy theoretical optimization support vector machines | |
| CN109784692A (en) | A kind of fast and safely constraint economic load dispatching method based on deep learning | |
| CN112017070A (en) | Method and system for evaluating transient stability of power system based on data enhancement | |
| CN107103167A (en) | For the DEH speed-regulating functions diagnostic method and system of removal of load operating mode | |
| CN112821424A (en) | Power system frequency response analysis method based on data-model fusion drive | |
| CN111814284A (en) | On-line voltage stability evaluation method based on correlation detection and improved random forest | |
| CN115526258A (en) | Power system transient stability evaluation method based on Spearman correlation coefficient feature extraction | |
| CN108805419B (en) | Power grid node importance calculation method based on network embedding and support vector regression | |
| CN105701265B (en) | A kind of double-fed fan motor unit modeling method and device | |
| He et al. | Power system frequency situation prediction method based on transfer learning | |
| CN116541748A (en) | Power grid fault diagnosis method and system based on artificial intelligence technology | |
| CN108183481A (en) | One kind quickly sentences steady method and system based on deep learning power grid | |
| CN115271070A (en) | Voltage stability margin evaluation method, equipment and medium based on improved Relief-F algorithm | |
| Zhang et al. | CNN‐LSTM based power grid voltage stability emergency control coordination strategy | |
| CN111553040B (en) | Power grid topology analysis high-performance computing method and device based on GPU acceleration |
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 |