CN106570657A - Power grid evaluation index weight determining method - Google Patents
Power grid evaluation index weight determining method Download PDFInfo
- Publication number
- CN106570657A CN106570657A CN201611000688.9A CN201611000688A CN106570657A CN 106570657 A CN106570657 A CN 106570657A CN 201611000688 A CN201611000688 A CN 201611000688A CN 106570657 A CN106570657 A CN 106570657A
- Authority
- CN
- China
- Prior art keywords
- data
- electrical network
- carried out
- boosting
- spark
- 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
- 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
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- 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
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Educational Administration (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Theoretical Computer Science (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a power grid evaluation index weight determining method which is very important to objectively and rationally determine the power grid evaluation index weight under the background of large data. The method provided by the invention comprises the steps that a Spark platform divides all data into a number of sub-data models; the efficiency of data processing is improved by parallel calculation; a data feature extraction method is used to derive valid data information; the acquired valid data information is input into a multivariable L2-Boosting regression model for training and learning, so as to acquire the trained multivariable L2-Boosting regression model; and test data are input into the multivariable L2-Boosting regression model for predicting to determine the power grid evaluation index weight. According to the invention, the L2-Boosting model is combined with a Spark data analysis framework; the power grid evaluation index weight is objectively and accurately determined; and the objectivity of the index weight value can be ensured.
Description
Technical field
The invention belongs to electrical network evaluation criterion weight field, especially a kind of to be based on Spark platforms and multivariate L2-
The electrical network evaluation criterion weight that Boosting regression models combine determines method.
Background technology
At present, China still in the economic development phase, power system infrastructure construction is still faced with huge pressure.Electricity
Network planning is drawn and decision-making plays vital effect during power system infrastructure are built.Although in recent years in electric power
Existing certain achievement in research in terms of the complex optimum and decision method of project, but decision-making is analyzed with these optimization methods
When need clear and definite candidate items.The proposition of current China power network construction project and the determination of Program Priority, subjective factorss rise
Dominance is acted on, and can cause the low problem of utilization rate of relatively low project decision level, electric network reconstruction less pertinence, fund.
Annual improved electrical network yet suffers from more problem, it is impossible to meet the demand of users very well.In order to adapt to national economy
And social development, the specific aim of Power Project decision-making being improved, strengthens economic benefit and social benefit, needs carry out more science and close
The Electric Power Network Planning of reason and decision-making.Electric Power Network Planning is the important base of electrical network decision-making, and electrical network evaluation is the key of Electric Power Network Planning,
Rational decision-making foundation is provided for Electric Power Network Planning, realizes that electric power resource is distributed rationally, promote the quick sane development of electrical network.Cause
This, carries out objective rational evaluation to electrical network particularly significant.
Objective rational evaluation is carried out to electrical network to be included building objective rational electrical network assessment indicator system, and which is evaluated refers to
The reasonable weight of mark distribution, finally carries out overall merit to electrical network.Objective rational electrical network assessment indicator system is built, is effectively to comment
The key of valency electrical network, and determine method using objective index weights, is to ensure that electrical network assessment indicator system is objective, effective pass
Key.Therefore, a set of objective rational electrical network evaluation criterion weight is built to the objective rational evaluation important in inhibiting of electrical network.
For many years, with the development and the continuous lifting of mathematical theory of the present computer technology, Weight Determination
It is evolving, progressively starts to combine with the multi-intelligence algorithm such as technology such as neutral net, fuzzy theory, genetic algorithm, and
Achieve good effect.But in the above-mentioned methods, the effectiveness of each model set up historical data accuracy is good, rule by force with
And data scale it is less when；When in the face of in large scale, complex degree of structure is high, the network system data with height real-time
When, the problems such as above-mentioned model will be made to occur that over-fitting, convergence rate are slow, easily local optimum is absorbed in, so that weight is objective
Reasonability is reduced.
With the quickening of electrical network intelligent Process, a large amount of intelligent electric meters and other testing equipments are deployed in each of electrical network
Critical positions, so that electric power data is presented the growth of " exponential ", the growth of annual data amount is also from GB (gigabyte)
Level rises to TB (terabyte) level.Quickly data increase so that electrical network evaluation index data dimension continuous enlargement, data knot
Structure is increasingly complex so that electrical network evaluation criterion weight determines that the difficulty of work increases.Therefore, the electrical network under big data background is commented
Valency index weights are objective rationally determine it is particularly significant.
The content of the invention
In order to solve the problems, such as accurate, objective, the reasonable determination of electrical network evaluation criterion weight, the present invention is commented with electrical network
Based on valency index big data, there is provided a kind of to be based on Spark platforms and multivariate L2The electricity that-Boosting regression models combine
Net evaluation criterion weight determines method.
The present invention is adopted the following technical scheme that：A kind of electrical network evaluation criterion weight determines method, and which comprises the steps：
Step 1：Total data is carried out into segmentation using Spark platforms and forms multiple subdata models, by parallel computation
To improve data-handling efficiency, valid data information, i.e. multivariate L are drawn using data characteristicses extracting method2- Boosting is returned
The input vector for returning model to need；
Step 2：The valid data information for drawing is input to into multivariate L2- Boosting regression models are trained
Practise, and multivariate L after being trained2- Boosting regression models；
Step 3：Test data brings multivariate L into2It is predicted in-Boosting regression models, determines that electrical network evaluation refers to
Target weight.
Spark platforms described in step 1, Spark platforms are new proposed on the basis of Hadoop MapReduce
For big data analytical framework, possess all advantages that Hadoop MapReduce possess, and Spark is by result of calculation
It is stored directly in internal memory so that operation efficiency is higher.In addition, Spark platforms can realize that machine learning on line, data are handed over
Mutually formula analysis etc., each thread task directly can transfer desired data from internal memory, realize data shared resources, so as to improve fortune
Calculate speed.With the development of intelligent grid so that on very first time line, the high efficiency of big data is excavated and application seems particularly heavy
Will, it is based on the distinctive property of Spark platforms, of the invention by L2- Boosting regression models are mutually tied with Spark data analysiss frameworks
Close, realize the determination of electrical network evaluation criterion weight.
To support that successive ignition computing, Spark platforms provide two key concepts：1. elasticity distribution formula data set
(Resilient Distributed Datasets,RDD).2. shared drive, each RDD is read-only, can pass through other
Batch operation on RDD is created, and in parallel work-flow, variable between task, between task and driver, variable is mutual
Shared.RDD provides four kinds of Important Operators.Wherein, it is input into operator：Data are input into Spark platforms, Jin Erzhuan from space outerpace
Service data block is turned to, and is managed by Block Manager.Transformation operator：Can be by transformation operator (such as fliter
Deng) data are operated, and RDD is converted into into new RDD data sets.Caching operator：Data can be delayed by Cache operators
Deposit and internal memory, it is to avoid re-call the trouble of data to improve operation efficiency.Action operator：By control data corporation by task
Assign in each RDD regions, each region task is performed by parallel computation.
Input operator：In Spark programs, data are input into Spark platforms from space outerpace, such as directly from HDFS receive datas
According to etc.；Data are into data space during Spark operations, and then are converted into service data block, and entered by Block Manager
Row management.
Transformation operator：After Spark data inputs form RDD data sets, can be right by transformation operator (such as fliter etc.)
Data are operated, and RDD is converted into new RDD data sets, are triggered Spark by action operators and are submitted operation to.
Caching operator：Data in Spark can be realized sharing, and its reason is to need reusable data, can be led to
Cache operators are crossed by data buffer storage and internal memory, it is to avoid the trouble of data is re-called to improve operation efficiency.
Action operator：Task assignment is performed by parallel computation in each RDD regions by each region by control data corporation
Task；After program end of run, space when Spark runs is output data to, and is stored in distributed storage space
Or in Scale data acquisition systems.
Multivariate L described in step 22- Boosting regression models.
Assume given n sample observation data { (X, Y) }, wherein response matrix is Y ∈ Rn×q, covariant moment matrix is X ∈ Rn ×p, give coefficient matrix B ∈ Rp×qWith error matrix E ∈ Rn×q, p is independent variable dimension, and to respond dimension, then multiple linear is returned q
Return model be illustrated by following formula：
Y=XB+E
Generally, the estimated value of coefficient matrixSolution be by adopting common least-squares algorithm (ordinaryleast
Square, OLS) obtain, its general expression is：However, with the increase of data volume, data dimension it is continuous
During expansion, the accuracy that OLS coefficient matrixes are estimated greatly will be reduced.Therefore, the present invention adopts improved boosting method
To covariant coefficient matrixIt is reconstructed.
In the structure of improved boosting method, need to redefine the loss function of algorithm and weak learning process, this
Invention is using negative gauss log-likelihood function (negative Gaussian likelihood function, NGLF) as damage
Function, branch's linear least-squares (component-wise linear least squares, CWLLS) are lost as weak study
Process.
The equation formulations of negative gauss log-likelihood function are：
Wherein, Π represents Gauss distribution likelihood function, and nq represents the product of n and q, y(i)For the response value of i-th sample point
(OK)；x(i)For the corresponding covariant moment matrix (OK) of the i-th sample point；The maximal possibility estimation parameter of B and formula OLS estimated result one
Cause, with covariance matrix Q independences.Generally, covariance matrix Q cannot determine in advance, therefore, above formula is adjusted and draws this specially
Loss function needed for sharp is：
Wherein, Φ-1For strengthen covariance matrix, which is used for the estimation of covariance matrix Q, when response matrix dimension q →
During ∞, can use Φ=I (I is unit matrix).
Fitting to covariant coefficient matrix B and study can constantly be strengthened using weak learning process, so as to reach ginseng
The purpose that number is estimated, it is assumed that given covariant moment matrix X and virtual responsive matrix R ∈ Rn×pIt is (R is simultaneously not equal to Y), minimum using branch
Two take advantage of study to carrying out least square regression fitting between X and R (arranging) so that loss function L (B) is minimum, so as to calculate
The estimated value of covariant coefficient of discharge is as follows for the weak learning process of element in covariant coefficient matrix B：
It is in above formula, rightDerivation so thatSolve：
OrderTherefore, multiple linear regression equations can be fixed by following formula
Justice：
Different virtual responsives R and regression equation are estimated repeatedly to be fitted and learnt by weak learning process, so as to can
Progressively to set up the multiple linear regression equations strengthened based on Boosting algorithmsWherein
Multivariate L based on branch's least-squares algorithm2The step of-boosting regressive prediction models, is as follows：
1. initialize.OrderInitialization iterationses are m=1.
2. calculate the deviation of element in current matrixAnd using weak
Learning process calculating parameterAnd then draw estimation function
3. equation updates：Wherein v≤1.
4. m=m+1, set-up procedure is made 2. to be circulated calculating, until meeting program determination condition.
In said process, multivariate L2- Boosting algorithms are the fitting to a certain column element in B in each iteration,
Iteration obtains the estimated value of one group of B every timeSo thatWhen program meets end condition
When,The estimation equation of as required regressive prediction model E [y | x=].Certainly, when covariant BjkTo response matrix R
Middle whole elements have an impact or without during impact, above-mentioned calculating process may simply obtain time figure of merit, i.e. algorithm fitting effect
Fruit will not reach ideal value.At this point it is possible to be fitted being all listed in homogeneous iteration in B, so as to select most in per generation
Excellent fitting is added in estimation equation, until reaching expected precision.However, calculate using parallel iteration causing calculating process exception
Complexity, conventional method can not meet such large-scale calculations, therefore, it is of the invention by multivariate L2- Boosting regression models
Combine with Spark platforms so that the suitability that index weights determine under big data is higher, it is more objective reasonable.
Further, using Spark platforms carry out data processing and extract input vector detailed process be：Electrical network evaluation refers to
Mark related data is stored in memory headroom by RDD conversions after collecting；Data are carried out by cluster Manager
Piecemeal；And then each piecemeal task assignment is carried out into parallel data process in N number of working node；In the data processing task assigned
In, missing data process is carried out using Neville methods；Bad data classification is carried out using subtractive clustering is improved；Differed by calculating
Cause rate carries out feature selection.
Further, the process that implements of Spark platforms is：
The data set collected is packaged in class Datapoint []；
Whole training datasets are carried out into a RDD process, by the parallelize () letter in Spark.Context
Number is translated into RDD collection, referred to as RDD1；
Spark parallel processings will be called to be iterated calculating in RDD1 bufferings and internal memory, and by result of calculation return in
In Feature [] set, RDD conversions, referred to as RDD2 are carried out again；
Data in RDD2 are processed by parallelization by construction input vector weights using map operators, is calculated using reduce
Vector value in each module is summed up by son, calculates the total weight vector of each iterationAnd it is stored in set weights
In [], RDD3 is designated as；
Finally, by selecting best initial weights building L2- Boosting weights determine model, and bring test set into model
Complete whole weights and determine work.
The device have the advantages that being：
(1) present invention is based on electrical network evaluation index data volume is big, baroque data characteristicses, missing data process,
On the basis of bad data classification and feature selection, establish based on Spark platforms and multivariate L2- Boosting regression models
Electrical network evaluation criterion weight determine method, to solve under the background of magnanimity high dimensional data, how effectively utilizes online data
Processing platform, this studies a question to be accurately determined electrical network evaluation criterion weight.
(2) present invention is based on the distinctive property of Spark platforms, by L2- Boosting models and Spark data analysiss frameworks
Combine, realize that electrical network evaluation criterion weight is objective, accurately determine, meet electrical network evaluation criterion weight objectivity and reasonability
Require, ensure that the objectivity of index weightses.
(3) it is of the invention by multivariate L2- Boosting regression models are combined with Spark platforms so that electricity under big data
The suitability that net evaluation criterion weight determines is higher, more objective reasonable.
Description of the drawings
Fig. 1 is the flow chart of the present invention.
Fig. 2-3 is the parallel algorithm flow process of the present invention.
Specific embodiment
With reference to Figure of description, embodiment of the present invention is described in detail.
It is a kind of to be based on Spark platforms and multivariate L2The electrical network evaluation criterion weight that-Boosting models combine determines
Method, as shown in Figure 1, Figure 2 shown in -3, including：
Electrical network evaluation index related data is stored in memory headroom by RDD conversions after collecting, by cluster
Data are carried out piecemeal by manager, and then each piecemeal task assignment is carried out parallel processing in N number of working node (workers).
In the data processing task assigned, the present invention is carried out missing data process using Neville methods, is entered using improvement subtractive clustering
Row bad data is classified and carries out feature selection by calculating inconsistent rate.
(1) missing data based on Neville methods is processed
Make Z={ z1,z2,…,znFor index weights sample data set, n is nodes, and z is interpolation point, makes p=1, q=
1, wherein p ∈ [1, n], q ∈ [1, p] then calculate Lagrange interpolation polynomial：
Lp,q=((z-zp,q)Lp,q-1-(z-zp)Lp-1,q-1)/(zp-zp-q)
Judge | Lp,q-Lp-1,q-1| whether ＜ η set up, if so, then export Lp,q, if not, then continue to calculate.
(2) based on the bad data classification for improving subtractive clustering
Monitoring device can cause to observe data abnormality due to reasons such as external faults, so that most of observations differ
Cause；Additionally, the generation of outside specific event can also cause data exception phenomenon, this disturbs electrical network evaluation to a certain extent and refers to
The change of mark rule, causes index weights to deviate objective circumstances.Accordingly, it would be desirable to the classification being necessary to bad data and place
Reason.The present invention is effectively classified to bad data using fuzzy C-means clustering method is improved, and its step is as follows：
1. clusters number and cluster centre are calculated using subtractive clustering method
2. calculate degree of membership：
Wherein, N is iteration count,
3. calculating target function value：
4. judge Φ (u, p)(N)Whether minima is reached, if reached, output clusters classification number z and cluster centre square
Battle array P；N=N+1 is otherwise made to update cluster centre matrix and recalculate degree of membership.Cluster centre matrix update formula is：
(3) feature selection based on inconsistent rate
Under magnanimity electrical network evaluation index related data, the purpose of feature selection is to distinguish related for evaluation criterion weight
The most strong character subset of property, so that L2The input vector of-Boosting regression models has stronger specific aim, reduces defeated
Enter the redundancy of information, so as to improve the reasonability of index weights determination.Swift nature selection is carried out by inconsistent rate method, is needed
Learn the computational methods of inconsistent rate；Thus, it is supposed that the index weights data collected possess g item features (such as subjective judgment, really
Determine method error etc.), respectively by G1,G2,…,GgValue represent；Assume again that standard class M possesses c classification, with N number of data reality
Example, uses zjiRepresent feature FiCorresponding eigenvalue, uses λiThe class value of M is represented, then data instance can be expressed as [zj,λi], its
Middle zj=[zj1,zj2,…,zjg], then the computing formula of the inconsistent rate of data is：
Wherein, fklTo belong to x in data setkThe number of data instance, x under the character subset pattern of patternkFor data set
Total P feature demarcation interval pattern (k=1,2 ..., p, p≤N).The step of adopting inconsistent rate to carry out feature selection for：
2. it is empty set Γ={ } to initialize optimal feature subset；
2. data set G under the character subset pattern constituted with each residue character in calculating Γ subsets1,G2,…,GgNo
Concordance rate；
3. select feature G corresponding to minimum inconsistent rateiFor optimal characteristics, then update optimal characteristics integrate as Γ=Γ,
Gi}；
4. using order search strategy forward, the inconsistent rate statistical table of character subset is calculated, and ascending is carried out
Arrangement；
5. the character subset for selecting Characteristic Number as far as possible littleJudgeIf so, thenFor selected spy
Levy subset.
In the implementing of Spark frameworks, first the data set collected is packaged in class Datapoint [], should
Class is made up of with vector y vector x, wherein x representative models input vector, and y represents index weights.Whole training datasets are carried out
RDD process, is translated into RDD collection, referred to as RDD1 by the parallelize () function in Spark.Context；
Secondly, Spark parallel processings will be called to be iterated calculating in RDD1 bufferings and internal memory, and by result of calculation return in
In Feature [] set, RDD conversions, referred to as RDD2 are carried out again；Furthermore, using map operators by the data in RDD2
Construction input vector weights are processed by parallelization, the vector value in each module is summed up using reduce operators, calculated
Go out the total weight vector of each iterationAnd be stored in set weights [], it is designated as RDD3；Finally, it is optimum by selecting
Weights are building L2- Boosting index weights determine model, and bring test set into model and complete whole weights and determine work.
This embodiment is only the present invention preferably specific embodiment, but protection scope of the present invention is not limited thereto,
Any those familiar with the art the invention discloses technical scope in, the change or replacement that can be readily occurred in,
Should all be included within the scope of the present invention.Therefore, protection scope of the present invention should be with scope of the claims
It is defined.
Claims (3)
1. a kind of electrical network evaluation criterion weight determines method, and which comprises the steps：
Step 1：Total data is carried out into segmentation using Spark platforms and forms multiple subdata models, carried by parallel computation
High data-handling efficiency, draws valid data information, i.e. multivariate L using data characteristicses extracting method2- Boosting returns mould
The input vector that type needs；
Step 2：The valid data information for drawing is input to into multivariate L2- Boosting regression models are trained study, and obtain
Multivariate L to after training2- Boosting regression models；
Step 3：Test data brings multivariate L into2It is predicted in-Boosting regression models, determines the power of electrical network evaluation index
Weight.
2. electrical network evaluation criterion weight according to claim 1 determines method, it is characterised in that entered using Spark platforms
Row data processing and extract input vector detailed process be：
Electrical network evaluation index related data is stored in memory headroom by RDD conversions after collecting；
Data are carried out by piecemeal by cluster Manager；
And then each piecemeal task assignment is carried out into parallel data process in N number of working node；
In the data processing task assigned, missing data process is carried out using Neville methods；
Bad data classification is carried out using subtractive clustering is improved；
Feature selection is carried out by calculating inconsistent rate.
3. electrical network evaluation criterion weight according to claim 1 and 2 determines method, it is characterised in that the tool of Spark platforms
Body realizes that process is：
The data set collected is packaged in class Datapoint []；
Whole training datasets are carried out into a RDD process, will by the parallelize () function in Spark.Context
Which is converted into RDD collection, referred to as RDD1；
Spark parallel processings will be called to be iterated calculating in RDD1 bufferings and internal memory, and by result of calculation return in
In Feature [] set, RDD conversions, referred to as RDD2 are carried out again；
Data in RDD2 are processed by parallelization by construction input vector weights using map operators, will using reduce operators
Vector value in each module is summed up, and is calculated the total weight vector w of each iteration, and is stored in set weights []
In, it is designated as RDD3；
Finally, by selecting best initial weights building L2- Boosting weights determine model, and bring test set into model and complete
Whole weights determine work.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611000688.9A CN106570657A (en) | 2016-11-14 | 2016-11-14 | Power grid evaluation index weight determining method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611000688.9A CN106570657A (en) | 2016-11-14 | 2016-11-14 | Power grid evaluation index weight determining method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106570657A true CN106570657A (en) | 2017-04-19 |
Family
ID=58542747
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611000688.9A Pending CN106570657A (en) | 2016-11-14 | 2016-11-14 | Power grid evaluation index weight determining method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106570657A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107633030A (en) * | 2017-09-04 | 2018-01-26 | 深圳市华傲数据技术有限公司 | Credit estimation method and device based on data model |
| CN112766640A (en) * | 2020-12-29 | 2021-05-07 | 国网浙江省电力有限公司宁波供电公司 | Full-voltage-grade target grid optimization evaluation method |
| CN112801417A (en) * | 2021-03-16 | 2021-05-14 | 贵州电网有限责任公司 | Optimized model parallel defect material prediction method |
| US11537893B2 (en) | 2019-12-05 | 2022-12-27 | Industrial Technology Research Institute | Method and electronic device for selecting deep neural network hyperparameters |
| WO2023197502A1 (en) * | 2022-04-11 | 2023-10-19 | 广西电网有限责任公司 | Comprehensive power evaluation method and apparatus |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104331628A (en) * | 2014-11-17 | 2015-02-04 | 国网河南省电力公司 | Power grid safety aggregative indicator evaluation method |
| CN105608512A (en) * | 2016-03-24 | 2016-05-25 | 东南大学 | Short-term load forecasting method |
| CN105678398A (en) * | 2015-12-24 | 2016-06-15 | 国家电网公司 | Power load forecasting method based on big data technology, and research and application system based on method |
-
2016
- 2016-11-14 CN CN201611000688.9A patent/CN106570657A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104331628A (en) * | 2014-11-17 | 2015-02-04 | 国网河南省电力公司 | Power grid safety aggregative indicator evaluation method |
| CN105678398A (en) * | 2015-12-24 | 2016-06-15 | 国家电网公司 | Power load forecasting method based on big data technology, and research and application system based on method |
| CN105608512A (en) * | 2016-03-24 | 2016-05-25 | 东南大学 | Short-term load forecasting method |
Non-Patent Citations (1)
| Title |
|---|
| 马天男等: "基于Spark平台和多变量L2-Boosting回归模型的分布式能源系统短期负荷预测", 《电网技术》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107633030A (en) * | 2017-09-04 | 2018-01-26 | 深圳市华傲数据技术有限公司 | Credit estimation method and device based on data model |
| US11537893B2 (en) | 2019-12-05 | 2022-12-27 | Industrial Technology Research Institute | Method and electronic device for selecting deep neural network hyperparameters |
| CN112766640A (en) * | 2020-12-29 | 2021-05-07 | 国网浙江省电力有限公司宁波供电公司 | Full-voltage-grade target grid optimization evaluation method |
| CN112801417A (en) * | 2021-03-16 | 2021-05-14 | 贵州电网有限责任公司 | Optimized model parallel defect material prediction method |
| WO2023197502A1 (en) * | 2022-04-11 | 2023-10-19 | 广西电网有限责任公司 | Comprehensive power evaluation method and apparatus |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106570657A (en) | Power grid evaluation index weight determining method | |
| CN105701596A (en) | Method for lean distribution network emergency maintenance and management system based on big data technology | |
| CN110348713A (en) | A kind of platform area line loss calculation method based on association analysis and data mining | |
| CN104809658B (en) | A kind of rapid analysis method of low-voltage distribution network taiwan area line loss | |
| CN106951695A (en) | Plant equipment remaining life computational methods and system under multi-state | |
| JP6784780B2 (en) | How to build a probabilistic model for large-scale renewable energy data | |
| CN105260738A (en) | Method and system for detecting change of high-resolution remote sensing image based on active learning | |
| CN106503731A (en) | A kind of based on conditional mutual information and the unsupervised feature selection approach of K means | |
| CN110072183B (en) | Passive positioning fingerprint database construction method based on crowd sensing | |
| CN113343427B (en) | Structural topology configuration prediction method based on convolutional neural network | |
| CN104461772A (en) | Method for recovering missed data | |
| CN108062566A (en) | A kind of intelligent integrated flexible measurement method based on the potential feature extraction of multinuclear | |
| CN111525587A (en) | Reactive load situation-based power grid reactive voltage control method and system | |
| CN112241606A (en) | Cooperative decision-making method for operation and maintenance of ship intelligent equipment based on CPS decision-making module | |
| CN103353895A (en) | Pre-processing method of power distribution network line loss data | |
| CN104778368A (en) | Pareto set individual ranking method aiming at high-dimensional multi-objective optimization problem | |
| CN110275868A (en) | A kind of multi-modal pretreated method of manufaturing data in intelligent plant | |
| CN114597970A (en) | Active power distribution network partitioning method based on graph convolution network | |
| CN116501444B (en) | Abnormal cloud edge collaborative monitoring and recovering system and method for virtual machine of intelligent network-connected automobile domain controller | |
| CN104062904A (en) | Large chemical process distributed modeling method based on CCA-PLS | |
| CN114006370A (en) | Power system transient stability analysis and evaluation method and system | |
| CN108763707A (en) | The approximate Analysis of structural reliability method of second order saddle point is based under Hybrid parameter matrix | |
| CN117156442A (en) | Cloud data security protection method and system based on 5G network | |
| CN108549696A (en) | A kind of time series data similarity query method calculated based on memory | |
| CN111405605A (en) | Wireless network interruption detection method based on self-organizing mapping |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170419 |