CN108596242A - Power grid meteorology load forecasting method based on wavelet neural network and support vector machines - Google Patents
Power grid meteorology load forecasting method based on wavelet neural network and support vector machines Download PDFInfo
- Publication number
- CN108596242A CN108596242A CN201810359331.2A CN201810359331A CN108596242A CN 108596242 A CN108596242 A CN 108596242A CN 201810359331 A CN201810359331 A CN 201810359331A CN 108596242 A CN108596242 A CN 108596242A
- Authority
- CN
- China
- Prior art keywords
- load
- data
- day
- prediction
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 54
- 238000012706 support-vector machine Methods 0.000 title claims abstract description 35
- 238000013277 forecasting method Methods 0.000 title claims abstract description 5
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 21
- 238000000513 principal component analysis Methods 0.000 claims abstract description 15
- 241001269238 Data Species 0.000 claims abstract description 6
- 230000002354 daily effect Effects 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 21
- 238000000034 method Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 10
- 238000009825 accumulation Methods 0.000 claims description 6
- 230000003203 everyday effect Effects 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 6
- 240000002853 Nelumbo nucifera Species 0.000 claims description 4
- 235000006508 Nelumbo nucifera Nutrition 0.000 claims description 4
- 235000006510 Nelumbo pentapetala Nutrition 0.000 claims description 4
- PEDCQBHIVMGVHV-UHFFFAOYSA-N Glycerine Chemical compound OCC(O)CO PEDCQBHIVMGVHV-UHFFFAOYSA-N 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 230000001537 neural effect Effects 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 abstract description 5
- 238000004364 calculation method Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 241001465754 Metazoa Species 0.000 description 1
- 244000131316 Panax pseudoginseng Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012067 mathematical method Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
-
- 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
- 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)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Tourism & Hospitality (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Quality & Reliability (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Entrepreneurship & Innovation (AREA)
- Water Supply & Treatment (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Development Economics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of power grid meteorology load forecasting method based on wavelet neural network and support vector machines.Initial data is extracted from electric power data, is combined to initial data reduction using clustering algorithm and Principal Component Analysis, standardization is remake, and the data after input standardization are trained to support vector machines, obtain load forecasting model and the first prediction load data;Neural network and the second prediction load data after data after input standardization are trained to wavelet neural network;Load forecasting model is obtained according to the one the second prediction load datas.The present invention fully considers influence of the meteorological data to load fluctuation, fully consider the scale of data, pass through clustering algorithm and Principal Component Analysis, reduce load and meteorological data amount simultaneously, the prediction model of proposition, the precision of prediction that ensure that support vector machines and wavelet-neural network model, improves precision of prediction, solves the problems, such as that the precision of prediction for considering meteorologic factor and bringing is low.
Description
Technical field
The present invention relates to a kind of network load data predication methods, and wavelet neural network is based on more particularly, to a kind of
With the power grid meteorology load data prediction technique of support vector machines.
Background technology
Electric system is made of power network and power consumer, and effect is exactly to be passed through as far as possible to all types of user of electric system
Ji ground provides reliable and the electric energy for requirement of satisfying the criteria meets burden requirement to meet the requirement of all types of user at any time.But
Be, in the present circumstance electric energy can't mass storage, this requires system power generation should at any time all with the change of system loading
Change and dynamic equilibrium is kept otherwise gently then to influence the quality for electricity consumption, the heavy then safety and stablization of entail dangers to system.And system
The acquisition of future load variation realizes therefore such Load Prediction In Power Systems just grow up by load prediction,
It is an important content in Automation of Electric Systems as important field of research in engineering science.
Load Prediction In Power Systems are using accurate statistical data and survey data as foundation, from the history of electricity consumption and existing
Shape sets out, and is fully considering some important system operating characteristics, increase-volume decision, natural conditions under conditions of social influence,
Research or the mathematical method handled over using set of system with future load.Under the meaning for meeting certain required precision,
Determine the load value of certain particular moment in future.
The purpose of load forecast is just to provide the state of development and level of load, is power generation department and management department
Door works out the production schedule and development plan provides foundation, determines the powering quantity in each power supply area, production programming etc..
The result of load forecast by the historical law of load itself other than being determined, also by numerous non-negative lotus factors
Influence, while and prediction theory, the prediction technique of use applied it is directly related.For many years, many scholars are to this project
It has made intensive studies, it is proposed that many methods.
The shortcomings that prior art:
1. most of prior art does not fully consider influence of the meteorological data to load fluctuation, not by meteorologic factor and
Load parameter is combined.
2. even if the load prediction of the prior art considers meteorologic factor, also due to huge meteorological data and load number
According to, and keep forecasting efficiency low.
3. even if the load prediction of the prior art considers meteorologic factor, also due to huge meteorological data and load number
According to, and keep precision of prediction low.
4. in existing Load Forecast Algorithm, simplifying for data is handled, just for load data or meteorological data, and
No while simplified processing load data and meteorological data.
Invention content
In order to solve the problems, such as background technology, the present invention propose it is a kind of based on wavelet neural network and support to
The power grid meteorology load data prediction technique of amount machine, while realizing and greatly reducing calculation amount and reached high-precision pre-
It surveys.
As shown in Figure 1, technical scheme of the present invention includes the following steps:
The first step:Extraction obtains the first history meteorological data, the first historical load data, the second history from electric network data
Meteorological data and the second historical load data constitute the first original number by the first history meteorological data and the first historical load data
According to, the second initial data is constituted by the second history meteorological data and the second historical load data, by the first history meteorological data and
Second history meteorological data constitutes original meteorological data, is made of the first historical load data and the second historical load data original
Load data;Wherein, the meteorological data in each days n before the first history meteorological data refers to, the first historical load data refers to preceding n
The load data in year each day, the second history meteorological data refers to the meteorological data in (n+1)th year each day, and the second historical load data is
Refer to the load data in (n+1)th year each day, daily meteorological data is made of multinomial meteorologic parameter, and daily load data is by multinomial
Load parameter is constituted;
Load data is made of load parameter, and meteorological data is made of meteorologic parameter.
Second step:It is original to the first initial data and second in such a way that clustering algorithm and Principal Component Analysis are combined
Data carry out simplifying processing, obtain the corresponding initial data in representative day remained, simplify data volume;
Third walks:The first initial data and the second initial data handle second step makees standardization, to it
In load data and meteorological data standardization is made using following formula as unit of the data of every day respectively:
Xi=XI is practical/XI is average
Wherein, XiIndicate normalized treated the i-th meteorology/load data for representing day, XI is practicalIndicate that i-th represents day
Original meteorology/load data, XI is averageIndicate the i-th average value for representing all original meteorology/load datas in day;
4th step:Load forecasting model is established using support vector machines, using per diem predicting, first after input standardization
It is trained in history meteorological data and the first historical load data to support vector machines, the load prediction mould after being trained
Then SVM prediction after the input training of second history meteorological data is obtained the first prediction load data by type;
5th step:Using monthly predicting, using the first history meteorological data after standardization as input layer, after standardization
The first historical load data as output layer, be trained using wavelet neural network;After the completion of training, by the second history gas
Image data is input to the neural network after training, and output obtains the second prediction load data;
Neural network is a kind of imitation animal nerve network behavior feature, carries out the algorithm number of distributed parallel information processing
Learn model.This network relies on the complexity of system, by adjusting the relationship being connected with each other between internal great deal of nodes, thus
Achieve the purpose that handle information.
6th step:Load forecasting model is obtained in conjunction with the result of four step of step the and the 5th step, when for predicting to be predicted
Between each day of section load data.
The second step, to the first initial data and second in such a way that clustering algorithm and Principal Component Analysis are combined
Initial data is respectively handled in the following ways:
Using average load parameter as second step clustering algorithm, simplification processing is required represents load parameter, with (n+1)th
The load data of the representative load parameter in year simplifies as second step clustering algorithm handles required data;
The representative day of representative load parameter that second step clustering algorithm simplification is handled is chosen as preceding n+1 every year
The representative day of whole load parameters;
Required for the meteorological data in the annual representative days n+1 is handled as second step Principal Component Analysis simplification before choosing
Data;
S1:Clustering algorithm is first used to be clustered to the load data in each day as unit of the load data in day, for poly-
It is every a kind of after class, it selects to be located at intermediate one day load data reservation in the one kind, remove from initial data in not being located at
Between one day load data, to greatly reduce load data amount, and be known as each day that processing obtains to represent day;
The clustering algorithm uses K clustering algorithms.
S2:Then it uses Principal Component Analysis to carry out processing to all meteorologic parameters in meteorological data and obtains each meteorology
The meteorologic parameter of the accumulation contribution rate of parameter, selection accumulation contribution rate to 80% retains, and removes from the meteorological data in each day
Contribution rate is not up to 80% meteorologic parameter, to reduce meteorological data amount.
The present invention reduces meteorological data amount and load data amount by above-mentioned two step, meteorological due to considering to alleviate
Factor and the big data brought calculates complicated problem.
4th step:Support vector machines is established load forecasting model and is handled in the following ways:
It inputs and is by the meteorological data of second step simplification and a certain representative day after third step standardization, to export
The same load data for representing day will represent day and be traversed from First Year to 1 year, to carry out load forecasting model training, instruction
The second history meteorological data, the first prediction load data for often representing day of output (n+1)th year are substituted into after the completion of practicing;It is pre- by first
It surveys load data and the second historical load data carries out precision of prediction verification, calculate the first precision of prediction E1。
In 5th step, the concrete structure of wavelet neural network is:Wavelet neural network be divided into input layer, hidden layer and
Three layers of output layer:Wherein input layer contains a*dxA input unit, a are the meteorologic parameter number that second step simplification is handled, dx
For total number of days in a month representative day of X that second step simplification is handled, each input unit represents one and represents day
One meteorologic parameter;Output layer is b*dxA output unit, b are the load parameter number that second step simplification is handled, dxIt is
Total number of days in the representative day for the X month that two step simplification are handled, each output unit represent one and represent the one of the world
A load parameter;Hidden layer has b*dxA hidden unit, each hidden unit are made of wavelet function, and wavelet function uses Morlet
Morther wavelet basic function;Input data is the meteorology obtained after second step simplifies and third walks standardization and load number
According to, by the whole in X month represent day output and input data as unit of, the whole in X month is represented into the defeated of day
Enter and substitutes into neural metwork training, the representative day in the annual X month for the preceding n that the first step is screened with output data
The first history meteorology and the first historical load data substitute into neural network successively, n times need to be trained altogether, deconditioning god after n times
Through network;X takes 1 to 12 successively, i.e., is all made of within annual 12 months the training that the above method carries out wavelet neural network, to
To 12 groups of wavelet-neural network models corresponding with 12 months.Using the moon as base unit, input the second history meteorological data profit
The wavelet neural network completed with training is predicted to obtain the second prediction load data, by the second prediction load data and the second history
Load data carries out precision of prediction verification, calculates the second precision of prediction E2。
It, can be further to the second precision of prediction E in specific implementation2Judge, if the second precision of prediction E2It is more than or equal to
90%, then it is assumed that wavelet neural network precision of prediction is met the requirements, into next step.Otherwise, wavelet function contraction-expansion factor is adjusted,
Shift factor, network connection weight, network connection threshold value and e-learning rate, re -training neural network, until meeting pre-
Survey required precision.
First precision of prediction E of the 4th step and the 5th step1With the second precision of prediction E2It calculates and obtains in the following ways:
The historical load data and the first/the in (n+1)th year each representative day that the 4th step and the 5th step are handled
Two prediction load datas use following formula, calculate (n+1)th year each first/second medium accuracy D for representing day1And D2:
Wherein, D is medium accuracy, is the first medium accuracy D of support vector machines load forecasting model1Or wavelet neural
Second medium accuracy D of network load prediction model2, n indicate load parameter number, i.e., daily peak load, day minimum load,
Day peak-valley difference and per day load,Indicate the second historical load data of load parameter i,Indicate the of load parameter i
One/the second prediction load data;
Then following formula is used to calculate precision of prediction again:
Wherein, E is precision of prediction, is the first precision of prediction E of support vector machines load forecasting model1And Wavelet Neural Network
Second precision of prediction E of network load forecasting model2, A indicate (n+1)th year in meet medium accuracy D be less than or equal to 7% representative day
Total number of days, B indicate that the (n+1)th Nian Zhong represents the total number of days in day.
6th step is specially:Calculate the first prediction load data and pass through that the 4th step is obtained by support vector machines
Between the second historical load data that neural network obtains on each load parameter in annual each representative day first
Medium accuracy D1, calculate the 5th step and pass through between the second obtained prediction load data of neural network and the second historical load data
The second medium accuracy D on each load parameter of every day2;By the second corresponding centre essence of the first medium accuracy
Degree compares, and to represent day as base unit, the first medium accuracy for choosing this day is corresponding with smaller in the second medium accuracy
Load forecasting model of the support vector machines/neural network as the representative day belongs to of a sort day use and generation with day is represented
Load forecasting model as table day, to obtain the load forecasting model for every day.
Multinomial meteorologic parameter in the meteorological data include maximum temperature, minimum temperature, mean temperature, relative humidity and
Rainfall.
Multinomial load parameter in the load data includes daily peak load, day minimum load, day peak-valley difference and Ping
Equal load.
In present invention specific implementation, moreover it is possible to per diem predict and monthly predict.
Per diem predict that it refers to utilizing the annual load feelings of data prediction future on the same day on the same day in historical data to refer to
Condition.Such as using on January 11st, 10, on January 11st, 11, on January 11st, 12, on January 11st, 13, on January 11st, 14 was predicted
On January 11st, 15, similarly and so on prediction January 12 etc. daily load data,
Monthly prediction refers to the historical load data of the previous year based on the moon to be predicted, and prediction waits for successively sequentially in time
The daily load of predicted month, for example, by using the total data in January, 10, the total data in January, 11, whole numbers in January, 12
According to, the total data in January, 13, the total data in January, 14 predicts the total data in January, 15.
In the present invention, select this four load parameters that can accurately represent the substantially variation feelings in one day internal loading
Condition selects this five meteorologic factors that can substantially reflect variation characteristic meteorological in one day.
In specific implementation, one month more additional than load data of meteorological data used is there is no load data, as waiting for
Predicted time section.
The beneficial effects of the invention are as follows:
Compared with the existing technology, the present invention fully considers influence of the meteorological data to load fluctuation, fully considers data
Scale, by clustering algorithm and Principal Component Analysis, while reducing load and meteorological data amount, and the prediction model of proposition passes through
The first/second precision of prediction calculation formula of definition, ensure that the precision of prediction of support vector machines and wavelet-neural network model,
Precision of prediction is improved, solves the problems, such as that the precision of prediction for considering meteorologic factor and bringing is low.
The power grid prediction technique of the prior art is usually to reduce meteorologic parameter, does not simplify load parameter, and the present invention passes through
Clustering and Principal Component Analysis greatly reduce meteorological data and load parameter simultaneously, to alleviate due to considering meteorologic factor
And the huge data volume brought, and by the first/second precision of prediction calculation formula of definition, ensure that support vector machines
With the precision of prediction of wavelet-neural network model, and as unit of day, the higher model of choice accuracy from two prediction models
Solves consideration as the final prediction model of this day to realize pinpoint accuracy while greatly improving forecasting efficiency
Meteorologic factor and the low problem of the precision of prediction that brings.
Description of the drawings
Fig. 1 is the method for the present invention logic diagram.
Fig. 2 is embodiment cluster result figure.
Fig. 3 is the comparison of prediction power and actual power in 2014 that embodiment is obtained according to SVM prediction model
Figure.
Fig. 4 is comparison of the embodiment according to prediction power and actual power in 2015 of prediction model based on wavelet neural network
Figure.
Specific implementation mode
The invention will be further described with reference to the accompanying drawings and examples.
The embodiment of the present invention is as follows:
The first step:Extraction obtains the first history meteorological data, the first historical load data, the second history from electric network data
Meteorological data and the second historical load data.
It (is adopted from the Power system load data on December 31st, 1 day 1 January in 2010 per 15min mono- in known somewhere
Sampling point, daily 96 points, dimension MW) and on January 31st, 1 day 1 January in 2010 meteorological data (max. daily temperature,
Daily minimum temperature, mean daily temperature, day relative humidity and daily rainfall).The day highest that this area's whole year is obtained through statistics is negative
Lotus, day minimum load, day peak-valley difference, daily load rate parameter load data.
First initial data is constituted by the first history meteorological data and the first historical load data, by the second history meteorology number
The second initial data is constituted according to the second historical load data;Wherein, the first history meteorological data refers to the meteorology in preceding 5 years each days
Data, the first historical load data refer to the load data in preceding 5 years each days, and the second history meteorological data refers to the 6th year each day
Meteorological data, the second historical load data refer to the load data in the 6th year each day.Daily meteorological data is by multinomial meteorologic parameter
It constitutes, daily load data is made of multinomial load parameter.Period to be predicted is January 31 1 day to 2016 January in 2016
Number
Second step:K-Means clustering algorithms are used to gather to the load data in each day as unit of the load data in day
Class is removed for per one kind, selecting one day load data for being located at centre in the one kind to retain after cluster from initial data
It is not located at one day intermediate load data, by taking the cluster in January, 2012 as an example, the number of days being connected with lines represents in the past few days
It can be classified as one kind, as shown in Figure 2, January 5,13,16,20,25, No. 27 in 2012 is the representative day of this month.
Third walks:Using Principal Component Analysis, using Principal Component Analysis to the institute of each meteorologic parameter for representing the world
There is meteorological data to carry out the accumulation contribution rate that processing obtains each meteorologic parameter, the meteorology ginseng of selection accumulation contribution rate to 80%
Number retains, and removes the meteorologic parameter that contribution rate is not up to 80% from the meteorological data in each day.
Table 1 is the weight shared by each parameter, by principal component analysis it is found that contribution rate shared by maximum temperature and minimum temperature
Highest, and and be more than 0.8, so choosing " maximum temperature, minimum temperature " taken in as main weather factor.
Table 1:Weight shared by each meteorologic factor
Weight | Maximum temperature | Minimum temperature | Mean temperature | Relative humidity | Rainfall |
Contribution rate | 0.651 | 0.2182 | 0.1236 | 0.0066 | 0.0006 |
4th step:Load forecasting model is established using support vector machines, using per diem predicting, first after input standardization
It is trained in history meteorological data and the first historical load data to support vector machines, the load prediction mould after being trained
Then SVM prediction after the input training of second history meteorological data is obtained the first prediction load data, by the by type
One prediction load data and the second historical load data carry out precision of prediction verification, calculate the first precision of prediction E1。
Further, if the first precision of prediction E1More than or equal to 90%, then it is assumed that hold the prediction of vector machine load forecasting model
Precision is met the requirements, into next step.Otherwise, relevant parameter is adjusted, until meeting precision of prediction requirement.
Its prediction result is by taking first 60 of daily peak load data in 2014 represent day as an example, as shown in Figure 3.
There is upper figure to can be seen that data variation trend in 2014 and the reality variation tendency in 2014 predicted basic
It is consistent, this shows the data and actual coincidence of prediction, has very high reliability.And first precision of prediction be 96%, table
Bright prediction accuracy is very high high with the practical goodness of fit.
5th step:Using monthly predicting, using the first history meteorological data after standardization as input layer, after standardization
The first historical load data as output layer, be trained using wavelet neural network;After the completion of training, by the second history gas
Image data is input to the neural network after training, and output obtains the second prediction load data;By the second prediction load data and the
Two historical load datas carry out precision of prediction verification, calculate the second precision of prediction E2。
Further, if the second precision of prediction E2More than or equal to 90%, then it is assumed that wavelet neural network precision of prediction meets
It is required that into next step.Otherwise, wavelet function contraction-expansion factor, shift factor, network connection weight, network connection threshold value are adjusted
With e-learning rate, re -training neural network, up to meeting precision of prediction requirement.
The prediction data of wavelet neural network is as shown in figure 4, with daily peak load data instance in 2015.
There is upper figure to can be seen that data variation trend in 2015 and the reality variation tendency in 2015 predicted basic
It is consistent, this shows the data and actual coincidence of prediction, has very high reliability.And first precision of prediction be 98.4%,
Show that prediction accuracy is very high high with the practical goodness of fit.
Specific implementation considers meteorological data and load data by two big algorithm of clustering algorithm and principal component analysis
Simplify processing, so as to greatly reduce the burden of the data in neural network prediction, accelerate forecasting efficiency, by last prediction essence
Degree using the corresponding algorithm that simplifies it is found that even if reduce data bulk, but influence and symbol of the meteorologic factor to load parameter
The overall variation trend of load parameter is retained, to ensure that computational accuracy.And it is pre- by the first/second of definition
Accuracy computation formula is surveyed, ensure that the precision of prediction of support vector machines and wavelet-neural network model.
6th step:Calculate the first prediction load data and obtained by neural network that the 4th step is obtained by support vector machines
To the second historical load data between the first medium accuracy D on each load parameter in annual each representative day1,
Calculate the 5th step by neural network obtain second prediction load data and the second historical load data between in every day
The second medium accuracy D on each load parameter2;The second corresponding medium accuracy of first medium accuracy is compared, with
Day is represented as base unit, chooses the first medium accuracy supporting vector corresponding with smaller in the second medium accuracy of this day
Load forecasting model of the machine/neural network as the representative day is used and is represented as day with representing day and belonging to of a sort day
Load forecasting model, to obtain the load forecasting model for every day.
The 6th step of embodiment as unit of day, from two prediction models the higher model of choice accuracy as this day most
Whole prediction model is solved and considers meteorologic factor and band to realizing pinpoint accuracy while greatly improving forecasting efficiency
The low problem of next precision of prediction.
The prediction result of No. 11-17 average load of period wherein to be predicted is as shown in table 2:
Table 2:The prediction result of No. 11-17 average load of period to be predicted
By prediction model result proposed by the present invention and the neural network prediction model for considering all meteorology and load data
The comparison on time and precision of prediction, such as the following table 3 are carried out, prediction data is the flat of on January 15,1 day to 2015 January in 2015
Equal load data.
Table 3
Algorithm | Average daily precision of prediction E | Calculate the time (unit second) |
Prediction model proposed by the present invention | 97.4% | 16.88 |
Neural network prediction model | 83.2% | 43.16 |
It can be seen that either precision of prediction still calculates on the time, based on wavelet neural network and support vector machines
Power grid meteorology load forecasting method embodies good meter efficiency and precision of prediction.
Claims (8)
1. a kind of power grid meteorology load forecasting method based on wavelet neural network and support vector machines, it is characterised in that:
The first step:It is meteorological to obtain the first history meteorological data, the first historical load data, the second history for extraction from electric network data
Data and the second historical load data constitute the first initial data by the first history meteorological data and the first historical load data,
Second initial data is constituted by the second history meteorological data and the second historical load data;Wherein, the first history meteorological data is
The meteorological data in each days n before referring to, the first historical load data refer to the load data in the preceding each days n, the second history meteorology number
According to the meteorological data for referring to (n+1)th year each day, the second historical load data refers to the load data in (n+1)th year each day, daily
Meteorological data is made of multinomial meteorologic parameter, and daily load data is made of multinomial load parameter;
Second step:To the first initial data and the second initial data in such a way that clustering algorithm and Principal Component Analysis are combined
It carries out simplifying processing, obtains the corresponding initial data in representative day remained;
Third walks:The first initial data and the second initial data handle second step makees standardization, to therein
Load data and meteorological data make standardization as unit of the data of every day using following formula respectively:
Xi=XI is practical/XI is average
Wherein, XiIndicate normalized treated the i-th meteorology/load data for representing day, XI is practicalIndicate the i-th original for representing day
Beginning meteorology/load data, XI is averageIndicate the i-th average value for representing all original meteorology/load datas in day;
4th step:It is carried out in the first history meteorological data and the first historical load data to support vector machines after input standardization
Training, the load forecasting model after being trained are then pre- by the support vector machines after the input training of the second history meteorological data
It surveys and obtains the first prediction load data;
5th step:Using monthly predicting, using the first history meteorological data after standardization as input layer, with after standardization
One historical load data is trained as output layer using wavelet neural network;After the completion of training, by the second history meteorology number
According to the neural network being input to after training, output obtains the second prediction load data;
6th step:Load forecasting model is obtained in conjunction with the result of four step of step the and the 5th step.
2. a kind of power grid meteorology load prediction side based on wavelet neural network and support vector machines according to claim 1
Method, it is characterised in that:
The second step, it is original to the first initial data and second in such a way that clustering algorithm and Principal Component Analysis are combined
Data are respectively handled in the following ways:
S1:Clustering algorithm is first used to be clustered to the load data in each day as unit of the load data in day, after cluster
Per a kind of, select to be located at each day that one day intermediate load data retains, and processing is obtained in the one kind and be known as representing
It;
S2:Then it uses Principal Component Analysis to carry out processing to all meteorologic parameters in meteorological data and obtains each meteorologic parameter
Accumulation contribution rate, meteorologic parameter of the selection accumulation contribution rate to 80% retain.
3. a kind of power grid meteorology load prediction side based on wavelet neural network and support vector machines according to claim 1
Method, it is characterised in that:4th step:Support vector machines is established load forecasting model and is handled in the following ways:Input is warp
It crosses second step and simplifies the meteorological data for walking a certain representative day after standardization with third, export as the same load for representing day
Data will represent day and be traversed from First Year to 1 year, and to carry out load forecasting model training, second is substituted into after the completion of training
History meteorological data, the first prediction load data for often representing day of output (n+1)th year;Load data and second is predicted by first
Historical load data carries out precision of prediction verification, calculates the first precision of prediction E1。
4. a kind of power grid meteorology load prediction side based on wavelet neural network and support vector machines according to claim 1
Method, it is characterised in that:In 5th step, the concrete structure of wavelet neural network is:Wavelet neural network is divided into input layer, hidden
Containing three layers of layer and output layer:Wherein input layer contains a*dxA input unit, a are the meteorologic parameter that second step simplification is handled
Number, dxFor total number of days in a month representative day of X that second step simplification is handled;Output layer is b*dxA output unit, b
For the load parameter number that second step simplification is handled, dxThe representative day of the X month handled for second step simplification
Total number of days;Hidden layer has b*dxA hidden unit, each hidden unit are made of wavelet function, and wavelet function is small using mother Morlet
Wave basic function;By the whole in X month represent day output and input data as unit of, whole in X month are represented
It the data that output and input substitute into neural metwork training, in a month of the annual X of the preceding n that the first step is screened
The the first history meteorology and the first historical load data for representing day substitute into neural network successively, need to train n times altogether, stop after n times
Training neural network;Using the moon as base unit, the second history meteorological data of input is pre- using the wavelet neural network that training is completed
It measures to the second prediction load data, the second prediction load data and the second historical load data is subjected to precision of prediction verification,
Calculate the second precision of prediction E2。
5. a kind of power grid meteorology load prediction side based on wavelet neural network and support vector machines according to claim 1
Method, it is characterised in that:First precision of prediction E of the 4th step and the 5th step1With the second precision of prediction E2It calculates and obtains in the following ways
:The historical load data and first/second in (n+1)th year each representative day that the 4th step and the 5th step are handled are predicted
Load data uses following formula, calculates (n+1)th year each first/second medium accuracy D for representing day1And D2:
Wherein, D is medium accuracy, is the first medium accuracy D of support vector machines load forecasting model1Or wavelet neural network
Second medium accuracy D of load forecasting model2, the number of n expression load parameters, i.e. daily peak load, day minimum load, Feng
Paddy difference and per day load,Indicate the second historical load data of load parameter i,Indicate the first/the of load parameter i
Two prediction load datas;
Then following formula is used to calculate precision of prediction again:
Wherein, E is precision of prediction, is the first precision of prediction E of support vector machines load forecasting model1It is negative with wavelet neural network
Second precision of prediction E of lotus prediction model2, A indicate (n+1)th year in meet medium accuracy D be less than or equal to 7% the total day in representative day
Number, B indicate that the (n+1)th Nian Zhong represents the total number of days in day.
6. a kind of power grid meteorology load prediction side based on wavelet neural network and support vector machines according to claim 1
Method, it is characterised in that:6th step is specially:Calculate the first prediction load data that the 4th step is obtained by support vector machines
Between the second historical load data obtained by neural network on each load parameter in annual each representative day
The first medium accuracy D1, calculate the second prediction load data and the second historical load number that the 5th step is obtained by neural network
The second medium accuracy D between on each load parameter of every day2;To represent day as base unit, the day is chosen
The first medium accuracy support vector machines/neural network corresponding with smaller in the second medium accuracy as the negative of the representative day
Lotus prediction model uses with representing day and belonging to of a sort day and represents the load forecasting model as day, to be directed to
The load forecasting model of every day.
7. a kind of power grid meteorology load prediction side based on wavelet neural network and support vector machines according to claim 1
Method, it is characterised in that:Multinomial meteorologic parameter in the meteorological data includes maximum temperature, minimum temperature, mean temperature, opposite
Humidity and rainfall.
8. a kind of power grid meteorology load prediction side based on wavelet neural network and support vector machines according to claim 1
Method, it is characterised in that:Multinomial load parameter in the load data includes daily peak load, day minimum load, day peak-valley difference
With per day load.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810359331.2A CN108596242B (en) | 2018-04-20 | 2018-04-20 | Power grid meteorological load prediction method based on wavelet neural network and support vector machine |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810359331.2A CN108596242B (en) | 2018-04-20 | 2018-04-20 | Power grid meteorological load prediction method based on wavelet neural network and support vector machine |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108596242A true CN108596242A (en) | 2018-09-28 |
CN108596242B CN108596242B (en) | 2021-03-23 |
Family
ID=63614224
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810359331.2A Active CN108596242B (en) | 2018-04-20 | 2018-04-20 | Power grid meteorological load prediction method based on wavelet neural network and support vector machine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108596242B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109345027A (en) * | 2018-10-25 | 2019-02-15 | 国网江苏省电力有限公司盐城供电分公司 | Micro-capacitance sensor short-term load forecasting method based on independent component analysis and support vector machines |
CN110617927A (en) * | 2019-09-20 | 2019-12-27 | 长安大学 | Structural settlement deformation prediction method based on EMD-SVR-WNN |
CN110826789A (en) * | 2019-10-30 | 2020-02-21 | 深圳市康必达控制技术有限公司 | Power load prediction method and device based on power system and terminal equipment |
CN110991638A (en) * | 2019-11-29 | 2020-04-10 | 国网山东省电力公司聊城供电公司 | Generalized load modeling method based on clustering and neural network |
CN112215459A (en) * | 2020-09-02 | 2021-01-12 | 南方电网能源发展研究院有限责任公司 | Power distribution method and device based on power grid investment scale prediction |
CN112418476A (en) * | 2019-08-23 | 2021-02-26 | 武汉剑心科技有限公司 | Ultra-short-term power load prediction method |
CN112418533A (en) * | 2020-11-25 | 2021-02-26 | 江苏电力交易中心有限公司 | Clean energy electric quantity decomposition prediction method |
CN115051864A (en) * | 2022-06-21 | 2022-09-13 | 郑州轻工业大学 | PCA-MF-WNN-based network security situation element extraction method and system |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2005122517A (en) * | 2003-10-17 | 2005-05-12 | Fuji Electric Holdings Co Ltd | Energy demand prediction method, energy demand prediction device and energy demand prediction program and recording medium |
CN102982387A (en) * | 2012-10-18 | 2013-03-20 | 安徽工程大学 | Method for predicting short-term power load |
CN103218675A (en) * | 2013-05-06 | 2013-07-24 | 国家电网公司 | Short-term load prediction method based on clustering and sliding window |
CN104008430A (en) * | 2014-05-29 | 2014-08-27 | 华北电力大学 | Method for establishing virtual reality excavation dynamic smart load prediction models |
CN104318332A (en) * | 2014-10-29 | 2015-01-28 | 国家电网公司 | Power load predicting method and device |
CN106208388A (en) * | 2016-08-31 | 2016-12-07 | 科大智能电气技术有限公司 | A kind of intelligence charging system and short term basis load prediction implementation method thereof in order |
CN106410781A (en) * | 2015-07-29 | 2017-02-15 | 中国电力科学研究院 | Power consumer demand response potential determination method |
-
2018
- 2018-04-20 CN CN201810359331.2A patent/CN108596242B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2005122517A (en) * | 2003-10-17 | 2005-05-12 | Fuji Electric Holdings Co Ltd | Energy demand prediction method, energy demand prediction device and energy demand prediction program and recording medium |
CN102982387A (en) * | 2012-10-18 | 2013-03-20 | 安徽工程大学 | Method for predicting short-term power load |
CN103218675A (en) * | 2013-05-06 | 2013-07-24 | 国家电网公司 | Short-term load prediction method based on clustering and sliding window |
CN104008430A (en) * | 2014-05-29 | 2014-08-27 | 华北电力大学 | Method for establishing virtual reality excavation dynamic smart load prediction models |
CN104318332A (en) * | 2014-10-29 | 2015-01-28 | 国家电网公司 | Power load predicting method and device |
CN106410781A (en) * | 2015-07-29 | 2017-02-15 | 中国电力科学研究院 | Power consumer demand response potential determination method |
CN106208388A (en) * | 2016-08-31 | 2016-12-07 | 科大智能电气技术有限公司 | A kind of intelligence charging system and short term basis load prediction implementation method thereof in order |
Non-Patent Citations (3)
Title |
---|
XI CHEN 等: "An improved load forecast model using factor analysis: An Australian case study", 《PROCEEDING OF THE 2017 IEEE INTERNATIONAL CONFERENCE ON INFORMATION AND AUTOMATION(ICIA) MACAU SAR,CHINA 》 * |
康重庆 等: "电力***负荷预测研究综述与发展方向的探讨", 《电力***自动化》 * |
***: "基于知识挖掘技术的智能协同电力负荷预测研究", 《中国博士学位论文全文数据库-工程科技Ⅱ辑》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109345027A (en) * | 2018-10-25 | 2019-02-15 | 国网江苏省电力有限公司盐城供电分公司 | Micro-capacitance sensor short-term load forecasting method based on independent component analysis and support vector machines |
CN109345027B (en) * | 2018-10-25 | 2021-11-23 | 国网江苏省电力有限公司盐城供电分公司 | Micro-grid short-term load prediction method based on independent component analysis and support vector machine |
CN112418476A (en) * | 2019-08-23 | 2021-02-26 | 武汉剑心科技有限公司 | Ultra-short-term power load prediction method |
CN110617927A (en) * | 2019-09-20 | 2019-12-27 | 长安大学 | Structural settlement deformation prediction method based on EMD-SVR-WNN |
CN110617927B (en) * | 2019-09-20 | 2022-04-05 | 长安大学 | Structural settlement deformation prediction method based on EMD-SVR-WNN |
CN110826789B (en) * | 2019-10-30 | 2023-06-06 | 深圳市康必达控制技术有限公司 | Power load prediction method and device based on power system and terminal equipment |
CN110826789A (en) * | 2019-10-30 | 2020-02-21 | 深圳市康必达控制技术有限公司 | Power load prediction method and device based on power system and terminal equipment |
CN110991638A (en) * | 2019-11-29 | 2020-04-10 | 国网山东省电力公司聊城供电公司 | Generalized load modeling method based on clustering and neural network |
CN110991638B (en) * | 2019-11-29 | 2024-01-05 | 国网山东省电力公司聊城供电公司 | Generalized load modeling method based on clustering and neural network |
CN112215459A (en) * | 2020-09-02 | 2021-01-12 | 南方电网能源发展研究院有限责任公司 | Power distribution method and device based on power grid investment scale prediction |
CN112418533A (en) * | 2020-11-25 | 2021-02-26 | 江苏电力交易中心有限公司 | Clean energy electric quantity decomposition prediction method |
CN115051864A (en) * | 2022-06-21 | 2022-09-13 | 郑州轻工业大学 | PCA-MF-WNN-based network security situation element extraction method and system |
CN115051864B (en) * | 2022-06-21 | 2024-02-27 | 郑州轻工业大学 | PCA-MF-WNN-based network security situation element extraction method and system |
Also Published As
Publication number | Publication date |
---|---|
CN108596242B (en) | 2021-03-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11581740B2 (en) | Method, system and storage medium for load dispatch optimization for residential microgrid | |
CN108596242A (en) | Power grid meteorology load forecasting method based on wavelet neural network and support vector machines | |
CN112508275B (en) | Power distribution network line load prediction method and equipment based on clustering and trend indexes | |
CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
Wang et al. | A seasonal GM (1, 1) model for forecasting the electricity consumption of the primary economic sectors | |
CN109508857B (en) | Multi-stage planning method for active power distribution network | |
CN107578124B (en) | Short-term power load prediction method based on multilayer improved GRU neural network | |
Lu et al. | US natural gas consumption prediction using an improved kernel-based nonlinear extension of the Arps decline model | |
CN110222882A (en) | A kind of prediction technique and device of electric system Mid-long Term Load | |
CN103310286A (en) | Product order prediction method and device with time series characteristics | |
US20210326696A1 (en) | Method and apparatus for forecasting power demand | |
Yuan et al. | Conditional style-based generative adversarial networks for renewable scenario generation | |
CN112149890A (en) | Comprehensive energy load prediction method and system based on user energy label | |
Mayer et al. | Probabilistic modeling of future electricity systems with high renewable energy penetration using machine learning | |
CN114021848A (en) | Generating capacity demand prediction method based on LSTM deep learning | |
CN115205068A (en) | Energy storage optimal peak-valley time interval dividing method considering net load demand distribution | |
CN114444805A (en) | Control method for smooth output of multi-photovoltaic power station shared energy storage system | |
CN114091776A (en) | K-means-based multi-branch AGCNN short-term power load prediction method | |
CN108537581B (en) | Energy consumption time series prediction method and device based on GMDH selective combination | |
CN113762591A (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM counterstudy | |
CN108615091A (en) | Electric power meteorology load data prediction technique based on cluster screening and neural network | |
CN112508734A (en) | Method and device for predicting power generation capacity of power enterprise based on convolutional neural network | |
CN116681173A (en) | Data-intensive power load prediction parallel optimization method based on RNN model | |
CN114676931B (en) | Electric quantity prediction system based on data center technology | |
CN110659775A (en) | LSTM-based improved electric power short-time load prediction algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |