CN106022614A - Data mining method of neural network based on nearest neighbor clustering - Google Patents
Data mining method of neural network based on nearest neighbor clustering Download PDFInfo
- Publication number
- CN106022614A CN106022614A CN201610343564.4A CN201610343564A CN106022614A CN 106022614 A CN106022614 A CN 106022614A CN 201610343564 A CN201610343564 A CN 201610343564A CN 106022614 A CN106022614 A CN 106022614A
- Authority
- CN
- China
- Prior art keywords
- data
- neural network
- neutral net
- cluster
- nearest neighbor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 35
- 238000007418 data mining Methods 0.000 title claims abstract description 19
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 28
- 230000007935 neutral effect Effects 0.000 claims description 44
- 238000012549 training Methods 0.000 claims description 25
- 230000001537 neural effect Effects 0.000 claims description 21
- 230000008569 process Effects 0.000 claims description 16
- 238000012360 testing method Methods 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 10
- 230000003044 adaptive effect Effects 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 238000012790 confirmation Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 238000005065 mining Methods 0.000 abstract 1
- 238000009826 distribution Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000007726 management method Methods 0.000 description 5
- 230000005611 electricity Effects 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000009897 systematic effect Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000012098 association analyses Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013523 data management Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000037361 pathway Effects 0.000 description 1
- 238000013439 planning Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
-
- 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—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- General Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a data mining method of the neural network based on nearest neighbor clustering. An improved nearest neighbor clustering learning algorithm is used to train the neural network, so that on the premise that the precision requirement is met, the number of nodes of a hidden layer is reduced, the network structure is simplified, the learning speed of the neural network is increased, and the learning efficiency and precision of the neural network are further improved; and further data mining is carried out on the neural network, and the mining efficiency for a database of a large practical power system via the neural network is further improved. The data mining method aims at ensuring the safe, stable, high-quality and economical operation of the power system, and satisfying higher requirements for mass data of the digital power system and reliability, consistency and sharing performance of data.
Description
Technical field
The present invention relates to a kind of data digging method being applied to power system relevant information, particularly relate to a kind of utilization
The Neural Network Data method for digging of NNCA algorithm training neutral net, the invention belongs to electric power system data analysis neck
Territory.
Background technology
Along with the popularizing in power system with information technology and computer technology that develop rapidly of power industry, digital
Change technology is used widely in recent years, occurs in that power informatization--digital power system.On-line monitoring system, transaction
System, geographic information management system, fault diagnosis, the various analytical calculations of off-line and planning system, and the day of electric power enterprise
The systems such as often issued transaction, communication and energy management are widely applied at electric power enterprise.But power system be one dynamically
Nonlinear big system, these systems constantly produce in running and accumulate substantial amounts of data, and these real time datas are
In blast growing trend.Additionally, PsS/E, the power system simulation software such as EMTP, PsAsP, BPA is at Electrical power system analysis and computing
In extensive application, also make system create substantial amounts of emulation data.Data management system based on traditional database, along with number
According to the increase of amount, statistical query performance significantly declines, and user can not optionally utilize these substantial amounts of data to carry out statistical
Analysis, and data user rate is low, and the historical data of magnanimity is being sunk into sleep dumbly, valuable feature extraction difficulty in data,
Cause utilizing these data that business is predicted in time and instructing the most relatively difficult.
Requirements at the higher level reliability, concordance and the sharing of data message proposed along with people, and preferably protect
Safe and stable, the high-quality of card power system, economic operation, problem the most in the urgent need to address in power system is exactly such as
What carries out integrated treatment to magnanimity, time-varying and mobile data, and the data collected are carried out data mining.The most permissible
More fully utilize service data, disclose rule and principle that historical data contains behind, find and more reasonably solve to ask
The method of topic, for the formulation of decision-making with perform to provide stronger scientific basis.Main in power system of data mining
Application has Load Prediction In Power Systems and classification, operational mode classification, operation states of electric power system and the equipment shape of power system
State monitoring and power scheduling optimization, power system modeling etc..
Data mining (Data Mninig) is exactly from substantial amounts of, incomplete, noisy, fuzzy, random data
In, extract lie in therein, people are the most ignorant but be information and the process of knowledge of potentially useful.Mainly have
Conceptual description, association analysis, cluster, automatic Prediction trend and a few class function such as behavior, separate-blas estimation.Cluster is exactly by data pair
As be divided into multiple class or bunch so that the object in same bunch is more similar, and the object in different bunches is dissimilar.Power system
Research worker carries out cluster analysis to different power consumers or Power Generation, to obtaining different category attributes.Supervise from electric power
Survey and dispatching patcher is extracted different regions different types of customer charge curve, carrying out electricity consumption Specialty aggregation analysis, for electric power
Company's marketing and load management provide foundation.
Neutral net provides the most relatively effective straightforward procedure of one for solving challenge.Neutral net has
There are the characteristics such as good robustness, self-organizing, self adaptation, self study, parallel processing, distribution storage and Error Tolerance, can basis
New input data adaptive adjusts network parameter.And neutral net has stronger ability to bear to noise data, to data
Classification accuracy is high, and available various algorithms carry out Rule Extraction.The more important thing is that neutral net is easy in parallel computation
Realize on machine, its node can be assigned to parallel computation on different CPU.Therefore, it can carry out by neutral net
Data mining.
But during utilizing neural networks for data mining, Learning Algorithm is it cannot be guaranteed that converge to
Preferably result；Neutral net is easy to overtraining, thus causes place of working on the training data fine, and in inspection data
Upper performance is not good enough.And the learning time length of neutral net affects its application in data mining, the training time of network
Length is relevant to the scale of problem, the complexity of network and training algorithm.The present invention uses RBF neural.Because RBF is refreshing
Through network have good approach in any nonlinear mapping and processing system the regular ability being difficult to Analytical Expression；
The topological structure of RBF neural not only makes pace of learning greatly speed up, and avoids local minimum problem；RBF is neural
Another outstanding advantages of network is that interpretability is good.The topological structure of RBF neural affects nerve net to a great extent
The performance of network self, and determine that the factor of RBF neural topological structure mainly has center vector, the Hidden nodes of four: RBF
Mesh, the width of RBF and hidden layer are to the weight matrix between output layer.The width parameter of kernel function determines hidden
The node response range to external input signal, affects the number at center, pace of learning and precision.K-means clustering algorithm only
Can reach to rely on the locally optimal solution of selected center initial value.General NNCA algorithm needs rule of thumb and priori
Determine suitable cluster radius, cannot change after determining, be unfavorable for the application of algorithm.And along with the increasing of input data
Adding, the nodes meeting monotonic increase of hidden layer, this will produce substantial amounts of redundant node, cause network structure the hugest, from
And can not relation between cooperated learning precision and pace of learning, affect the result of data mining.
The Neural Network Data method for digging based on nearest neighbor classifier that the present invention proposes, is based on above demand, real
Show and electric power system data has been excavated more efficiently.
Summary of the invention
Present invention aims to the deficiency of available data digging technology, it is provided that one can be for electric power system data
Carry out the in hgher efficiency data digging method utilizing nearest neighbor classifier optimization neural network.The method mainly solves neutral net
The problem that in data mining process, network is complicated, learning time is long, it is ensured that obtain optimal electric power system data mining effect.
To achieve these goals, the technical solution used in the present invention is as follows:
A kind of Neural Network Data method for digging based on nearest neighbor classifier being applicable to power system, described method are provided
Comprise the following steps:
Step 1: electric power data is carried out and selects
Electric power data has higher-dimension, discrete data and continuous data mixing, the time response of data and statistical property, deposits
In the uncertainty such as problem such as noise, data incomplete, the Data Warehouse as data initial set is a lot, but the most only needs
Want a portion data for a certain decision-making.It is thus desirable to the data for this data mining are selected.General feelings
Under condition, which parameter is it is important that ignorant for a certain decision-making, but neutral net can assist solution, and this is asked
Topic, it can set up a model relevant to this parameter.
Step 2: to electric power data pretreatment and conversion
Data prediction is exactly the process that the clean data selected carry out enhancement process.Neural Network Data is excavated
For, also need to convert the data into a kind of form that can be accepted by Neural Network Data mining algorithm.Neutral net can only be located
Managing numerical data, text data needs to be converted to the numerical data that neutral net is capable of identify that.Most of neutral nets
Model only accepts the data value of (0,1) or (-1,1) scope, and the data in power system exist relatively on the order of magnitude of numerical value
Big difference, therefore, it is interval that data must be normalized this to training sample.Scalar data value is the most uniform
Ground distribution within a certain range, can map directly to interval (0, l)；If numeric distribution is uneven, available segment linear equation
Or logarithmic equation changes, it is scaled down to the most again specify interval；Discrete data is by encoding it with 0 and 1
Represent.
The maximum assuming the training sample set of system is Dmax, minima is Dmin, data are originally as Di, then
Normalized is extremely important to network training, beneficially convergence during neural metwork training, it is possible to be effectively improved neutral net
Pace of learning, reduces the training time, it is to avoid neutral net is the sensitiveest or insensitive to a certain input quantity.
Step 3: the management of data set
Initial data is randomly divided into training dataset, test data set and three data sets of confirmation data set, front
Two data sets are used for training neutral net, the precision of test network carrys out constructing neural network model, confirm data set independently
Test network, the ratio of these three data set is respectively 80%, 10% and 10%.
Step 4: determine neural network type, algorithm and train neutral net；
The present invention uses three layers of feedforward RBF neural, and uses the closest clustering algorithm of improvement to train this god
Through network.Neutral net input, the number of output node are determined by the concrete decision-making of power system, the number of hidden layer node by
NNCA algorithm determines.Training sample set equal intervals after normalization is chosen data and is trained, according to specifically wanting
Ask and training precision is set.The improvement closest clustering algorithm that the present invention uses can adjust cluster radius automatically, is meeting
Under the requirement of systematic function, by the adjustment to cluster radius so that it is reach the value of a satisfaction, so that cluster centre number
Reach optimal, be that parameter and two processes of structure of RBF neural carry out online adaptive adjustment.
Step 5: data result display output, and Result is analyzed.
Effectively result:
The present invention provides a kind of Neural Network Data method for digging based on nearest neighbor classifier being applicable to power system, can
The mass data in power system is analyzed, processes, reasoning, prediction, the condition finally set according to user, it is achieved
Excellent scheme.
Use the closest clustering algorithm Training RBF Neural Network becoming cluster radius, make neutral net in satisfied essence
On the premise of degree requires, reduce the number of hidden nodes, thus simplify network structure, accelerate the pace of learning of neutral net.This side
Method can make RBF neural can carry out the self-adaptative adjustment of network parameter and network structure online simultaneously, can eliminate master
See the impact selecting fixing cluster radius to network performance, reach to improve further the mesh of neural network learning efficiency and precision
, on this basis with neural networks for data mining, to improve neutral net further to large-scale practical power systems number
Efficiency when excavating according to storehouse.
Accompanying drawing explanation
Fig. 1 is applicable to power system Neural Network Data method for digging flow chart；
The topology diagram of Fig. 2 RBF neural；
Fig. 3 becomes the NNCA algorithm flow process of cluster radius；
Detailed description of the invention
Below in conjunction with the accompanying drawings embodiments of the invention are elaborated: this gives detailed embodiment and
Specific implementation process, but protection scope of the present invention is not limited to following embodiment.
The data source of Data Warehouse for Power System mostlys come from the EMS (EMS) of power system, electricity consumption battalion
Industry data, GIS-Geographic Information System etc..EMS system saves the method for operation of electrical network, real time execution parameter enters the whole network load, trend
Distribution, maincenter voltage, system frequency etc.；Electric business data include the data such as subscriber data, sale of electricity, electricity price, metering；Geographical letter
Breath system includes the geographical location information of user, power equipment；Other Data Source includes economic situation, meteorological condition, hands
The data etc. of work typing.As it is shown in figure 1, a kind of Neural Network Data method for digging based on nearest neighbor classifier.Including following step
Rapid:
Step 1: electric power data is carried out and selects；
Electric power data has higher-dimension, discrete data and continuous data mixing, the time response of data and statistical property, deposits
In the uncertainty such as problem such as noise, data incomplete.Data Warehouse as data initial set is a lot, but the most only needs
Want a portion data for a certain decision-making, need the data for this data mining are selected.Generally,
Which parameter is it is important that ignorant for a certain decision-making, but neutral net can assist this problem of solution, it
A model relevant to this parameter can be set up.
Step 2: to electric power data pretreatment and conversion；
Data prediction is exactly the process that the clean data selected carry out enhancement process.Neural Network Data is excavated
For, also need to convert the data into a kind of form that can be accepted by Neural Network Data mining algorithm.Neutral net can only be located
Managing numerical data, text data needs to be converted to the numerical data that network is capable of identify that.Most of neural network models
Only accept the data value of (0,1) or (-1,1) scope, and the data in power system exist bigger on the order of magnitude of numerical value
Difference, therefore, it is interval that data must be normalized this to training sample.Scalar data value is essentially homogeneously divided
Cloth within a certain range, can map directly to interval (0, l)；If numeric distribution is uneven, available segment linear equation or right
Number equation is changed, and is scaled down to specify interval the most again；Discrete data carrys out table by it being carried out coding with 0 and 1
Show.
The maximum assuming the training sample set of system is Dmax, minima is Dmin, data are originally as Di, then after normalization
Data be
Normalized is extremely important to network training, beneficially convergence during neural metwork training, it is possible to be effectively improved
E-learning speed, reduces the training time, it is to avoid neutral net is the sensitiveest or insensitive to a certain input quantity.
Step 3: the management of data set；
Data later for pretreatment are randomly divided into training dataset, test data set and confirmation data set three number
According to collection, the first two data set is used for training neutral net, and the precision of test network carrys out constructing neural network model, confirms data set
Test network independently, the ratio of these three data set is respectively 80%, 10% and 10%.
Step 4: determine neural network type, algorithm and training neutral net；
The present invention uses three layers of feedforward RBF neural, and topology of networks figure is as shown in Figure 2.Optimal in order to obtain
Approximation capability, is released Regularized RBF Network by fuzzy inference system
Wherein, ωjFor RBF neural hidden layer to output layer weights, Rj(xk) it is hidden layer unit jth hidden node
It is output as:
In formula: xkInput vector is tieed up for n；cjCenter for hidden layer jth Gaussian function；σjGauss for jth hidden unit
Function widths；M is the number of hidden unit.||xk-cj| | represent xkAnd cjBetween distance.
RBF neural realize from the input space to the nonlinear transformation in hidden layer space depend on RBF center number,
Position and action scope width, i.e. radius r.Radius r is very big on the impact of cluster, the one of traditional RBF center selection algorithm
Individual major defect is to be desirable that Center Number is fixed in advance, then classification capacity and the generalization ability of network will be produced by the selected value of r
Raw appreciable impact.Choose r difficulty by the method artificially trying to gather relatively big, and can not ensure the reasonability of cluster.For cluster
The importance of radius r, the present invention proposes a kind of NNCA algorithm based on adjustment cluster radius, is meeting systematic function
Under requirement, by the adjustment to cluster radius so that it is reach the value of a satisfaction, so that cluster centre number i.e. RBF hidden layer
Nodes reaches optimal, so that cluster centre number reaches optimal, makes the parameter of RBF neural and two processes of structure enter
Row online adaptive adjusts.The idiographic flow of this algorithm is as shown in Figure 3.
In clustering algorithm, introduce adaptive law, choose suitable cluster radius r, make performance index function reach to set
Value, its algorithm is as follows:
Step1: an initial clustering radius r is set, by normalized data xkRead in, defeated as RBF neural
Enter, and calculate the minimum euclidean distance with other data existing, obtain distance d of minimummin, and position is designated as p；
Step2: if dmin＞ r, then cluster numbers adds 1 is m=m+1, and current sample send new cluster centre ci, otherwise pth
Cluster member adds 1, and revises cluster correlation variable；
Step3: all kinds of output vector sums is designated as A (l), represents with enumerator B (l) and belongs to all kinds of for statistics
Number of samples, wherein l is classification number, calculate neutral net hidden layer to output layer weight vector
Wi=A (i)/B (i) (4)
Wherein, i represents i-th iteration, and A (i) represents all kinds of output vector sums during i-th iteration, and B (i) represents i-th
Different categories of samples number sum during individual iteration, W (i) represents i-th output layer weight vector；
(4) Step4: obtain according to the output of Regularized RBF Network
Wherein, ciCenter for hidden layer i-th Gaussian function；
Step5: calculation of performance indicators functional expression
Wherein, N is current iteration total degree, xkInput vector, y (x is tieed up for nk) andBe respectively target data to
The output of neutral net.Performance indications are judged: if J0＞ ε (ε is the least number being previously set), then turn Step6, no
Then turn Step7；
Step6: select certain suitably change step h, then cluster radius is r=r-h, returns Step1；
Step7: cluster terminates；
Step 5: extracting rule from the neutral net trained；
Algorithm based on search, for arbitrary given hidden node or output node, first extracts symbolic rule, then to life
The rule become is attached by the pathway of network and arranges, and is converted to certain intelligible form and expresses, finally
By data result display output, and Result is analyzed and assesses.
Step 6: the rule extracted is estimated.
The rule test data set being extracted and confirmation data set carry out the test of correctness, and detection is in neutral net
In also have how much knowledge to be not extracted by out, exist not between the detection rule and the neutral net trained that are extracted
Conforming place etc..
Claims (7)
1. a Neural Network Data method for digging based on nearest neighbor classifier, it is characterised in that: use nearest neighbor classifier study
Neutral net is trained by algorithm, reduces the number of hidden nodes of neutral net, simplifies network structure, accelerates of neutral net
Practising speed, by neutral net, large-scale practical power systems data base is carried out data mining on this basis, concrete steps include:
Step 1: the electric power data in practical power systems data base is carried out and selects；
Step 2: to the electric power data pretreatment after step 1 processes and conversion；
Step 3: the electric power data after step 2 processes is carried out data set management；
Step 4: the data set produced for step 3, determines neural network type, algorithm and trains neutral net；
Step 5: extracting rule from the neutral net trained；
Step 6: the rule extracted is estimated.
A kind of Neural Network Data method for digging based on nearest neighbor classifier the most according to claim 1, it is characterised in that:
In described step 1, step 2, to the data in power system, it is carried out according to concrete target and needs and selects, rejecting
Unwanted data；Then logarithm value type data acquisition method for normalizing processes, and text data is converted to numeric data
Process.
A kind of Neural Network Data method for digging based on nearest neighbor classifier the most according to claim 1, it is characterised in that:
In described step 3, data later for pretreatment are randomly divided into three data sets: training dataset, test data set and
Confirming data set, described training dataset is in order to train neutral net, and described test data set is in order to the precision of test network, institute
State confirmation data set in order to test network independently, and the extracting rule producing step 6 is estimated.
A kind of Neural Network Data method for digging based on nearest neighbor classifier the most according to claim 1, it is characterised in that:
In described step 4, described neural network type uses three layers of feedforward RBF neural；Described algorithm uses the arest neighbors improved
Cluster learning algorithm trains this neutral net, neutral net input, the number of output node to be determined by the decision-making that power system is concrete
Fixed, the output of neutral net is released regularization output by fuzzy inference system, and the number of hidden layer node is calculated by nearest neighbor classifier
Method determines.
A kind of Neural Network Data method for digging based on nearest neighbor classifier the most according to claim 4, it is characterised in that:
By the adjustment to cluster radius r, cluster centre number i.e. RBF the number of hidden nodes is made to reach optimal, the ginseng to RBF neural
Number and two processes of structure carry out online adaptive adjustment simultaneously, and concrete steps include:
Step1: arranging initial clustering radius is r, by normalized data xkRead in, as the input of RBF neural, and count
Calculate xkWith the minimum euclidean distance of other data existing, obtain distance d of minimummin, and position is designated as p；
Step2: if dmin> r, then cluster numbers adds 1 is that m=m+1, m represent that cluster numbers, current sample send new cluster centre ci, no
Then pth cluster member adds 1, and revises cluster correlation variable；
Step3: all kinds of output vector sums is designated as A (l), represents for adding up the sample belonging to all kinds of with enumerator B (l)
This number, wherein l is classification number, calculating neutral net hidden layer to output layer weight vector Wi=A (i)/B (i)；
Wherein, i represents i-th iteration, and A (i) represents all kinds of output vector sums during i-th iteration, and B (i) represents that i-th is repeatedly
For time Different categories of samples number sum, W (i) represents i-th output layer weight vector；
Step4: obtain according to the output of Regularized RBF Network
Wherein, ciCenter for hidden layer i-th Gaussian function；
Step5: calculation of performance indicators functional expressionWherein, N is current iteration total degree, xkFor N
Dimension input vector, y (xk) andBe respectively target data to and the output of neutral net；
Performance indications are judged: if J0> ε, ε be the threshold value being previously set, then turn Step6, otherwise turn Step7；
Step6: set change step h, then cluster radius is r=r-h, returns Step1；
Step7: cluster terminates.
A kind of Neural Network Data method for digging based on nearest neighbor classifier the most according to claim 5, it is characterised in that:
In Step6, described change step h adjusts the cluster half of RBF neural according to performance index function online adaptive in Step5
Footpath, makes cluster centre number i.e. RBF the number of hidden nodes reach optimal, so that the structure of RBF neural is optimal.
A kind of Neural Network Data method for digging based on nearest neighbor classifier the most according to claim 1, it is characterised in that:
In step 5, step 6, after neutral net network training terminates, use searching algorithm extracting rule, the rule being extracted is used
Test data set and confirmation data set are tested, the correctness that checking Neural Network Data is excavated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610343564.4A CN106022614A (en) | 2016-05-22 | 2016-05-22 | Data mining method of neural network based on nearest neighbor clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610343564.4A CN106022614A (en) | 2016-05-22 | 2016-05-22 | Data mining method of neural network based on nearest neighbor clustering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106022614A true CN106022614A (en) | 2016-10-12 |
Family
ID=57096900
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610343564.4A Pending CN106022614A (en) | 2016-05-22 | 2016-05-22 | Data mining method of neural network based on nearest neighbor clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106022614A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107729990A (en) * | 2017-07-20 | 2018-02-23 | 上海寒武纪信息科技有限公司 | Support the device and method for being used to perform artificial neural network forward operation that discrete data represents |
| CN108255854A (en) * | 2016-12-29 | 2018-07-06 | 南京南瑞继保电气有限公司 | Fuzzy neural network data digging method based on particle group optimizing |
| CN108446712A (en) * | 2018-02-02 | 2018-08-24 | 广东省电信规划设计院有限公司 | ODN nets Intelligent planning method, apparatus and system |
| CN108629087A (en) * | 2018-04-13 | 2018-10-09 | 广州供电局有限公司 | Disturbance event model building method, device and disturbance event recognition methods, device |
| CN108717570A (en) * | 2018-05-23 | 2018-10-30 | 电子科技大学 | A kind of impulsive neural networks parameter quantification method |
| CN108847686A (en) * | 2018-07-02 | 2018-11-20 | 国电南瑞科技股份有限公司 | A kind of photovoltaic DC-to-AC converter failure prediction method |
| CN109639524A (en) * | 2018-12-13 | 2019-04-16 | 国网上海市电力公司 | Communication network data method for visualizing, device and equipment based on volume forecasting |
| CN110457687A (en) * | 2019-07-23 | 2019-11-15 | 福建奇点时空数字科技有限公司 | A kind of data mining and mask method based on complex neural network modeling |
| CN110866777A (en) * | 2019-10-12 | 2020-03-06 | 马征峰 | Tourism big data business platform system |
| CN111723873A (en) * | 2020-06-29 | 2020-09-29 | 南方电网科学研究院有限责任公司 | Power sequence data classification method and device |
| CN112862326A (en) * | 2021-02-19 | 2021-05-28 | 同济大学 | System scheduling optimization method based on big data mining |
| CN112883020A (en) * | 2021-03-02 | 2021-06-01 | 北京虹信万达科技有限公司 | Big data application-based analysis and management system |
| CN113837458A (en) * | 2020-12-29 | 2021-12-24 | 国网江西省电力有限公司南昌供电分公司 | Edge calculation model of power distribution network data, and distribution network equipment defect elimination auxiliary system and defect elimination method based on edge calculation model |
| CN116304931A (en) * | 2023-05-12 | 2023-06-23 | 山东英伟电子技术有限公司 | Electric power data mining method based on big data |
-
2016
- 2016-05-22 CN CN201610343564.4A patent/CN106022614A/en active Pending
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108255854A (en) * | 2016-12-29 | 2018-07-06 | 南京南瑞继保电气有限公司 | Fuzzy neural network data digging method based on particle group optimizing |
| CN107729990A (en) * | 2017-07-20 | 2018-02-23 | 上海寒武纪信息科技有限公司 | Support the device and method for being used to perform artificial neural network forward operation that discrete data represents |
| CN108446712B (en) * | 2018-02-02 | 2020-12-11 | 广东省电信规划设计院有限公司 | ODN network intelligent planning method, device and system |
| CN108446712A (en) * | 2018-02-02 | 2018-08-24 | 广东省电信规划设计院有限公司 | ODN nets Intelligent planning method, apparatus and system |
| CN108629087A (en) * | 2018-04-13 | 2018-10-09 | 广州供电局有限公司 | Disturbance event model building method, device and disturbance event recognition methods, device |
| CN108717570A (en) * | 2018-05-23 | 2018-10-30 | 电子科技大学 | A kind of impulsive neural networks parameter quantification method |
| CN108847686A (en) * | 2018-07-02 | 2018-11-20 | 国电南瑞科技股份有限公司 | A kind of photovoltaic DC-to-AC converter failure prediction method |
| CN108847686B (en) * | 2018-07-02 | 2021-11-30 | 国电南瑞科技股份有限公司 | Photovoltaic inverter fault prediction method |
| CN109639524A (en) * | 2018-12-13 | 2019-04-16 | 国网上海市电力公司 | Communication network data method for visualizing, device and equipment based on volume forecasting |
| CN110457687A (en) * | 2019-07-23 | 2019-11-15 | 福建奇点时空数字科技有限公司 | A kind of data mining and mask method based on complex neural network modeling |
| CN110866777A (en) * | 2019-10-12 | 2020-03-06 | 马征峰 | Tourism big data business platform system |
| CN110866777B (en) * | 2019-10-12 | 2024-03-15 | 马征峰 | Tourism big data business platform system |
| CN111723873A (en) * | 2020-06-29 | 2020-09-29 | 南方电网科学研究院有限责任公司 | Power sequence data classification method and device |
| CN113837458A (en) * | 2020-12-29 | 2021-12-24 | 国网江西省电力有限公司南昌供电分公司 | Edge calculation model of power distribution network data, and distribution network equipment defect elimination auxiliary system and defect elimination method based on edge calculation model |
| CN112862326A (en) * | 2021-02-19 | 2021-05-28 | 同济大学 | System scheduling optimization method based on big data mining |
| CN112883020A (en) * | 2021-03-02 | 2021-06-01 | 北京虹信万达科技有限公司 | Big data application-based analysis and management system |
| CN116304931A (en) * | 2023-05-12 | 2023-06-23 | 山东英伟电子技术有限公司 | Electric power data mining method based on big data |
| CN116304931B (en) * | 2023-05-12 | 2023-08-04 | 山东英伟电子技术有限公司 | Electric power data mining method based on big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106022614A (en) | Data mining method of neural network based on nearest neighbor clustering | |
| Dong et al. | Hourly energy consumption prediction of an office building based on ensemble learning and energy consumption pattern classification | |
| Li et al. | Wind power forecasting considering data privacy protection: A federated deep reinforcement learning approach | |
| Zhao et al. | Correlation-constrained and sparsity-controlled vector autoregressive model for spatio-temporal wind power forecasting | |
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| Cao et al. | Hybrid ensemble deep learning for deterministic and probabilistic low-voltage load forecasting | |
| Zhang et al. | Typical wind power scenario generation for multiple wind farms using conditional improved Wasserstein generative adversarial network | |
| Ghadimi et al. | PSO based fuzzy stochastic long-term model for deployment of distributed energy resources in distribution systems with several objectives | |
| De Silva et al. | A data mining framework for electricity consumption analysis from meter data | |
| CN104022552B (en) | A kind of intelligent detecting method controlled for charging electric vehicle | |
| Rahman et al. | Discretization of continuous attributes through low frequency numerical values and attribute interdependency | |
| CN104573106A (en) | Intelligent urban construction examining and approving method based on case-based reasoning technology | |
| CN104537433A (en) | Sold electricity quantity prediction method based on inventory capacities and business expansion characteristics | |
| CN105005878B (en) | A kind of comprehensive estimation method of strong intelligent grid | |
| Ge et al. | Short-term load forecasting of regional distribution network based on generalized regression neural network optimized by grey wolf optimization algorithm | |
| CN106529732A (en) | Carbon emission efficiency prediction method based on neural network and random frontier analysis | |
| Yi et al. | Intelligent prediction of transmission line project cost based on least squares support vector machine optimized by particle swarm optimization | |
| Lashgari et al. | Transportation energy demand forecasting in Taiwan based on metaheuristic algorithms | |
| Vasanthkumar et al. | Improving energy consumption prediction for residential buildings using Modified Wild Horse Optimization with Deep Learning model | |
| Kong et al. | Real-time pricing method for VPP demand response based on PER-DDPG algorithm | |
| Liu et al. | A novel electricity load forecasting based on probabilistic least absolute shrinkage and selection operator-Quantile regression neural network | |
| Guo et al. | Distribution network topology identification based on gradient boosting decision tree and attribute weighted naive Bayes | |
| Wang et al. | Sustainable investment forecasting of power grids based on the deep restricted Boltzmann machine optimized by the lion algorithm | |
| CN108537581A (en) | Based on the GMDH energy consumption Time Series Forecasting Methods selectively combined and device | |
| Gao et al. | Multi-objective optimal allocation of distributed generation considering the spatiotemporal correlation of wind-photovoltaic-load |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161012 |