CN106485594A - A kind of main distribution integration incident response decision method - Google Patents
A kind of main distribution integration incident response decision method Download PDFInfo
- Publication number
- CN106485594A CN106485594A CN201610305304.8A CN201610305304A CN106485594A CN 106485594 A CN106485594 A CN 106485594A CN 201610305304 A CN201610305304 A CN 201610305304A CN 106485594 A CN106485594 A CN 106485594A
- Authority
- CN
- China
- Prior art keywords
- main distribution
- fault
- attribute
- node
- decision
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Abstract
A kind of main distribution integration incident response decision method, step one, sets up main distribution integration and runs scene collection and fault set；Step 2, global coordination control strategy knowledge base is set up based on main distribution integration high-speed simulation result；Step 3, using feature extraction, decision tree generation technique generate emulation scene decision ruless storehouse；Step 4, after main distribution running status cognition technology and matching process of making decisions on one's own realize fault main distribution make decisions on one's own coupling response.The method that the present invention sets up main distribution integration incident response decision-making, make anticipation before major network and distribution actually occur fault, there is provided Control Measure to major network operations staff and distribution operations staff, or be quickly given urgent after fault actually occurs and recover control strategy.Improve the overall operation stability of electrical network, safety, reliability and economy.
Description
Technical field
The present invention relates to power system accident response decision-making method, more particularly to a kind of main distribution integration incident response
Decision method.
Background technology
In the complex accident Response Decision of bulk power grid, maximum difficulty is that the decision-making time is short and main distribution is integrated and imitates
The true analytical calculation time is long, and real-time status estimated result is relied on big.Based on this, this patent uses hypercube Latin sampling method
Generate main distribution integrative simulation and run scene collection, and according to probability of malfunction and value-at-risk, carry out screening life from primary fault collection
Become simulation analysis fault set；Input runs scene collection and fault set carries out main distribution integration high-speed simulation, that is, carry out main distribution
Global optimization obtains fault optimum recovery control strategy；Record running status and control strategy before and after electric network fault truly, set up
Global coordination plan knowledge storehouse；By the crucial spy to operation of power networks state in global coordination plan knowledge storehouse for the Feature Extraction Technology
Levy attribute to be extracted, set up emulation scene decision ruless storehouse；When the electrical network (major network or containing distributed power source in real time execution
Distribution) occur physical fault after, current operating conditions are obtained using state aware technology, so adopt characteristic matching search technique
Search from emulation scene decision ruless storehouse obtains corresponding global coordination control strategy, realizes fast quick-recovery and the sound of fault
Should.
Traditionally the accident analysiss Response Decision of major network and distribution is mainly isolated is carried out.The line causing after major network fault
Road transmission blocking problem or voltage problem to eliminate mainly by the controllable resources controlling major network and to recover, and does not consider the coordination of distribution
And support；And the service restoration strategy causing after feeder line fault of distribution network is also only counted and the power supply capacity of distribution, do not consider major network
Coordinate and support.This undoubtedly causes in many cases, control decision less economical, a lot of available resources idle plus profit
With in some instances it may even be possible to cause load cannot full recovery, power supply reliability reduce.Access power distribution network with a large amount of DG, power distribution network is certainly
The controllability of body and motility have large increase, there has also been large increase to the support ability of major network.Existing major network and distribution
Incident response decision-making is isolated the mode carrying out and is no longer adapted to.
At present, it is badly in need of a kind of method setting up main distribution integration incident response decision-making, actually occur in major network and distribution
Make anticipation before fault, provide Control Measure to major network operations staff and distribution operations staff, or in actual of fault
Quickly be given urgent after life and recover control strategy.
Content of the invention
It is an object of the invention to provide a kind of improve the overall operation stability of electrical network, safety, reliability and economy
Main distribution integration incident response decision method.
For achieving the above object, the technical scheme of present invention offer is：
A kind of main distribution integration incident response decision method, the method comprising the steps of：
Step one, set up main distribution integration and run scene collection and fault set；
Step 2, global coordination control strategy knowledge base is set up based on main distribution integration high-speed simulation result；
Step 3, using feature extraction, decision tree generation technique generate emulation scene decision ruless storehouse；
Step 4, main distribution after main distribution running status cognition technology and matching process of making decisions on one's own realize fault
Coupling of making decisions on one's own response.
Further, in step one, described operation scene collection, predictive power are generated by Latin hypercube simulation
With forecast error distribution, each input stochastic variable is sampled it is ensured that random distribution region can be sampled a little covers completely
Lid；Change putting in order of each stochastic variable sampled value, so that the dependency of the sampled value of mutually independent random variables is tended to
Little, put in order using Cholesky decomposition method,
Sampled value formula is
Wherein, XnmRepresent stochastic variable XnM-th sampled value, UmRepresent sampled value.
Further, in step one, by providing the comprehensive measurement of each fault rate and seriousness, screened with this
Fault, generates described fault set, and the mathematic(al) representation of comprehensive measurement is：
Wherein, XfFor current system running status；EiFor i-th forecast failure；Pr(Ei) it is EiThe probability occurring；Sev(Ei)
For EiThe order of severity of system loss after generation.
Further, in step one, failure probability model formula is：
Pr(Fi)=1-exp (- λits),
Wherein, FiBreak down for i-th line road；Pr(Fi) probability that breaks down for i-th line road；λiFor i-th line
The fault rate on road；
After system jam, the overload penalty values computing formula of branch road is：
Wherein, ωLiRepresent overload penalty values, LiThe ratio of the actual fed power for branch road i and power allowances；L0For setting
Threshold value, if LiLess than L0, dispatcher thinks that this branch road does not have overload risk；
Define branch road i overload order of severity computing formula be：
Wherein, a, c are positive number, ωLiRepresent overload penalty values；
Voltage out-of-limit penalty values ω of system jam posterior nodal point iViComputing formula is：
Wherein, UiVoltage magnitude for node i；Uimax、UiminIt is respectively the upper and lower limit of the voltage magnitude of node i；
Define node i voltage out-of-limit order of severity computing formula be：
Wherein, a, c are positive number, ωviRepresent voltage out-of-limit penalty values；
By Contingency screening and ranking, some faults coming in ranking results above are generated described fault set.
Further, in step 2, described knowledge base is by the running status before electric network fault, the operation after electric network fault
State, fault is optimum to recover control strategy composition, carries out main distribution integration by input operation scene collection and fault set quick
Emulation, carries out main distribution global optimization and obtains fault optimum recovery control strategy, before and after the fault that record electrical network occurs and fault
Running status, control strategy, set up described global coordination plan knowledge storehouse.
Further, the step setting up knowledge base is as follows：
Step 2a master, distribution intelligent body obtain current time electrical network section information, by main distribution integration state estimation
Set up respective trend initial state；
Step 2b main distribution intelligent body collects weather information data in respective electrical network, passes through in conjunction with electrical network typical operation modes
Latin Hypercube Sampling technology obtains operation of power networks scene collection.Obtain grid simulation fault set；
For in failure collection, each treats simulated fault to step 2c, controls optimization to ask using main distribution integration fault recovery
The solution of topic, obtains main distribution coordination control strategy；
Electric network state after electric network state before fault, fault and control strategy write global coordination control strategy are known by step 2d
Know storehouse, original storehouse is updated.
Further, in step 3, described emulation scene decision ruless storehouse is to overall situation association by Feature Selection
The key feature attribute adjusting operation of power networks state in plan knowledge storehouse is extracted, and the characteristic attribute selected is belonged to as input
Property, automatically generate decision tree, extract finely rule from decision tree, set up emulation scene decision ruless storehouse.
Further, separated based on attribute and carry out feature selection, higher-dimension combinatorial problem is reduced to low-dimensional combinatorial problem, attribute
Between the computing formula of degree of association be：
Wherein, H (X), H (Y) are respectively the entropy of attribute X and Y；I(X；Y) it is mutual information between attribute X and Y；
Select characteristic attribute formula be：
Wherein, k is the quantity of attributive classification, ΩiIt is expressed as the i-th class.mi、WithIt is respectively ΩiCurrent offer
Characteristic attribute number, characteristic attribute set, provide quantity of information；F is characterized selection result property set；M is that the feature in F belongs to
Property number；C is objective attribute target attribute.
Further, described decision Tree algorithms are to detect all of attribute, select the maximum attribute of information gain-ratio to produce
Decision tree nodes, set up branch by the different values of this attribute, then set up decision-making to subset recursive call the method for each branch
The branch of tree node, till all subsets only comprise same category of data.
Further, in step 4, described main distribution running status sensing method, in main distribution range of operation, cognitive,
Understand the running status of each equipment and operational factor in electrical network real time execution, all information identified and convert based on distribution
Real-time running state amount, and screen and eliminate wherein wrong, invalid information, obtained by analyzing external data and internal data
Obtain running status and parameter after electrical network is disturbed, by being calculated the characteristic quantity information of coupling of making decisions on one's own.
Further, in step 4, accelerate the decision-making coupling of rule base using Rete algorithm, build Rete matching network
Step include：
Step 4a creates root node, forms the entrance of matching network；
Step 4b extracts new rule from scene decision ruless storehouse, extracts each state variable from regular condition
The condition meeting：
I) check in this regular condition and whether occur in that new state variable, if new state variable, add one
Individual type node；
Ii) check condition that each state variable in this rule meets whether Already in α node, if there is
Then record the position of this α node, if it is not, the condition that this state variable is met adds as a new α node
To in network；
Iii) repeat step ii), until the condition that state variable all of in this rule is met all is disposed；
Iv) combine β node：First β node is formed by two α combination of nodes, and each β node thereafter is by last layer
β node and new α node are combined；
V) repeat step iv), until generating all β nodes；
Vi) this regular corresponding execution action is packaged into final node；
Step 4c repeat step 4b, until the strictly all rules in rule base is all disposed.
Further, in step 4, based on Rete algorithm carry out main Distribution Network Failure make decisions on one's own coupling with response when,
In addition to needing each state variable in main distribution is mated when first fit, carry out every a cycle afterwards
During coupling, only need to differentiate the currently main distribution state variable part big compared to last mechanical periodicity, according to the shape changing greatly
State variable corresponding α node continues directly to mate work from β network, eliminates the state variable repeated matching little to change
Process.
Using technique scheme, the present invention has the advantages that：
Main distribution integration incident response decision method proposed by the present invention only can start real when physical fault occurs
When state quick sensing and fault make decisions on one's own coupling and response function, you can automatically derive optimal control policy, solve
In the complex accident Response Decision of bulk power grid, the decision-making time is short, the main distribution integration parallel-distributed processing analytical calculation time
Long, and big problem is relied on to real-time status estimated result.
The method that the present invention sets up main distribution integration incident response decision-making, does before major network and distribution actually occur fault
Go out anticipation, provide Control Measure to major network operations staff and distribution operations staff, or quick after fault actually occurs
Be given urgent and recover control strategy.Improve the overall operation stability of electrical network, safety, reliability and economy.
Brief description
Fig. 1 is the present invention main distribution integration incident response decision flow diagram；
Fig. 2 is the pre-structure schematic diagram that breaks down in main distribution in the embodiment of the present invention；
Fig. 3 is the structural representation that in the embodiment of the present invention, main distribution occurs during Distribution Network Failure；
Fig. 4 is the optimal control policy figure that in the embodiment of the present invention, main distribution occurs after Distribution Network Failure；
Fig. 5 is structural representation when there is major network fault in main distribution in the embodiment of the present invention；
Fig. 6 is the optimal control policy figure after there is major network fault in main distribution in the embodiment of the present invention.
Specific embodiment
In order that the objects, technical solutions and advantages of the present invention become more apparent, below in conjunction with the accompanying drawings and embodiment, right
The present invention is further elaborated.It should be appreciated that structure chart described herein and specific embodiment are only in order to explain this
Invention, is not intended to limit the present invention.
Fig. 1 is the present invention main distribution integration incident response decision flow diagram, as shown in figure 1, the invention provides a kind of
Main distribution integration incident response decision method, method comprises the following steps：
Step one, set up main distribution integration and run scene collection and fault set；
Step 2, global coordination control strategy knowledge base is set up based on main distribution integration high-speed simulation result；
Step 3, using feature extraction, decision tree generation technique generate emulation scene decision ruless storehouse；
Step 4, main distribution after main distribution running status cognition technology and matching process of making decisions on one's own realize fault
Coupling of making decisions on one's own response.
Embodiment 1
In step：The operation scene collection design of main distribution integrative simulation is considered as distributed power source under meteorological condition
Uncertainty with load.The present invention generates a series of operation scenes by Latin Hypercube Sampling technical modelling.
Latin Hypercube Sampling (Latin hypercube sampling, LHS) be a kind of sampled value can effectively reflect with
The multiple-dimensional hierarchical method of sampling of machine variable overall distribution, this method ensure that all of sample area can be sampled a covering.
Latin hypercube is generally divided into sampling and sequence 2 steps：First each input stochastic variable is sampled it is ensured that random
Distributed areas can be sampled and a little be completely covered；Then change putting in order of each stochastic variable sampled value, make separate
The dependency of the sampled value of stochastic variable tends to minimum.The present invention adopts Latin hypercube, each according to wind-powered electricity generation, photovoltaic etc.
Prediction under the conditions of typical meteorological is exerted oneself and forecast error distribution, the Run-time scenario that simulation generates wind-powered electricity generation, photovoltaic is exerted oneself.
(1) Interval Sampling
Assume there is N number of independent random variable X obeying certain probability distribution1, X2..., XN, wherein XnFor wherein arbitrarily with
Machine variable, cumulative probability function is represented by：
Yn=Fn(Xn) n=1,2 ..., N (1)
The concrete method of sampling of Latin Hypercube Sampling is：M is taken to be sampling scale, due to adding up generally of each stochastic variable
Rate function YnIt is continuous monotonically increasing function, its corresponding valued space [0,1] is averagely divided into M not overlapping interval, that is,
[0,1/M], [1/M, 2/M] ..., [(M-1)/M, 1], randomly choose one as sampled value from each equidistant interval,
Each sampled value is represented by：
Wherein sampled value UmSpan be：
Each interval can only generate a sampled value, not repeated sampling at random simultaneously.
Obtain M interval YnAfter stochastical sampling value, using inverse functionX can be calculatednSampled value：
In formula (4)：XnmFor stochastic variable XnM-th sampled value.
To stochastic variable XnAfter sampling terminates, obtain M sampled value, arrange the line n for sampling matrix.When K variable is complete
After the completion of portion's sampling, sampled value constitutes initial samples matrix Xs, exponent number is N × M.
(2) sort
Forming initial samples matrix XsAfterwards, need with reference to dependency control algolithm, it to be ranked up, by changing
The order of each stochastic variable sampled value is reducing the dependency between them.The present invention adopts Cholesky decomposition method to sampling square
Battle array rearrangement, is broadly divided into following steps：
1) form sequential matrix L.The exponent number of L and XsIdentical, i.e. N × M rank, in L, every a line is by 1～M number random alignment
Formed, each element represents XsPosition in new matrix for the middle corresponding line sampled value；
2) renewal of sequential matrix L.Assume symmetric positive definite matrix ρLCorrelation series matrix for L, then pass through
What Cholesky decomposed obtains matrix D, and D is a nonsingular lower triangle real number matrix, meets following relation：
ρL=DDT(5)
Then structural matrix G：
G=D-1L (6)
G is N × M rank matrix, and its correlation matrix N × N rank unit matrix, in this explanation matrix G each row to
There is not dependency between amount.Row element each in matrix L is sorted from big to small according to corresponding element in G, obtains new order square
Battle array L；
3) obtain new XsMatrix.By XsMiddle element is replaced by the sequence valve in L, realizes XsThe fall of matrix correlation
Low.
For the ease of calculating, typically first arrange post-sampling when main distribution integrative simulation runs scene collection generating, that is, first
Genesis sequence matrix L, is then sampled again, forms Xs.For example, it is contemplated that when wind-powered electricity generation, photovoltaic etc. are exerted oneself, XsMiddle row element represents
The sampled value that under certain meteorological condition, wind-powered electricity generation, photovoltaic are exerted oneself.
The present invention is to remaining attribute such as generator output, payload in operation scene all using Latin hypercube
Analog sampling, will form multiple operation scenes after all sampled result permutation and combination, generate the operation of main distribution integrative simulation
Scene collection.
The fault set design of main distribution integrative simulation, because in actual electric network, number of faults is very huge, and not institute
Faulty all can cause serious consequence, if being all put in faulty for institute each operation scene and concentrate that to carry out main distribution integration quick
Emulation, then cross and amount of calculation can be led to excessive.So the present invention considers according to the real time execution situation of equipment, electrical network actual schedule need
Ask, the factor such as the impact to electrical network for the fault, provide the comprehensive measurement of each fault rate and seriousness, to be screened with this therefore
Barrier, generates the fault set of main distribution integrative simulation.The mathematic(al) representation of comprehensive measurement is：
In formula：XfFor current system running status；EiFor i-th forecast failure；Pr(Ei) it is EiThe probability occurring；Sev(Ei)
For EiThe order of severity of system loss after generation.
(1) failure probability model
For certain circuit, according to Poisson distribution, i-th line road is in time interval tsThe probability inside breaking down is：
Pr(Fi)=1-exp (- λits) (8)
In formula：FiBreak down for i-th line road；Pr(Fi) probability that breaks down for i-th line road；λiFor i-th line
The fault rate on road.
Substance fault is had
Computing formula in the case of multiple failure can be analogized.
(2) fault severity level evaluation model
The impact that line fault is stopped transport to power system is mainly shown as that branch road overload and node voltage are out-of-limit, and branch road transships
Cascading failure may be caused, node voltage is out-of-limit may to lead to collapse of voltage.For this reason, system failure loss is divided into branch road to transship
2 classes out-of-limit with node voltage.
Should be able to reflect relative between the out-of-limit Chengdu of running status amount and different faults in view of fault severity level function
The order of severity, the present invention to measure failure effect using utility function.
Overload penalty values ω of branch road i after system jamLiDefinition be：
In formula：LiThe ratio of the actual fed power for branch road i and power allowances；L0For set threshold value it is preferable that this
It is taken as 0.9 in bright.If LiLess than L0, dispatcher thinks that this branch road does not have overload risk.
Define branch road i the overload order of severity be：
In formula：A, c are positive number.
By formula (11) as can be seen that the overload order of severity is all higher than with regard to 1 order derivative and 2 order derivatives transshipping penalty values
0, this represents increases with breakdown loss, and the dissatisfaction of operations staff and its rate of change all increase, and have fully demonstrated fortune
The psychological bearing capability to failure effect for the administrative staff, meets power system practical operation situation.
In the same manner, voltage out-of-limit penalty values ω of system jam posterior nodal point iViDefinition be：
In formula：UiVoltage magnitude for node i；Uimax、UiminIt is respectively the upper and lower limit of the voltage magnitude of node i.If Ui?
In allowed band, operations staff thinks that this node does not have voltage limit risk.
Define node i the voltage out-of-limit order of severity be：
The present invention carries out Contingency screening and ranking by fault comprehensive metric, screens from thousand of forecast failures
Go out catastrophe failure and it is sorted, form main distribution integrative simulation fault set.By Contingency screening and ranking, by sequence knot
In fruit, front M fault generates main distribution integrative simulation fault set.
In view of the impact of meteorological factor, under dense fog, heavy rain, high temperature, frost, the specific bad weather condition of typhoon,
The probability of the device fails of some areas can significant raising, now fault higher for wherein probability of happening should be also added to
In fault set.
Embodiment 2
The step 2 of the present invention：
Knowledge base is made up of three partial contents：Running status before electric network fault, the running status after electric network fault, fault
Optimum recovery control strategy.Scene collection is run by input and fault set carries out main distribution integration high-speed simulation, that is, led
Distribution global optimization obtains fault optimum recovery control strategy, runs shape before and after the fault that true record electrical network occurs and fault
State, control strategy, set up global coordination plan knowledge storehouse.
Break down in distribution based on Fig. 2 pre-structure schematic diagram, as shown in Fig. 2 the master read in after a certain operation scene joins
There is the structural representation before Distribution Network Failure in net.Fig. 3 is structural representation during fault, according to Fig. 3, stores in knowledge base
Fault before running status include the exerting oneself of electromotor 1,2 in major network, the tap position of transformator 1, the switching of capacitor bank 1,2
Group number, in distribution interconnection switch 1～3 cut-off state, the operation of power networks state such as exert oneself of DG1～4.Read in a certain in fault set
There is line disconnection fault as shown in Figure 3, now the main distribution running status after record fault in Distribution Network Failure, such as distribution 1
Measure and be stored in knowledge base, afterwards the optimum recovery of fault is obtained by the main distribution integration high-speed simulation shown in Fig. 4 and control
Strategy.Fault for obtaining after this main Distribution Network Failure is optimum to recover control strategy, increases including electromotor in major network 2 and exerts oneself, joins
In net, interconnection switch 3 closes, DG1,4 increases the strategy such as exert oneself, and by these policy store in knowledge base, sets up global coordination control
Plan knowledge storehouse processed.There is structural representation during major network fault, if occurring as shown in Figure 5 in main distribution in distribution based on Fig. 5
Major network in electromotor 2 fault of stop, then by main distribution integration high-speed simulation, to can get fault optimum as shown in Figure 6 extensive
Multiple control strategy, increases including electromotor in major network 1 and exerts oneself, in distribution, DG1～4 increase and exert oneself.
Main distribution integration global coordination controls the establishment step of knowledge base to be：
(1) main, distribution intelligent body obtains current time electrical network section information, is built by main distribution integration state estimation
Stand respective trend initial state.
(2) main distribution intelligent body collects weather information data in respective electrical network, passes through to draw in conjunction with electrical network typical operation modes
Fourth hypercube Sampling techniques obtain operation of power networks scene collection, obtain grid simulation fault set simultaneously；
(3) in failure collection, each treats simulated fault, controls optimization problem using main distribution integration fault recovery
Solution, obtain main distribution coordination control strategy；
(4) electric network state after electric network state before fault, fault and control strategy are write the global coordination of present invention definition
Control strategy knowledge base, is updated to original storehouse.
Embodiment 3
In the step 3 of the present invention：
Main distribution integrative simulation scene decision ruless storehouse is to global coordination plan knowledge storehouse by Feature Selection
The key feature attribute of middle operation of power networks state is extracted, then using the characteristic attribute selected as inputting attribute, automatically
Generate decision tree, extract finely rule from decision tree, set up main distribution integrative simulation scene decision ruless storehouse.
Feature selection is the feasible method extracting characteristic attribute.Power system has decoupled active and reactive, layering and zoning is adjusted
The features such as spend.In consideration of it, the present invention is using feature selection approach (the mutual information based on attributive classification
feature selection based on classification of attributes,MIFS-C).This method will belong to first
Property is classified, the dependency very little between inhomogeneous attribute.Feature selection can be carried out respectively in each apoplexy due to endogenous wind, thus by higher-dimension
Combinatorial problem is reduced to low-dimensional combinatorial problem, reduces amount of calculation.The process of MIFS-C feature selection is as follows：
(1) calculate the degree of association between each attribute.Formula (14) defines the degree of association between attribute X and Y：
In formula：H (X), H (Y) are respectively the entropy of attribute X and Y；I(X；Y) it is mutual information between attribute X and Y.
In formula：Sx, SyFor X, the possible value set of Y.P (x) is the probability that X takes x, the implication of p (y) and p (x) class
Seemingly；P (x, y) is that X takes x, Y to take the joint probability of y.The present invention in the calculating of entropy and mutual information, connection attribute has been carried out from
Dispersion is processed.
From formula (14), R (X；Y span) is [0,1], and it reflects the dependency between attribute X and Y,
If big, R (X；Y) attribute X and Y strong correlation；Otherwise then attribute X is weak to Y related.
(2) redundant attributes are rejected
If R is (X；Y) >=α (present invention takes α=0.95), then attribute X and Y strong correlation.Further, if H (X)≤H (Y),
Then reject X；Conversely, rejecting Y.If the property set after rejecting redundant attributes is Ω.
(3) attributive classification
If Ω1、Ω2With any 2 mutually disjoint subsets of Ω, the degree of association defining between them is：
R(Ω1；Ω2)=max { R (Zi；Zj),Zi∈Ω1,Zj∈Ω2} (16)
Classification is exactly that Ω is divided into several mutually disjoint subsets, random subset ΩiAnd ΩjBetween be satisfied by R (Ωi；Ωj)
＜ β (present invention takes β=0.04), that is, belong to the dependency very littles between inhomogeneous attribute.Present invention classification adopts recurrence
Algorithm, its detailed process is as follows：1. there is n element in current Ω, by each element separately as 1 class, so, Ω is divided into
N class；2. determine the degree of association between any 2 classes using formula (16)；3. 2 maximum classes of degree of association are merged into 1 class,
N=n-1；4. the degree of association of any 2 classes of n apoplexy due to endogenous wind is determined using formula (16)；If 5. n ≠ 2, go to step 3.；If 6. n=2,
Illustrate that Ω has been divided into 2 classes, be designated as Ω here1、Ω2If, R (Ω1；Ω2) ＜ β, Ω is divided into Ω1、Ω2, otherwise Ω labelling
For " inseparable "；7. recursive procedure：To each subclass, repeat step 1.～6. classified again, until all subclasses all " can not
Point ", complete to classify.
(4) select characteristic attribute
If classification obtains K attributive classification, use ΩiIt is expressed as the i-th class.mi、WithIt is respectively ΩiWork as premise
For characteristic attribute number, characteristic attribute set, provide quantity of information；F is characterized selection result property set；M is the feature in F
Attribute number；C is objective attribute target attribute, definition：
In formula：Attributeη is constant, and the present invention takes η=0.3.
Select the process of M characteristic attribute as follows：
(1) initialize：M=0 is rightI(Fi 0；C)=0.RightTo mi=0, in ΩiIn withIt is to the maximum
1 attribute composition of target selection
(2) m increases by 1, i.e. m ← m+1 on the basis of original, and the step of m characteristic attribute of selection is：1. rightCalculate letter
Breath gainIf 2. class ΩiInformation gainMaximum, then update characteristic attribute set：3. prepare for next step feature selection：To Ωi, by miIncrease by 1 on the basis of original, i.e. mi←mi+
1, withIt is target to the maximum and select miIndividual attribute composition
(3) if m is ＜ M, (2) are gone to step；Otherwise output characteristic property set F.
Rule in rule base generates the traditional decision-tree employing based on C4.5 algorithm, and the root node of decision tree is to leaf segment
Every paths of point all correspond to an IF-THEN conditional plan.The present invention is with main distribution integration high-speed simulation result as sample
Space, the running status key feature attribute after the electric network fault selected using Feature Extraction Technology and fault are optimum to recover control
System strategy, as input attribute, as objective attribute target attribute, generates decision tree based on C4.5 algorithm.
C4.5 decision Tree algorithms, based on theory of information, as judgement and are selected using comentropy and information gain-ratio for attribute
Standard, the mode using the top-down search of greedy algorithm generates decision tree.Its concrete grammar is：Detect all of attribute,
Select the maximum attribute of information gain-ratio to produce decision tree nodes, set up branch by the different values of this attribute, then to each branch
Subset recursive call the method set up the branch of decision tree nodes, until all subsets only comprise same category of data be
Only.
If S is the set of s sample.Hypothetical classification attribute has m different value, and defining m different value is Ci(i=
1,...,m).If siFor class CiIn sample number.Then expectation information sample classification needed for given to one is：
In formula：pi=si/ s belongs to C for sampleiProbability.Due to information binary coding, then logarithmic function with 2 is
Bottom.
If attribute A has v different value { a1,a2,...,av, then with attribute A, S is divided into v subset { S1,S2,...,
Sv, wherein SjIn sample identical value a is had on attribute Aj(j=1,2 ..., v).If sijIt is subset SjMiddle class CiSample
This number.To given subset Sj, it is desired for by the entropy or information of A dividing subset：
In formula：pij=sij/sjIt is SjMiddle sample belongs to CiProbability.
In the obtainable information gain of attribute A upper branch it is then：
Gain (A)=I (s1,s2,...,sm)-E(A) (20)
The comentropy of the set of sample point is：
Then information gain-ratio is：
Decision tree generation method process based on C4.5 algorithm is as follows：
(1) decision tree has represented the individual node of the whole record of whole sample set as root node；
(2) if sample record broadly falls into same class, this node is leaf node, and with the class belonging to sample in this node
It is marked；
(3) otherwise, the metric of algorithm use information ratio of profit increase is as heuristic information, from sample attribute candidate collection
Selection can be by the attribute of sample optimal classification, and this attribute is referred to as " test " or " judgement " attribute of this node, for the present invention
Operation of power networks status attribute have many successive values, need first by its discretization；
(4) each given value to selected testing attribute, creates a branch, and accordingly sample data is divided into each
In branch；
(5) the equally applicable said process of algorithm recurrence, forms the subsample decision tree of each division, once an attribute quilt
It is elected to be the testing attribute for a node it is not necessary to consider further that the probability that testing attribute is made on the spawn of this node；
(6) recurrence partiting step stops only when one of following condition is set up：
1) all sample datas giving on node belong to same class, and that is, the value of all record class label attributes is identical；
2) remaining candidate attribute is not had to can be used to Further Division sample, in the case, using majority vote method,
This need compulsory the node treating point row is converted into leaf node, and record affiliated class with most in this node sample data
Labelling it；
3) after dividing row, in certain branch, there is no sample record, in this case, create one with the many several classes ofs in sample data
Individual leaf node.
Embodiment 4
In the step 4 of the present invention：
The operation situation perception of main distribution refers in main distribution range of operation, cognitive, understand each in electrical network real time execution
The running status of individual equipment and operational factor, all information are identified and convert based on distribution real-time running state amount, and discriminate
Not and eliminate wherein wrong, invalid information.Fault in the present invention mainly comes from electrical network external environmental factor and interior
Disturbance to electrical network during the equipment operation characteristic change in portion.Therefore, situational awareness techniques mainly perceive this disturbance, to it
Carry out a series of analyses, obtain operation of power networks state and operational factor after this disturbance.The fault main contents of perception specifically may be used
To come from the data of different data sources, mostly come from external data and internal data in the present invention, external data is main
Refer to the high accuracy meteorological disaster data that local meteorological department is provided；Internal data can be divided into dynamic data and static data again
Two parts, dynamic data is primarily referred to as the electrical network real-time running state amount data extracted from EMS/DMS and SCADA, static data
It is primarily referred to as the property parameters of electrical network electric components and the geodata of power grid GIS, the geodata obtaining from GIS mainly has
The geography of local power distribution network is along Butut, SVG figure of circuit line chart and components and parts etc..Therefore, by the run mode of main distribution
Gesture cognition technology can obtain the huge data source of the above in real time, based on distribution integration incident response decision system under
Basis is set up in one step analysis.
In a word, the situational awareness techniques being run by main distribution integration, can be obtained master from real-time data of power grid and join
The information of the main distribution running status required for net integrative simulation, and the current fault message occurring.And according to these
Raw information, by being calculated the characteristic quantity information of coupling of making decisions on one's own.
1. main Distribution Network Failure is made decisions on one's own coupling and is responded
The real-time running state information of main distribution is continually changing in actual motion, after system jam is detected
Need to match from scene decision ruless storehouse as early as possible quickly through coupling of making decisions on one's own and meet current main distribution running status
Rule is executing.Complicated yet with grid operating conditions, in main distribution integration scene decision ruless storehouse fuzzy rules mistake
In huge, if mated to whole rule base every time, its efficiency will be very low, be improved autonomous using certain method
The efficiency of decision-making coupling is imperative, accelerates the decision-making matching process of rule base therefore in the present invention using Rete algorithm.
(1) principle of Rete algorithm
Rete algorithm is current one forward direction chain matching algorithm of efficiency highest, and its core concept is by detached coupling
Item mates tree according to content dynamic construction, to reach the effect significantly reducing amount of calculation.During matched rule, rule
Might have a lot of identical Rule of judgment in premise, therefore in matched rule premise, substantial amounts of repetitive operation will be carried out, this
Sample just brings time redundancy sex chromosome mosaicism.For example：
RULE1:IF(A>B) and C or D THEN E=100
RULE2:IF(A>B)and (B<C) THEN E=200
RULE3:IF(A>B)or (B<C) THEN E=300
During to mate this 3 rule, three calculating to be carried out for expression formula A ＞ B, B ＜ C is needed to calculate twice.
Rete adopt method be：Make M=A ＞ B, N=B ＜ C, then above-mentioned three rules are rewritable is：
RULE1:IF M and C or D THEN E=100
RULE2:IF M and N THEN E=200
RULE3:IF M or N THEN E=300
So only when A or B changes, just recalculate M；Equally when B or C changes, just recalculate
N.Such processing method avoids when carrying out rule match every time all double counting identical expression formulas, as long as and detecting phase
Whether off status variable changes to decide whether more new-standard cement, so saves the plenty of time in the matching process and opens
Pin, thus improve matching efficiency.As can be seen that Rete algorithm needs to store extra match information, it is one and is changed with space
Take the algorithm of time.
In actual motion, the change of most of state variable is continuous and slow, in the shorter time period for main distribution
Be not in inside too big fluctuation, even if break down in somewhere in major network or distribution, from its most of feelings of part electrical network farther out
Can't be much affected under condition, if all of state variable is all mated in each circulation, efficiency will be very low,
Using Rete algorithm, change less state variable in matched rule and be not repeated to mate, become just for changing big state
Amount is compared, and can be greatly enhanced efficiency.
(2) network structure of Rete figure and foundation
The realizing flow process and can scheme to illustrate by Rete of Rete algorithm.Rete figure containing type node (Type Node),
Alpha node (Alpha Node) and Beta node (Beta Node), the reality of make decisions on one's own in main Distribution Network Failure coupling and response
During existing, type node may be configured as the state variable of main distribution being compared when mating, and Alpha node is every rules and regulations
The concrete condition meeting of each state variable, the result that Beta node is then combined successively by Alpha node in then.
Strictly all rules in scene decision ruless storehouse is built according to the structure of Rete figure shown in 7 and can form into Rete coupling
Network.Its specific building process is as follows：
1) create root node, form the entrance of matching network；
2) extract new rule from scene decision ruless storehouse, extract each state variable from regular condition and meet
Condition：
A) check in this regular condition and whether occur in that new state variable, if new state variable, add one
Individual type node；
B) check condition that each state variable in this rule meets whether Already in Alpha node, if
Exist and then record the position of this Alpha node, if it is not, the condition that this state variable is met is new as one
Alpha node is added in network；
C) repeat b), until the condition that state variable all of in this rule is met all is disposed；
D) combine Beta node：First Beta node is formed by two Alpha combination of nodes, and each Beta thereafter saves
Point is combined by the Beta node of last layer and new Alpha node；
E) repeat d), until generating all Beta nodes；
F) this regular corresponding execution action is packaged into final node；
3) 2 are repeated), until the strictly all rules in rule base is all disposed.
(3) coupling is made decisions on one's own based on the main Distribution Network Failure of Rete algorithm and respond
Based on Rete algorithm carry out main Distribution Network Failure make decisions on one's own coupling with response when, except during first fit need to master
Beyond each state variable in distribution is mated, afterwards when a cycle is mated, only need to differentiate and currently lead
The distribution state variable part big compared to last mechanical periodicity, according to the state variable changing greatly corresponding Alpha node
Continue directly to follow-up coupling work from Beta network, eliminate to the mistake changing little part state variable repeated matching
Journey.Its concrete matching process is as follows：
(1), during first fit, relatively in main distribution, the information of all state variables, finds rule accordingly；
(2) every certain cycle, the state variable in detecting system, compared to the variable quantity in last cycle, screens out it
Middle change exceedes the state variable of decision tree respective conditions place discrete segment；
(3) the Alpha node corresponding to the state variable according to change, the minimum Beta section of the wherein corresponding number of plies of search
Point, begins look for the Rete figure meeting the rule of condition and recording this coupling from this Beta node；
(4) the corresponding action of executing rule, responds action.
Embodiment described above only have expressed embodiments of the present invention, and its description is more concrete and detailed, but can not
Therefore it is interpreted as the restriction to the scope of the claims of the present invention.It should be pointed out that for the person of ordinary skill of the art,
Without departing from the inventive concept of the premise, some deformation can also be made and improve, these broadly fall into the protection model of the present invention
Enclose.Therefore, the protection domain of patent of the present invention should be defined by claims.
Claims (12)
1. a kind of main distribution integration incident response decision method it is characterised in that：The method comprising the steps of：
Step one, set up main distribution integration and run scene collection and fault set；
Step 2, global coordination control strategy knowledge base is set up based on main distribution integration high-speed simulation result；
Step 3, using feature extraction, decision tree generation technique generate emulation scene decision ruless storehouse；
Step 4, after main distribution running status cognition technology and matching process of making decisions on one's own realize fault, main distribution is autonomous
Decision-making coupling response.
2. main distribution integration incident response decision method according to claim 1 is it is characterised in that in step one,
Described operation scene collection is generated by Latin hypercube simulation, predictive power and forecast error are distributed, to each input with
Machine variable is sampled it is ensured that random distribution region can be sampled a little is completely covered；Change the row of each stochastic variable sampled value
Row order, is made the dependency of the sampled value of mutually independent random variables tend to minimum, is arranged using Cholesky decomposition method
Row order,
Sampled value formula is
Wherein, XnmRepresent stochastic variable XnM-th sampled value, UmRepresent sampled value.
3. main distribution integration incident response decision method according to claim 1 is it is characterised in that in step one,
By providing the comprehensive measurement of each fault rate and seriousness, fault is screened with this, generate described fault set, comprehensive measurement
Mathematic(al) representation be：
Wherein, XfFor current system running status；EiFor i-th forecast failure；Pr(Ei) it is EiThe probability occurring；Sev(Ei) it is Ei
The order of severity of system loss after generation.
4. main distribution integration incident response decision method according to claim 1 is it is characterised in that in step one,
Failure probability model formula is：
Pr(Fi)=1-exp (- λits),
Wherein, FiBreak down for i-th line road；Pr(Fi) probability that breaks down for i-th line road；λiFor i-th line road
Fault rate；
After system jam, the overload penalty values computing formula of branch road is：
Wherein, ωLiRepresent overload penalty values, LiThe ratio of the actual fed power for branch road i and power allowances；L0For the threshold setting
Value, if LiLess than L0, dispatcher thinks that this branch road does not have overload risk；
Define branch road i overload order of severity computing formula be：
Wherein, a, c are positive number, ωLiRepresent overload penalty values；
Voltage out-of-limit penalty values ω of system jam posterior nodal point iViComputing formula is：
Wherein, UiVoltage magnitude for node i；Uimax、UiminIt is respectively the upper and lower limit of the voltage magnitude of node i；
Define node i voltage out-of-limit order of severity computing formula be：
Wherein, a, c are positive number, ωviRepresent voltage out-of-limit penalty values；
By Contingency screening and ranking, some faults coming in ranking results above are generated described fault set.
5. main distribution integration incident response decision method according to claim 1 is it is characterised in that in step 2,
By the running status before electric network fault, the running status after electric network fault, fault is optimum to recover control strategy group to described knowledge base
Become, scene collection is run by input and fault set carries out main distribution integration high-speed simulation, carry out main distribution global optimization and obtain
Fault is optimum to recover control strategy, running status, control strategy before and after the fault that record electrical network occurs and fault, sets up described complete
Office's coordination strategy knowledge base.
6. main distribution integration incident response decision method according to claim 5 is it is characterised in that set up knowledge base
Step is as follows：
Step 2a master, distribution intelligent body obtain current time electrical network section information, are set up by main distribution integration state estimation
Each trend initial state；
Step 2b main distribution intelligent body collects weather information data in respective electrical network, passes through Latin in conjunction with electrical network typical operation modes
Hypercube Sampling techniques obtain operation of power networks scene collection.Obtain grid simulation fault set；
For in failure collection, each treats simulated fault to step 2c, controls optimization problem using main distribution integration fault recovery
Solve, obtain main distribution coordination control strategy；
Step 2d is by electric network state after electric network state before fault, fault and control strategy write global coordination control strategy knowledge
Storehouse, is updated to original storehouse.
7. main distribution integration incident response decision method according to claim 1 is it is characterised in that in step 3,
Described emulation scene decision ruless storehouse is to operation of power networks state in global coordination plan knowledge storehouse by Feature Selection
Key feature attribute is extracted, and using the characteristic attribute selected as input attribute, automatically generates decision tree, from decision tree
Extract finely rule, set up emulation scene decision ruless storehouse.
8. main distribution integration incident response decision method according to claim 7 is it is characterised in that separated based on attribute
Carry out feature selection, higher-dimension combinatorial problem is reduced to low-dimensional combinatorial problem, between attribute, the computing formula of degree of association is：
Wherein, H (X), H (Y) are respectively the entropy of attribute X and Y；I(X；Y) it is mutual information between attribute X and Y；
Select characteristic attribute formula be：
Wherein, k is the quantity of attributive classification, ΩiIt is expressed as the i-th class.mi、WithIt is respectively ΩiThe current spy providing
Attribute number, characteristic attribute set, the quantity of information providing are provided；F is characterized selection result property set；M is the characteristic attribute in F
Number；C is objective attribute target attribute.
9. main distribution integration incident response decision method according to claim 7 is it is characterised in that described decision tree is calculated
Method is to detect all of attribute, selects the maximum attribute of information gain-ratio to produce decision tree nodes, by the different values of this attribute
Set up branch, subset recursive call the method for more each branch is set up with the branch of decision tree nodes, until all subsets are only wrapped
Till same category of data.
10. main distribution integration incident response decision method according to claim 1 is it is characterised in that in step 4,
Described main distribution running status sensing method, in main distribution range of operation, cognitive, understand each equipment in electrical network real time execution
Running status and operational factor, all information are identified and convert based on distribution real-time running state amount, and screen and reject
Go out wherein wrong, invalid information, by analyze external data and internal data obtain after electrical network is disturbed running status and
Parameter, by being calculated the characteristic quantity information of coupling of making decisions on one's own.
11. main distribution integration incident response decision methods according to claim 1 it is characterised in that in step 4,
Accelerate the decision-making coupling of rule base using Rete algorithm, the step building Rete matching network includes：
Step 4a creates root node, forms the entrance of matching network；
Step 4b extracts new rule from scene decision ruless storehouse, extracts each state variable and meet from regular condition
Condition：
I) check in this regular condition and whether occur in that new state variable, if new state variable, add a class
Type node；
Ii) check condition that each state variable in this rule meets whether Already in α node, if there is then remembering
The position of record this α node lower, if it is not, the condition that this state variable is met is added to net as a new α node
In network；
Iii) repeat step ii), until the condition that state variable all of in this rule is met all is disposed；
Iv) combine β node：First β node is formed by two α combination of nodes, and each β node thereafter is saved by the β of last layer
Point and new α node are combined；
V) repeat step iv), until generating all β nodes；
Vi) this regular corresponding execution action is packaged into final node；
Step 4c repeat step 4b, until the strictly all rules in rule base is all disposed.
12. main distribution integration incident response decision methods according to claim 11 are it is characterised in that in step 4
In, based on Rete algorithm carry out main Distribution Network Failure make decisions on one's own coupling with response when, except during first fit need to main distribution
In each state variable mated beyond, afterwards when a cycle is mated, only need to differentiate currently main distribution
The state variable part big compared to last mechanical periodicity, according to the state variable changing greatly corresponding α node from β network
Continue directly to mate work, eliminate to the process changing little state variable repeated matching.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610305304.8A CN106485594A (en) | 2016-05-10 | 2016-05-10 | A kind of main distribution integration incident response decision method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610305304.8A CN106485594A (en) | 2016-05-10 | 2016-05-10 | A kind of main distribution integration incident response decision method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106485594A true CN106485594A (en) | 2017-03-08 |
Family
ID=58238715
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610305304.8A Pending CN106485594A (en) | 2016-05-10 | 2016-05-10 | A kind of main distribution integration incident response decision method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106485594A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107609760A (en) * | 2017-08-30 | 2018-01-19 | 清华大学 | The key feature system of selection of power system and device |
| CN108288855A (en) * | 2018-01-23 | 2018-07-17 | 国电南京自动化股份有限公司 | A kind of dynamic strategy simulation and optimization method based on micro-capacitance sensor |
| CN108390371A (en) * | 2018-02-11 | 2018-08-10 | 中国电力科学研究院有限公司 | Network system protects control strategy modeling method in on-line analysis |
| CN108416443A (en) * | 2018-03-29 | 2018-08-17 | 北京润科通用技术有限公司 | A kind of method for diagnosing faults and device |
| CN110222962A (en) * | 2019-05-24 | 2019-09-10 | 东方银谷(北京)科技发展有限公司 | Data configuration method and device for financial business risk control |
| CN111356959A (en) * | 2017-11-24 | 2020-06-30 | 西门子股份公司 | Method for the computer-aided control of a technical system, in particular a power plant |
| CN112308330A (en) * | 2020-11-09 | 2021-02-02 | 清华大学 | Digital accident database construction method and device and computer equipment |
| CN113033003A (en) * | 2021-03-30 | 2021-06-25 | 华北电力大学（保定） | Load recovery model generation method and device, computer equipment and storage medium |
| CN113051553A (en) * | 2019-12-27 | 2021-06-29 | 北京安控科技股份有限公司 | Accident rule generation method and system |
| CN113450612A (en) * | 2021-05-17 | 2021-09-28 | 云南电网有限责任公司 | Development method of complete teaching device applied to relay protection training |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104810826A (en) * | 2015-05-07 | 2015-07-29 | 东南大学 | Bidirectional iteration parallel probability load flow calculation method combining Latin hypercube sampling |
| CN105305444A (en) * | 2015-11-30 | 2016-02-03 | 国网浙江省电力公司绍兴供电公司 | Distribution network intelligent alarm and processing method based on integrated allocation |
-
2016
- 2016-05-10 CN CN201610305304.8A patent/CN106485594A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104810826A (en) * | 2015-05-07 | 2015-07-29 | 东南大学 | Bidirectional iteration parallel probability load flow calculation method combining Latin hypercube sampling |
| CN105305444A (en) * | 2015-11-30 | 2016-02-03 | 国网浙江省电力公司绍兴供电公司 | Distribution network intelligent alarm and processing method based on integrated allocation |
Non-Patent Citations (2)
| Title |
|---|
| 罗少杰等: "基于概率性分析的配电网风险评估研究", 《中国电业(技术版)》 * |
| 顾伟等: "在线电压安全评估的多重动态决策树方法", 《中国电机工程学报》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107609760B (en) * | 2017-08-30 | 2020-11-24 | 清华大学 | Key feature selection method and device for power system |
| CN107609760A (en) * | 2017-08-30 | 2018-01-19 | 清华大学 | The key feature system of selection of power system and device |
| CN111356959A (en) * | 2017-11-24 | 2020-06-30 | 西门子股份公司 | Method for the computer-aided control of a technical system, in particular a power plant |
| US11720069B2 (en) | 2017-11-24 | 2023-08-08 | Siemens Aktiengesellschaft | Method for the computer-aided control of a technical system, more particularly a power generation installation |
| CN108288855A (en) * | 2018-01-23 | 2018-07-17 | 国电南京自动化股份有限公司 | A kind of dynamic strategy simulation and optimization method based on micro-capacitance sensor |
| CN108288855B (en) * | 2018-01-23 | 2020-11-24 | 国电南京自动化股份有限公司 | Dynamic strategy simulation and optimization method based on micro-grid |
| CN108390371A (en) * | 2018-02-11 | 2018-08-10 | 中国电力科学研究院有限公司 | Network system protects control strategy modeling method in on-line analysis |
| CN108390371B (en) * | 2018-02-11 | 2021-03-02 | 中国电力科学研究院有限公司 | Power grid system protection control strategy modeling method in online analysis |
| CN108416443A (en) * | 2018-03-29 | 2018-08-17 | 北京润科通用技术有限公司 | A kind of method for diagnosing faults and device |
| CN110222962A (en) * | 2019-05-24 | 2019-09-10 | 东方银谷(北京)科技发展有限公司 | Data configuration method and device for financial business risk control |
| CN113051553A (en) * | 2019-12-27 | 2021-06-29 | 北京安控科技股份有限公司 | Accident rule generation method and system |
| CN112308330A (en) * | 2020-11-09 | 2021-02-02 | 清华大学 | Digital accident database construction method and device and computer equipment |
| CN112308330B (en) * | 2020-11-09 | 2021-07-09 | 清华大学 | Digital accident database construction method and device and computer equipment |
| CN113033003B (en) * | 2021-03-30 | 2022-04-05 | 华北电力大学（保定） | Load recovery model generation method and device, computer equipment and storage medium |
| CN113033003A (en) * | 2021-03-30 | 2021-06-25 | 华北电力大学（保定） | Load recovery model generation method and device, computer equipment and storage medium |
| CN113450612A (en) * | 2021-05-17 | 2021-09-28 | 云南电网有限责任公司 | Development method of complete teaching device applied to relay protection training |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106485594A (en) | A kind of main distribution integration incident response decision method | |
| Xie et al. | A review of machine learning applications in power system resilience | |
| Li et al. | A hierarchical data-driven method for event-based load shedding against fault-induced delayed voltage recovery in power systems | |
| CN107179503A (en) | The method of Wind turbines intelligent fault diagnosis early warning based on random forest | |
| CN109635928A (en) | A kind of voltage sag reason recognition methods based on deep learning Model Fusion | |
| CN107563122A (en) | The method of crime prediction of Recognition with Recurrent Neural Network is locally connected based on interleaving time sequence | |
| CN108876163A (en) | The transient rotor angle stability fast evaluation method of comprehensive causality analysis and machine learning | |
| Kulluk et al. | Fuzzy DIFACONN-miner: A novel approach for fuzzy rule extraction from neural networks | |
| CN102722719A (en) | Intrusion detection method based on observational learning | |
| CN106649479A (en) | Probability graph-based transformer state association rule mining method | |
| Qu et al. | False data injection attack detection in power systems based on cyber-physical attack genes | |
| CN112734154A (en) | Multi-factor public opinion risk assessment method based on fuzzy number similarity | |
| Rizvi et al. | Data-driven short-term voltage stability assessment using convolutional neural networks considering data anomalies and localization | |
| CN106526465A (en) | High-voltage circuit breaker fault intelligent diagnosis method based on improved fuzzy Petri network | |
| Atrigna et al. | A machine learning approach to fault prediction of power distribution grids under heatwaves | |
| Ren et al. | A super-resolution perception-based incremental learning approach for power system voltage stability assessment with incomplete PMU measurements | |
| CN106202805B (en) | A kind of switchgear method for evaluating reliability | |
| Khosla et al. | Learning knowledge and strategy of a neuro-expert system architecture in alarm processing | |
| Kumar et al. | Deep Learning based Fault Detection in Power Transmission Lines | |
| CN116956702A (en) | Electricity safety early warning method, medium and system | |
| CN115733730A (en) | Power grid fault detection method and device based on graph neural network | |
| Haiwen et al. | Fuzzy Petri nets reasoning for application of electric control system fault diagnosis | |
| Abberley et al. | Modelling road congestion using a fuzzy system and real-world data for connected and autonomous vehicles | |
| CN110336280B (en) | Power system cascading failure analysis method based on dictionary set acceleration | |
| CN115598459A (en) | Power failure prediction method for 10kV feeder line fault of power distribution network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |