CN109034453A - A kind of Short-Term Load Forecasting Method based on multiple labeling neural network - Google Patents
A kind of Short-Term Load Forecasting Method based on multiple labeling neural network Download PDFInfo
- Publication number
- CN109034453A CN109034453A CN201810642238.2A CN201810642238A CN109034453A CN 109034453 A CN109034453 A CN 109034453A CN 201810642238 A CN201810642238 A CN 201810642238A CN 109034453 A CN109034453 A CN 109034453A
- Authority
- CN
- China
- Prior art keywords
- data
- load
- cluster
- multiple labeling
- prediction
- 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
- 238000002372 labelling Methods 0.000 title claims abstract description 36
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 16
- 238000013277 forecasting method Methods 0.000 title claims abstract description 15
- 238000000034 method Methods 0.000 claims abstract description 23
- 238000012545 processing Methods 0.000 claims abstract description 12
- 238000003064 k means clustering Methods 0.000 claims abstract description 9
- 238000012549 training Methods 0.000 claims description 16
- 241001269238 Data Species 0.000 claims description 8
- 238000013500 data storage Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 239000012141 concentrate Substances 0.000 claims description 3
- 235000006508 Nelumbo nucifera Nutrition 0.000 claims 1
- 240000002853 Nelumbo nucifera Species 0.000 claims 1
- 235000006510 Nelumbo pentapetala Nutrition 0.000 claims 1
- 230000005611 electricity Effects 0.000 description 3
- 241000209094 Oryza Species 0.000 description 2
- 235000007164 Oryza sativa Nutrition 0.000 description 2
- 235000009566 rice Nutrition 0.000 description 2
- 238000012731 temporal analysis Methods 0.000 description 2
- 238000000700 time series analysis Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000003245 coal Substances 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 230000011218 segmentation 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
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Marketing (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Quality & Reliability (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Entrepreneurship & Innovation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Operations Research (AREA)
- Evolutionary Biology (AREA)
- Development Economics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Present invention discloses a kind of Short-Term Load Forecasting Methods based on multiple labeling neural network, firstly, historical data is carried out the processing such as segment encoding and standardization；Secondly, initial data is divided into k cluster using k-means clustering algorithm；Again, learn the similarity between load to be predicted out and k cluster using the multiple labeling algorithm based on K-NN；Finally, BPNN model is respectively trained using each cluster data, the prediction result of load is obtained.This method is experimentally confirmed all to be very significantly improved in terms of precision of prediction and runing time.
Description
Technical field
The present invention relates to a kind of Short-Term Load Forecasting Methods based on multiple labeling neural network, and can be related to micro-capacitance sensor has
Effect operation technical field.
Background technique
In recent years, environmental problem constantly highlights, and requirement of the people to air quality is higher and higher, and traditional coal electricity then can
Serious pollution is caused to atmosphere, to accelerate the development of new energy, wherein micro power network as it is a kind of collect luminous energy, wind energy in
One has the characteristics that miniaturization, distributed operation, is widely used in generation of electricity by new energy.In micro-capacitance sensor, in order to
Control is coordinated and optimized between the normal operation of guarantee system and each component, it is very heavy for predicting the operating status of modules in advance
It wants.Wherein, load forecast is just important a link.
People have had long time, domestic and foreign scholars, expert and engineering people about the exploration of the method for load prediction
Member has done a large amount of work, proposes many prediction models and relevant forecasting system.Including time series analysis, ash
Color theoretical method and neural network method etc..Time series analysis method needs to establish a kind of effective mathematical model, prediction essence
Degree is not general high；Grey forecasting model is theoretically suitable for any nonlinear-load prediction, and the differential equation is suitable for having
It predicts the load prediction of growth trend, but is difficult to improve the precision of gray prediction.
As the complexity of electric power pool is higher and higher, the continuous expansion of data scale, neural network calculates (such as BP mind
Through
Network etc.) it is also increasingly used in micro-capacitance sensor in load forecast.
Summary of the invention
The object of the invention is to propose a kind of based on multiple labeling mind to solve the above-mentioned problems in the prior art
Short-Term Load Forecasting Method through network.
A kind of the purpose of the invention will be achieved through the following technical solutions: short-term electricity based on multiple labeling neural network
Power load forecasting method, method includes the following steps:
S1: data prediction step；
Data are acquired, the storage of collected data are handled into database S, by the number after pretreatment
Database S is arrived according to storage*；
S2:k-means clustering algorithm is by database S*In data clusters, historical data is gathered with k-means algorithm
At 8 classes, i.e. k=8 in k-means algorithm, k represents cluster number, and cluster centre randomly selects, similitude Euclidean between data
Distance metric obtains being indicated in 8 clusters with C, C={ c1, c2, c3, c4, c5, c6, c6, c7, c8, wherein c1-c8Respectively represent
1 to the 8th cluster；
S3: load to be predicted is obtained with the multiple labeling algorithm based on K-NN and clusters the similarity of C；:
S31: the K=100 in setting K-NN；
S32: assuming that the non-load attribute of load F to be predicted is F={ T ', L ', D ' }, by F and training data T, L,
D } dimension does similarity calculation, using euclidean distance metric, Candidate Set of the smallest 100 data of selected distance as F；
S33: multiple labeling algorithm calculates 100 datas in the Candidate Set of F in c1-c8In how many each item, to obtain F
With cluster c1-c8Similarity；The Candidate Set of F is respectively in cluster C={ c1, c2, c3, c4, c5, c6, c6, c7, c8In number be N
={ n1,n2,n3,n4,n5,n6,n7,n8Assume N={ 10,20,30,11,5,1,0,0 }, it can be obtained by multiple labeling algorithm each poly-
Class and the similarity of load F to be predicted are expressed as with SI Namely SI=0.19,0.2,0.3,0.25,0.05,0.01,
0,0 }；
S4:BPNN algorithm trains prediction model；The load and cluster set to be predicted obtained according to S3 step multiple labeling algorithm
The similarity set SI for closing C, when si is not equal to zero, which is used to train BPNN model；{ the si known in S3 step1,
si2, si3, si4, si5, si6It is the cluster of non-zero, then just gathering { c with cluster1, c2, c3, c4, c5, c6, c6Be respectively trained
Respective BPNN prediction model MODEL={ M1, M2, M3, M4, M5, M6, and respective predicted load is provided, with set
RP expression, RP={ Mp1, Mp2, Mp3, Mp4, Mp5, Mp6}；
S5: prediction result is obtained；By S4 step obtained each cluster data train Lai BPNN model prediction it is negative
Charge values RP={ Mp1, Mp2, Mp3, Mp4, Mp5, Mp6, data preprocessing phase load data be normalized into [- 1,1] it
Between, RP is reduced into true value, obtains RP '={ Mp '1, Mp '2, Mp '3, Mp '4, Mp '5, Mp '6, then, prediction result is multiplied
Final prediction result namely RP ' × SI is obtained with respective similarityT。
Preferably, data needed for acquiring electric power data load prediction, which includes power load related data, i.e., electric
Press data (U), current data (I), temperature data (T), rainfall product data (L), date type data (D) or power data (P).
Preferably, in S1 step, the data prediction the following steps are included::
S11: missing values processing, to there are the data of a small amount of missing values to be filled using correlation method, to missing values mistake
More data are directly deleted；
S12: data normalization will be every using method for normalizing formula (1) due to the having differences property of data of different dimensions
Under the data unification to same scale of a dimension, rice was restored prediction result using formula (2) for load data；Normalizing
Change method formula is (1), and load data formula is (2):
X=(xMAX-xMIN)/2+(xMAX-xMIN)/2 (2)
In formula, xMAXAnd xMINRepresent maximum value and minimum value that training sample concentrates each dimension.
S13: symbol data coding and temperature data segment encoding will be to differences if date type data are text datas
Date type do digital coding, influence of the temperature to load in certain section be it is comparable, the temperature in this section is united
One coding, by the data storage after data prediction to database S*In.
Preferably, in database S*In, every data attribute includes: temperature data (T), rainfall product data (L), date class
Type data (D) and power data (P).
Preferably, wherein power data (P) is one day load data, and time granularity is a hour namely P=
[p1, p2..., pi..., p24], wherein piAs soon as i-th hour load value, every training data have 27 dimensions in representing day,
Middle power data (P) is that load namely the dimensional information of day to be predicted to be predicted only have { T, L, D }.
Preferably, in S13 step, segment processing is carried out to temperature, when temperature is in some optimum range to load
Influence it is smaller, be affected when temperature is raised and lowered to a certain extent to load.
The advantages of technical solution of the present invention, is mainly reflected in: this method, first by historical data carry out segment encoding and
The processing such as standardization；Secondly, initial data is divided into k cluster using k-means clustering algorithm；Again, using based on K-NN
Multiple labeling algorithm learn load to be predicted out and k cluster between similarity；Finally, being instructed respectively using each cluster data
Practice BPNN model, obtains the prediction result of load.This method is experimentally confirmed all to be obtained in terms of precision of prediction and runing time
To very big improvement.
Detailed description of the invention
Fig. 1 is a kind of algorithm flow of Short-Term Load Forecasting Method based on multiple labeling neural network of the invention
Figure.
Specific embodiment
The purpose of the present invention, advantage and feature, by by the non-limitative illustration of preferred embodiment below carry out diagram and
It explains.These embodiments are only the prominent examples using technical solution of the present invention, it is all take equivalent replacement or equivalent transformation and
The technical solution of formation, all falls within the scope of protection of present invention.
Present invention discloses a kind of Short-Term Load Forecasting Method based on multiple labeling neural network, this method firstly,
Historical data is subjected to the processing such as segment encoding and standardization；Secondly, initial data is divided into using k-means clustering algorithm
K cluster；Again, learn the similarity between load to be predicted out and k cluster using the multiple labeling algorithm based on K-NN；Most
Afterwards, BPNN model is respectively trained using each cluster data, obtains the prediction result of load.
As shown in Figure 1, method includes the following steps:
S1: data prediction step；
Data are acquired, the storage of collected data are handled into database S, by the number after pretreatment
Database S is arrived according to storage*；
S2:k-means clustering algorithm is by database S*In data clusters, historical data is gathered with k-means algorithm
At 8 classes, i.e. k=8 in k-means algorithm, k represents cluster number, and cluster centre randomly selects, similitude Europe between data
Family name's distance metric obtains being indicated in 8 clusters with C, C={ c1, c2, c3, c4, c5, c6, c6, c7, c8, wherein c1-c8It respectively represents
1st to the 8th cluster；
S3: load to be predicted is obtained with the multiple labeling algorithm based on K-NN and clusters the similarity of C；:
S31: the K=100 in setting K-NN；
S32: assuming that the non-load attribute of load F to be predicted is F={ T ', L ', D ' }, by F and training data T, L,
D } dimension does similarity calculation, using euclidean distance metric, Candidate Set of the smallest 100 data of selected distance as F；
S33: multiple labeling algorithm calculates 100 datas in the Candidate Set of F in c1-c8In how many each item, to obtain F
With cluster c1-c8Similarity；The Candidate Set of F is respectively in cluster C={ c1, c2, c3, c4, c5, c6, c6, c7, c8In number be
N={ n1, n2, n3, n4, n5, n6, n7, n8, it is assumed that N={ 10,20,30,11,5,1,0,0 } can be obtained each by multiple labeling algorithm
It clusters and is expressed as with the similarity of load F to be predicted with SI Namely SI=0.19,0.2,0.3,0.25,0.05,0.01,
0,0 }；
S4:BPNN algorithm trains prediction model；The load and cluster set to be predicted obtained according to S3 step multiple labeling algorithm
The similarity set SI for closing C, when si is not equal to zero, which is used to train BPNN model；From in S3 step
{si1, si2, si3, si4, si5, si6It is the cluster of non-zero, then just gathering { c with cluster1, c2, c3, c4, c5, c6, c6Respectively
The respective BPNN prediction model MODEL={ M of training1, M2, M3, M4, M5, M6, and respective predicted load is provided, with collection
Closing RP indicates, RP={ Mp1, Mp2, Mp3, Mp4, Mp5, Mp6}；
S5: prediction result is obtained；By S4 step obtained each cluster data train Lai BPNN model prediction it is negative
Charge values RP={ Mp1, Mp2, Mp3, Mp4, Mp5, Mp6, it is normalized between [- 1,1] in data preprocessing phase load data,
RP is reduced into true value, obtains RP '={ Mp '1, Mp '2, Mp '3, Mp '4, Mp '5, Mp '6, then, by prediction result multiplied by each
From similarity obtain final prediction result namely RP ' × SIT。
Data needed for acquiring electric power data load prediction: 1) power load related data (voltage data (U), electric current number
According to (I), temperature data (T), rainfall product data (L), date type data (D), power data (P)), by collected data
It stores in database S.
Data prediction: 1) missing values are handled, to there are the data of a small amount of missing values to be filled using correlation method；It is right
The excessive data of missing values are directly deleted.2) data normalization utilizes normalization due to the having differences property of data of different dimensions
Method will be under the data unification to same scale of each dimension.3) symbol data and segment data coding, such as date type
Data are text datas, therefore digital coding, influence of the temperature to load in certain section are done to different date types
It is comparable, it is possible to the temperature Unified coding in this section.By the data storage after data prediction to database
S*In.
Algoritic module, 1) training data k-means clustering algorithm progress clustering processing, k cluster is divided the data into,
2) using be based on k-NN (k- neighbour) multiple labeling algorithm, obtain sample to be predicted with k cluster similitude probability, 3) by
Training set of the highest cluster of similitude as sample to be predicted, 4) use BP (back propagation) neural metwork training
Model, and calculate load value to be predicted.5) output load prediction result.
S1: data prediction；
1) missing values are handled, to there are the data of a small amount of missing values to be filled using correlation method；It is excessive to missing values
Data directly delete.Here data needed for data refer to acquisition electric power data load prediction: 1) power load related data
(temperature data (T), rainfall product data (L), date type data (D), power data (P)), collected data storage is arrived
In database S.
2) data normalization will be every using method for normalizing (formula (1)) due to the having differences property of data of different dimensions
Under the data unification to same scale of a dimension.Rice was restored prediction result using formula (2) for load data.
X=(xMAX-xMIN)/2+(xMAX-xMIN)/2 (2)
In formula, xMAXAnd xMINRepresent maximum value and minimum value that training sample concentrates each dimension.
Step 3:
1) symbol data encodes, for example date type data are text datas, therefore to do number to different date types
Word coding, as shown in table 1:
1 date data of table coding
| Date type divides | Date type quantifies value |
| Working day | 0.0 |
| Saturday | 0.2 |
| Sunday | 0.3 |
| Working day and the official holiday of being | 0.5 |
| Saturday and the official holiday of being | 0.6 |
| Sunday and the official holiday of being | 0.7 |
| Working day and be foreign red-letter day (Christmas and Valentine's Day etc.) | 0.8 |
| Saturday and be foreign red-letter day (Christmas and Valentine's Day etc.) | 0.9 |
| Sunday and be foreign red-letter day (Christmas and Valentine's Day etc.) | 1 |
2) temperature data segment encoding is handled:
The conversion of temperature, experiments have shown that the influence to load is similar when temperature changes in a certain range, so to temperature
Segment processing can be carried out, and the influence when temperature is in some optimum range to load is smaller, when temperature increases or drops
As low as to a certain degree when be affected to load, therefore segmentation and quantification treatment are carried out to temperature, processing is as shown in table 2:
The processing of 2 temperature data of table
By the data storage after pretreatment to database S*.Database S*In, every data attribute includes: temperature data
(T), rainfall product data (L), date type data (D) and power data (P), wherein power data (P) is one day load number
According to time granularity is a hour namely P=[p1, p2..., pi..., p24], wherein piRepresent in one day i-th hour negative
Charge values, in this way, every training data just has 27 dimensions, wherein power data (P) is load namely the dimension of day to be predicted to be predicted
It spends information and there was only { T, L, D }.
S2::k-means clustering algorithm clusters historical data；
With k-means algorithm historical data be polymerized to 8 classes (namely k=8 in k-means algorithm, k represent cluster
Number), cluster centre randomly selects, similitude euclidean distance metric between data.It obtains being indicated in 8 clusters with C, C=
{c1, c2, c3, c4, c5, c6, c6, c7, c8, wherein c1-c8Respectively represent the 1st to the 8th cluster.
S3: load to be predicted is obtained with the multiple labeling algorithm based on K-NN and clusters the similarity of C.
1) K=100 in K-NN is set；
2) the non-load attribute for assuming load F to be predicted is F={ T ', L ', D ' }, by { T, L, the D } in F and training data
Dimension is similarity calculation (using euclidean distance metric), candidate of the smallest by 100 (K=100) data of selected distance as F
Collection；
3) multiple labeling algorithm calculates 100 datas in the Candidate Set of F in c1-c8In how many each item, thus obtain F with
Cluster c1-c8Similarity.The Candidate Set of F is respectively in cluster C={ c1, c2, c3, c4, c5, c6, c6, c7, c8In number be N=
{n1, n2, n3, n4, n5, n6, n7, n8, it is assumed that N={ 10,20,30,11,5,1,0,0 } can be obtained each poly- by multiple labeling algorithm
Class and the similarity of load F to be predicted are expressed as with SI Namely SI=0.19,0.2,0.3,0.25,0.05,0.01,
0,0 }.
S4:BPNN algorithm trains prediction model.The load and cluster set to be predicted obtained according to S3 step multiple labeling algorithm
The similarity set SI for closing C, when si is not equal to zero, which is used to train BPNN model.From in S3 step
{si1, si2, si3, si4, si5, si6It is the cluster of non-zero, then just gathering { c with cluster1, c2, c3, c4, c5, c6, c6Respectively
The respective BPNN prediction model MODEL={ M of training1, M2, M3, M4, M5, M6, and respective predicted load is provided, with collection
Closing RP indicates, RP={ Mp1, Mp2, Mp3, Mp4, Mp5, Mp6}。
S5: prediction result is obtained；By S4 step obtained each cluster data train Lai BPNN model prediction it is negative
Charge values RP={ Mp1, Mp2, Mp3, Mp4, Mp5, Mp6, because of the needs of BPNN model, in data preprocessing phase load data
It is normalized between [- 1,1], so RP is reduced into true value with formula (2), obtains RP '={ Mp '1, Mp '2, Mp '3,
Mp′4, Mp '5, Mp '6}.Then, by prediction result multiplied by respective similarity obtain final prediction result namely RP ' ×
SIT。
A kind of Short-Term Load Forecasting Method based on multiple labeling neural network of invention, firstly, by history
Data carry out segment encoding and standardization；Secondly, initial data is divided into k cluster using k-means clustering algorithm；
Again, learn the similarity between load to be predicted out and k cluster using the multiple labeling algorithm based on K-NN；Finally, utilizing
BPNN model is respectively trained in each cluster data, obtains the prediction result of load.
Still there are many embodiment, all technical sides formed using equivalents or equivalent transformation by the present invention
Case is within the scope of the present invention.
Claims (6)
1. a kind of Short-Term Load Forecasting Method based on multiple labeling neural network, it is characterised in that: this method includes following
Step:
S1: data prediction step；
Data are acquired, the storage of collected data is handled into database S, the data after pretreatment are deposited
Store up database S*；
S2: training data k-means clustering algorithm is carried out clustering processing, divides the data into k cluster；K-means cluster is calculated
Method is by database S*In data clusters, historical data is polymerized to 8 classes, i.e. k=in k-means algorithm with k-means algorithm
8, k represent cluster number, and cluster centre randomly selects, similitude euclidean distance metric between data, obtain using in 8 clusters
C expression, C={ c1, c2, c3, c4, c5, c6, c6, c7, c8, wherein c1-c8Respectively represent the 1st to the 8th cluster；
S3: the similarity of load to be predicted with k cluster C is obtained with the multiple labeling algorithm based on K-NN；
S31: the K=100 in setting K-NN；
S32: assuming that the non-load attribute of load F to be predicted is F={ T ', L ', D ' }, { T, L, the D } in F and training data is tieed up
Degree does similarity calculation, using euclidean distance metric, Candidate Set of the smallest 100 data of selected distance as F；
S33: multiple labeling algorithm calculates 100 datas in the Candidate Set of F in c1-c8In how many each item, to obtain F and poly-
Class c1-c8Similarity；The Candidate Set of F is respectively in cluster C={ c1, c2, c3, c4, c5, c6, c6, c7, c8In number be N=
{n1, n2, n3, n4, n5, n6, n7, n8, it is assumed that N={ 10,20,30,11,5,1,0,0 } can be obtained each poly- by multiple labeling algorithm
Class and the similarity of load F to be predicted are expressed as with SI Namely SI=0.19,0.2,0.3,0.25,0.05,
0.01,0,0 }；
S4:BPNN algorithm trains prediction model；The load to be predicted obtained according to S3 step multiple labeling algorithm is with cluster set C's
Similarity set SI, when si is not equal to zero, which is used to train BPNN model；{ the si known in S3 step1, si2,
si3, si4, si5, si6It is the cluster of non-zero, then just gathering { c with cluster1, c2, c3, c4, c5, c6, c6Be respectively trained it is respective
BPNN prediction model MODEL={ M1, M2, M3, M4, M5, M6, and respective predicted load is provided, it is indicated with set RP, RP
={ Mp1, Mp2, Mp3, Mp4, Mp5, Mp6}；
S5: prediction result is obtained；By S4 step obtained each cluster data train Lai BPNN model prediction load value
RP={ Mp1, Mp2, Mp3, Mp4, Mp5, Mp6, it is normalized between [- 1,1] in data preprocessing phase load data, by RP
It is reduced into true value, obtains RP '={ Mp '1, Mp '2, Mp '3, Mp '4, Mp '5, Mp '6, then, by prediction result multiplied by respective
Similarity obtains final prediction result namely RP ' × SIT。
2. a kind of Short-Term Load Forecasting Method based on multiple labeling neural network according to claim 1, feature
Be: acquisition electric power data load prediction needed for data, which includes power load related data, i.e., voltage data (U),
Current data (I), temperature data (T), rainfall product data (L), date type data (D) or power data (P).
3. a kind of Short-Term Load Forecasting Method based on multiple labeling neural network according to claim 1, feature
Be: in S1 step, the data prediction the following steps are included::
S11: missing values processing；
To there are the data of a small amount of missing values to be filled using correlation method, the data excessive to missing values are directly deleted；
S12: data normalization；
Due to the having differences property of data of different dimensions, the data of each dimension are uniformly arrived using method for normalizing formula (1)
Under same scale, prediction result reduction is come using formula (2) for load data；Method for normalizing formula is (1), is born
Lotus data formula is (2):
X=(xMAX-xMIN)/2+(xMAX-xMIN)/2 (2)
In formula, xMAXAnd xMINRepresent maximum value and minimum value that training sample concentrates each dimension.
S13: symbol data coding and temperature data segment encoding；
If date type data are text datas, digital coding, the temperature pair in certain section are done to different date types
The influence of load be it is comparable, the temperature Unified coding in this section, by the data storage after data prediction to data
Library S*In.
4. a kind of Short-Term Load Forecasting Method based on multiple labeling neural network according to claim 1, feature
It is: in database S*In, every data attribute includes: temperature data (T), rainfall product data (L), date type data (D)
With power data (P).
5. a kind of Short-Term Load Forecasting Method based on multiple labeling neural network according to claim 4, feature
Be: wherein power data (P) is one day load data, and time granularity is a hour namely P=[p1, p2...,
pi..., p24], wherein piAs soon as i-th hour load value, every training data have 27 dimensions in representing day, wherein power data
It (P) is that load namely the dimensional information of day to be predicted to be predicted only have { T, L, D }.
6. a kind of Short-Term Load Forecasting Method based on multiple labeling neural network according to claim 3, feature
It is: in S13 step, segment processing is carried out to temperature, the influence when temperature is in some optimum range to load is smaller,
It is affected when temperature is raised and lowered to a certain extent to load.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810642238.2A CN109034453A (en) | 2018-06-21 | 2018-06-21 | A kind of Short-Term Load Forecasting Method based on multiple labeling neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810642238.2A CN109034453A (en) | 2018-06-21 | 2018-06-21 | A kind of Short-Term Load Forecasting Method based on multiple labeling neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109034453A true CN109034453A (en) | 2018-12-18 |
Family
ID=64610141
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810642238.2A Pending CN109034453A (en) | 2018-06-21 | 2018-06-21 | A kind of Short-Term Load Forecasting Method based on multiple labeling neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109034453A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109508835A (en) * | 2019-01-01 | 2019-03-22 | 中南大学 | A kind of wisdom Power Network Short-Term Electric Load Forecasting method of integrated environment feedback |
| CN113673741A (en) * | 2021-05-25 | 2021-11-19 | 石化盈科信息技术有限责任公司 | Steady state real-time optimization method and device, storage medium and electronic equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106447133A (en) * | 2016-11-03 | 2017-02-22 | 上海交通大学 | Short-term electric load prediction method based on deep self-encoding network |
| CN107590567A (en) * | 2017-09-13 | 2018-01-16 | 南京航空航天大学 | Recurrent neural network short-term load prediction method based on information entropy clustering and attention mechanism |
-
2018
- 2018-06-21 CN CN201810642238.2A patent/CN109034453A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106447133A (en) * | 2016-11-03 | 2017-02-22 | 上海交通大学 | Short-term electric load prediction method based on deep self-encoding network |
| CN107590567A (en) * | 2017-09-13 | 2018-01-16 | 南京航空航天大学 | Recurrent neural network short-term load prediction method based on information entropy clustering and attention mechanism |
Non-Patent Citations (1)
| Title |
|---|
| XIAOKUI SUN ETC: "Short-Term Load Forecasting Model Based on Multi-label and BPNN", 《ADVANCED COMPUTATIONAL METHODS IN LIFE SYSTEM MODELING AND SIMULATION. SPRINGER, SINGAPORE》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109508835A (en) * | 2019-01-01 | 2019-03-22 | 中南大学 | A kind of wisdom Power Network Short-Term Electric Load Forecasting method of integrated environment feedback |
| CN109508835B (en) * | 2019-01-01 | 2020-11-24 | 中南大学 | Smart power grid short-term power load prediction method integrating environmental feedback |
| CN113673741A (en) * | 2021-05-25 | 2021-11-19 | 石化盈科信息技术有限责任公司 | Steady state real-time optimization method and device, storage medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| CN110909912B (en) | Park electric power system net load combination prediction method based on self-adaptive error feedback | |
| CN110348624B (en) | Sand storm grade prediction method based on Stacking integration strategy | |
| CN104598986B (en) | Methods of electric load forecasting based on big data | |
| CN110490385A (en) | The unified prediction of electric load and thermic load in a kind of integrated energy system | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN112465256A (en) | Building power consumption prediction method and system based on Stacking model fusion | |
| CN115099511A (en) | Photovoltaic power probability estimation method and system based on optimized copula | |
| Akpinar et al. | Forecasting natural gas consumption with hybrid neural networks—Artificial bee colony | |
| Čurčić et al. | Gaining insights into dwelling characteristics using machine learning for policy making on nearly zero-energy buildings with the use of smart meter and weather data | |
| Agga et al. | Short-term load forecasting: based on hybrid CNN-LSTM neural network | |
| CN109034453A (en) | A kind of Short-Term Load Forecasting Method based on multiple labeling neural network | |
| CN117494906A (en) | Natural gas daily load prediction method based on multivariate time series | |
| CN109214610A (en) | A kind of saturation Methods of electric load forecasting based on shot and long term Memory Neural Networks | |
| Ajagunsegun et al. | Machine learning-based system for managing energy efficiency of public buildings: An approach towards smart cities | |
| Guo et al. | Wind speed forecasting of genetic neural model based on rough set theory | |
| CN117391257A (en) | Road congestion condition prediction method and device | |
| CN117200352A (en) | Photovoltaic power generation regulation and control method and system based on cloud edge fusion | |
| CN116862043A (en) | Load prediction method and system based on multilayer feedforward network | |
| CN110659775A (en) | LSTM-based improved electric power short-time load prediction algorithm | |
| CN114997475B (en) | Kmeans-based fusion model photovoltaic power generation short-term prediction method | |
| CN116485582A (en) | Heat supply optimization regulation and control method and device based on deep learning | |
| Thirunavukkarasu et al. | Very Short-Term Solar Irradiance Forecasting using Multilayered Long-Short Term Memory | |
| Kayakuş | The Estimation of Turkey's Energy Demand Through Artificial Neural Networks and Support Vector Regression Methods | |
| CN113807027A (en) | Health state evaluation model, method and system for wind turbine generator |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181218 |
|
| RJ01 | Rejection of invention patent application after publication |