CN106408223A - Short-term load prediction based on meteorological similar day and error correction - Google Patents
Short-term load prediction based on meteorological similar day and error correction Download PDFInfo
- Publication number
- CN106408223A CN106408223A CN201611079983.8A CN201611079983A CN106408223A CN 106408223 A CN106408223 A CN 106408223A CN 201611079983 A CN201611079983 A CN 201611079983A CN 106408223 A CN106408223 A CN 106408223A
- Authority
- CN
- China
- Prior art keywords
- error
- value
- day
- meteorological
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06313—Resource planning in a project environment
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0637—Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
- G06Q10/06375—Prediction of business process outcome or impact based on a proposed change
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Educational Administration (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention, which relates to the scheduling, operating, and planning field of the power system, discloses a short-term load prediction based on meteorological similar day and error correction. According to the invention, a meteorological factor regression analysis is carried out by using SPSS software; meteorological factors affecting a load most obviously in all seasons are selected; and weights of all factors are determined and are used as a basis for selecting a meteorological similar day. A historical prediction error data sample set is established; for a certain prediction day, an error data sample establishment set of a similar day is extracted, and a probability density distributed fitting model is established. An error fluctuation situation of a prediction point is analyzed to obtain a compensation value of a predicted error; and an error sample value closest to the error compensation value is selected as an error fitting value of this time and is superposed on the prediction value. Therefore, the prediction precision is improved.
Description
Technical field
The present invention relates to electric power system dispatching, operation and planning field, particularly to a kind of based on meteorological similar day and mistake
The short-term load forecasting method of difference correction.
Background technology
As short-term load forecasting become the important component part in electric power system dispatching, operation and planning.In China's electricity
Lixing industry is moved towards under the trend of the marketization, and the result of load prediction has increasingly constituted electric power enterprise and formulated production & marketing plan, carries
The key of high economic benefit.The load of power system has its own intrinsic periodic regularity, is also subject to factors simultaneously
Impact, such as weather conditions, the level of economic development, energy supply mode etc..Due to the difference of part throttle characteristics between each department, for
The load prediction work of different regions all should consider load in conjunction with local practical situation on the basis of Load Characteristic Analysis
Influence factor, then be predicted from suitable method, to improve precision of prediction.
Meteorological factor is to cause one of major reason of electric power short term change, and the polytropy of meteorological condition is to power train
System load prediction work causes no small puzzlement.Have many literature research analysis and processing method of meteorological factor at present,
And load prediction work is carried out based on this.The roadmap of meteorological factor is varied, sums up and mainly has two kinds:One be from
The angle of part throttle characteristics is started with, using the thinking decomposing load or layered modeling.Such as load sequence is decomposed into fundamental frequency, low frequency
And high fdrequency components, introduce hourly weather factors in low, high-frequency load component are predicted and adopt multiple models;Such as according to load
By the difference of meteorological factor influence degree, daily load is decomposed into by two parts using wavelet transformation, is respectively adopted recurrent nerve net
Network forecast model and linear session sequence ARMA forecast model, improve precision of prediction and modeling efficiency;Being equally such as will be total
Load and weather sensitive load based on load decomposition, right respectively with multilayer feedforward neural network model using Grey System Model
Two kinds of loads are predicted.Second roadmap is to start with model from prediction algorithm, and the effectiveness using algorithm is special by day
The impact levying meteorological factor is taken into account, such as a kind of SVM prediction method based on data mining pretreatment, utilizes
Data mining technology is from meteorological angle Selection similar day, thus simplifying the training data with optimal prediction model, to reach more preferably
Prediction effect such as pass through to analyze the corresponding relation of meteorological factor and network load, give experimental correction model storehouse
Method for building up, and the factors such as temperature, humidity, air pressure are set up with correction model, it is preferably pre- that test shows that this forecast model has
Survey precision.
But it is based on above-mentioned theory, the polytropy of meteorological condition still causes not little to Load Prediction In Power Systems work
Puzzlement, the internal relation between meteorological condition and forecast error numerical value characteristic does not also disclose, and does not also disclose in prior art
Contacting between the main weather factor of impact load variations and load, these cause puzzlement to work about electric power person.
Content of the invention
The technical problem to be solved is to provide a kind of short term based on meteorological similar day and error correction
Forecasting Methodology, it is different from above two main flow thinking, introduces the thought of error correction, sets up historical forecast error information
Collection, analyzes the internal relation between meteorological condition and forecast error numerical value characteristic using statistical method, thus according to meteorological condition pair
Forecast error is fitted, to reach the purpose that short term is predicted.
For solving above-mentioned technical problem, the technical solution used in the present invention is:One kind is based on meteorological similar day and error
Correction short-term load forecasting method it is characterised in that:Its step is:
Initially set up multiple linear regression model, by historical data:Load data, meteorological data, carry out the pre- place of data
Reason, carries out stepwise regression analysis showed according still further to multiple linear regression model to historical data, selects impact load in each season
The most significant meteorological factor, determines the weight of each factor and in this, as the foundation selecting meteorological similar day;Build by season type
Vertical historical forecast error information sample set, for a certain prediction day, extracts the error information Sample Establishing set of its similar day, and
By season, probability density distribution matching is carried out to it;The fluctuating error situation of future position is analyzed, obtains forecast error and mend
Repay value, choose the error sample value closest to forecast error offset as the error fit value in this moment, and the prediction that is added to
In value, as final load prediction results.
Further technical scheme is, described multiple linear regression model is to be set up based on principle of least square method,
The analytical parameters of its model are the reaction coefficient of determination of regression effect, the regression effect test value in variance analyses, its significance
Test value, the test value of regression coefficient.
Further technical scheme is, described load data is daily peak load, day minimum load, per day load;
Described meteorological data is max. daily temperature, Daily minimum temperature, mean daily temperature, day relative humidity, daily rainfall.
Further technical scheme is, the determination method of described forecast error offset is in history day forecast error
Choose multiple long-term relative erroies as the criterion of recent relative error analysis, then calculate long-term variance level and matching
Straight slope absolute value, in conjunction with the undulatory property of the two approximate analyses last-period forecast relative error, obtains future position relative error
Estimated value;Relatively large with the threshold values error of little error with previous moment error, the big error of division finally according to prediction point tolerance
Little, determine relative error offset.
Further technical scheme is, described error sample value adopts systematic sampling in the error information sample of similar day
It is sampled in this foundation set;The step of its systematic sampling is as follows:
When the sample that the totally middle extraction capacity for N for the capacity is n, if being totally divided into uniform stem portion, then press
According to the rule pre-establishing, extract an individual from each section, obtain required sample;
Wherein,
1) overall capacity N is larger;
2) pitch requirements equal (generally K=N/n) of overall segmentation;
3) in each sample segments using stochastic sampling it is ensured that the probability that is pumped to of each sample is equal.
Further technical scheme is, described error fit value is according to forecast error offset size, error to be sampled
Value is ranked up, and chooses with forecast error offset immediate error sample value as error fit value.
Further technical scheme is, described multiple linear regression model is
Y=y '+μ=b0+b1X1+b2X2+...+biXi+...+bkXk(1)
Further technical scheme is, the analysis method of the undulatory property of described relative error is to see table:
Wherein, σlFor long-term variance level, σsFor in the recent period relative to the variance of calculation error, klAbsolute for fitting a straight line slope
The marginal value of value, ksThe slope absolute value of the fitting a straight line calculating for recent relative error.
Further technical scheme is, the determination method of described error compensation value see table:
Wherein, δiFor predicting point tolerance, δi-1For future position previous moment error, δ0For dividing the big error of future position and prediction
The threshold values error of the little error of point.
Further technical scheme is, described error sample value in sampling, by the mistake in similar day error sample set
Difference data was divided according to the period, 24 hours of a day was divided into 6 periods it is ensured that sampling is uniformly entered in day part
OK;
Wherein, choose centered on peak value in each matched curve, probability be 85% part as actual sampling interval.
Have the beneficial effects that using produced by technique scheme:The present invention first with SPSS software carry out meteorological because
Plain regression analyses, find the main weather factor of impact load variations, and the dependency relation according to both is assigned to influence factor weigh
Weight, and meteorological similar day is selected with this.Then extracting the forecast error of similar day in history error sample set, to set up probability close
Degree fitting of distribution model, and combine the error fit value that the undulatory property analysis of prediction point tolerance obtains this point, in order to predictive value
It is corrected, obtain final predicted load.Demonstrate the correctness and effectively of the put forward forecast reason of the present invention with embodiment
Property.
Brief description
The present invention is further detailed explanation with reference to the accompanying drawings and detailed description.
The corresponding relative error distribution of Fig. 1 Various Seasonal;
Fig. 2 Forecasting Methodology flow chart;
Fig. 3 wavelet neural network algorithm routine flow chart;
Fig. 4 A1~A4 relative error fitting result;
Predict the outcome before and after Fig. 5 A1~A4 correction contrast.
Specific embodiment
With reference to the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Ground description is it is clear that described embodiment is only a part of embodiment of the present invention, rather than whole embodiments.It is based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of not making creative work
Embodiment, broadly falls into the scope of protection of the invention.
Elaborate a lot of details in the following description in order to fully understand the present invention, but the present invention is acceptable
To be implemented different from alternate manner described here using other, those skilled in the art can be without prejudice to intension of the present invention
In the case of do similar popularization, therefore the present invention is not limited by following public specific embodiment.
Technical scheme provided by the present invention is, the short-term load forecasting side based on meteorological similar day and error correction
Method, its step is as follows:
Initially set up multiple linear regression model, by historical data:Load data, meteorological data, carry out the pre- place of data
Reason, carries out stepwise regression analysis showed according still further to multiple linear regression model to historical data, selects impact load in each season
The most significant meteorological factor, determines the weight of each factor and in this, as the foundation selecting meteorological similar day;Build by season type
Vertical historical forecast error information sample set, for a certain prediction day, extracts the error information Sample Establishing set of its similar day, and
By season, probability density distribution matching is carried out to it;The fluctuating error situation of future position is analyzed, obtains forecast error and mend
Repay value, choose the error sample value closest to forecast error offset as the error fit value in this moment, and the prediction that is added to
In value, as final load prediction results.
Wherein, multiple linear regression model can be set up based on principle of least square method, the analytical parameters of its model
For reacting the coefficient of determination of regression effect, regression effect test value in variance analyses, its significance test value, regression coefficient
Test value.
Wherein, load data mostly is daily peak load, day minimum load, per day load;Described meteorological data mostly is day
Maximum temperature, Daily minimum temperature, mean daily temperature, day relative humidity, daily rainfall.
Wherein, the determination method of forecast error offset is to choose multiple long-term relative erroies in history day forecast error
As the criterion of recent relative error analysis, then calculate long-term variance level and fitting a straight line slope absolute value, in conjunction with
The undulatory property of the two approximate analyses last-period forecast relative error, obtains the estimated value of future position relative error;Finally according to prediction
Point tolerance and the relative size of previous moment error, the threshold values error dividing big error and little error, determine that relative error compensates
Value.
Wherein, error sample value is taken out in the error information Sample Establishing set of similar day using systematic sampling
Sample;Its systematic sampling is as follows:
When the sample that the totally middle extraction capacity for N for the capacity is n, if being totally divided into uniform stem portion, then press
According to the rule pre-establishing, extract an individual from each section, obtain required sample;
Wherein,
1) overall capacity N is larger;
2) pitch requirements equal (generally K=N/n) of overall segmentation;
3) in each sample segments using stochastic sampling it is ensured that the probability that is pumped to of each sample is equal.
Wherein, error fit value is according to forecast error offset size, error sample value to be ranked up, and chooses and pre-
Survey error compensation value immediate error sample value as error fit value.
Wherein, multiple linear regression model is specially
Y=y '+μ=b0+b1X1+b2X2+...+biXi+...+bkXk(1)
Wherein, the analysis method of the undulatory property of relative error is to see table:
Wherein, σlFor long-term variance level, σsFor in the recent period relative to the variance of calculation error, klAbsolute for fitting a straight line slope
The marginal value of value, ksThe slope absolute value of the fitting a straight line calculating for recent relative error.
Wherein, the determination method of described error compensation value see table:
Wherein, δiFor predicting point tolerance, δi-1For future position previous moment error, δ0For dividing the big error of future position and prediction
The threshold values error of the little error of point.
Wherein, error sample value, in sampling, needs to enter the error information in similar day error sample set according to the period
Row divides, and 24 hours of a day is divided into 6 periods it is ensured that sampling is uniformly carried out in day part;
Wherein, choose centered on peak value in each matched curve, probability be 85% part as actual sampling interval.
Based on above-mentioned discussion, as follows for the concrete exploration of the present invention, research:
Embodiment one
The 1 meteorological factor regression analyses based on SPSS
The foundation of 1.1 Multivariable regressive analysis model
In actual life, people often will carry out statistical analysiss to certain dependent variable, but affects certainly becoming of this dependent variable
Amount often more than one.For example, it is desired to consider k independent variable X1, X2..., XkDuring relation and dependent variable y between, using minimum
Square law principle sets up multiple linear regression model:
Y=y '+μ=b0+b1X1+b2X2+...+biXi+...+bkXk(1)
By formula (1) as can be seen that dependent variable y is made up of two parts, Part I y ' is the estimated value of dependent variable y, represents
The part that can be determined by variable;U is residual error, represents the part not determined by independent variable.U for the model judging current foundation is
No establishment, if also have other variable to need a series of problems, such as introduce model extremely important.B in formula (1)0For constant term, table
Show the intercept of equation;biFor partial regression coefficient, represent when other independent variables are constant, independent variable XiOften y ' during 1 unit of changeization
Variable quantity.Multiple linear regression not only needs to carry out the inspection of regression coefficient, estimates the confidence interval of regression coefficient, carries out pre-
The discussion of the aspect such as survey and hypothesis testing, in addition it is also necessary to consider the relation between each independent variable, such as whether there is altogether between them
Linear problem.
Before carrying out multiple regression analysis using SPSS, first data should be organized (load data and meteorological data
It was made the pretreatment of data), then complete in the function menu of SPSS multiple linear regression analysis, concrete operations are such as
Under:
(1) select menu Analyze → Regression → Linear;
(2) dependent variable is selected into Dependent (dependent variable);
(3) one or more independent variables are selected in Independent (independent variable);
(4) Part and partial correlations, Collinearity are selected in Statistics option
Diagnostics (collinearity diagnostics), Estimates, Model fit (system default options);
(5) in Plots option, * ZPRED (probability) is selected into Y, * ZRESID (residual error) is selected into X, and chooses simultaneously
Histogram and Normal probability plot, draws the standardized residual sequence rectangular histogram of regression result and normal state is divided
Cloth accumulated probability figure, carries out the analysis of regression accuracy, clicks on OK.
By R in F value in analysis of variance table and mould general introduction table2Value determines fitting effect, by regression coefficient and significance test
Dependency in table and synteny statistic determine whether there is multiple conllinear problem.If there is multiple conllinear problem, need
Dependent variable is screened.In order to overcome conllinear problem, simplified model, increase prediction accuracy, the present invention selects progressively first
Homing method, realizing step is:On the basis of aforesaid operations, select Stepwise (progressively) in Method frame, then click on OK,
Output result.
Output result explanation:
(1)R2For the coefficient of determination, it reflects regression effect, better closer to 1 effect;
(2) in variance analyses, F checks for regression effect, and F value bigger explanation regression effect is better;
(3) F test value Sig.<The regression coefficient of 0.05 explanation at least 1 independent variable is not zero, the recurrence mould set up
Type is statistically significant;
(4) significance being directed to the t inspection of regression coefficient determines that can relevant variable enter recurrence side as explanatory variable
Journey;
(5) t test value Sig.<0.05 explanation relevant variable coefficient has statistical significance;
1.2 meteorological factor influence analyses
To somewhere 2013-2014 daily peak load quarterly, day minimum load, per day load with each meteorological because
The relation of element has carried out stepwise regression analysis showed, and regression analyses parameter is shown in Table 1, table 2, and X1, X2, X3, X4, X5 in each table divide
Do not represent max. daily temperature in meteorological factor, Daily minimum temperature, mean daily temperature, day relative humidity, daily rainfall.
1 2013 season fraction of the year of table multivariate regressive analysis
2 2014 season fraction of the year of table multivariate regressive analysis
Note:Represent Regression Analysis Result not statistically significant
Parameter R, F value in summary analysis 1, table 2 Regression Analysis Result and its significance test value, regression coefficient t
Test value, can obtain and affect daily peak load, day minimum load, the main meteorological of per day load respectively under regional 2 different seasons
Factor is:
(1) main weather factor of impact spring daily peak load and day minimum load is minimum temperature and mean temperature,
The main weather factor of impact spring per day load is mean temperature;
(2) main weather factor of impact summer daily peak load is mean temperature and relative humidity, and impact day summer is
The main weather factor of underload is mean temperature and rainfall;
(3) main weather factor of impact autumn daily peak load is mean temperature and relative humidity, and impact day summer is
The main weather factor of underload is mean temperature and rainfall;
(4) main weather factor of impact winter daily peak load is maximum temperature and relative humidity, and impact is per day negative
The main weather factor of lotus is maximum temperature.
1.3 meteorological similar day systems of selection
Results by multivariate regression analysis meteorological factor being carried out by 2013-2014 season fraction of the year can be definitely different
The main weather factor affecting load variations in season is different.In conjunction with the qualitative analyses to influence factor above, the present invention
Quantitative analyses are carried out to loading effects factor using this index of partial correlation coefficient.
The present invention have chosen max. daily temperature, Daily minimum temperature, mean daily temperature, day relative humidity, daily rainfall this five
Individual meteorological index constitutes day character vector, for carrying out the selection of similar day.Due to the impact journey to electric load for each factor
Degree is not quite similar, and there is also certain influencing each other between each meteorological factor, therefore selects partial correlation coefficient as each gas of calculating
Important evidence as factor weight.Calculate the weight shared by each index in five indexs, formula is as follows:
Wherein, γiPartial correlation coefficient for each index and the per day load in this area.
Determine the weight coefficient of meteorological factor by season, as shown in table 3:
Table 3 meteorological factor weight coefficient
Day character vector and the prediction day of i-th day after obtaining each factor weight, is calculated using Weighted Similarity formula
Weighted Similarity sim (i) of day character vector, formula is as follows:
Wherein, pjI () is the numerical value after i-th day j-th influence factor's normalization, ε is a smaller number.
Calculate all history days in sample and the Weighted Similarity of prediction day, carry out descending, select to come before
History day as load prediction similar day.
2 forecast error sample values and the generation of offset
2.1 forecast error distributions
Above the regression analyses of meteorological factor be may indicate that with the meteorological factor of this area and the significantly correlated of load variations
Property, the polytropy of meteorological condition simultaneously makes the meteorological condition of each history day mutually variant.Find through statistical study, meteorological condition
Distribution to forecast error also has a certain impact.According to the weather data analysis in sample set, this area is throughout the year
The climatic characteristic having nothing in common with each other, the impact to load variations is also different, the mistake being produced using same load forecasting method
Difference cloth also has certain change because climatic characteristic is different.The present invention presses season type analysis this area and produces in prediction
Error characteristic distributions, establish history error sample set, and carried out error fit by season it will therefore be readily appreciated that either which
Plant season, the distributed image of relative error is all approximately symmetrical with regard to y-axis, and probability density function generally subtracts with the increase of | x |
Little, and level off to x-axis.The error distribution in season in summer in winter two is concentrated the most, i.e. variances sigma2 Winter、σ2 SummerMinimum, figure relatively " high thin ", and
Expected value μSummerClosest to zero, winter expected value μWinterIt is significantly greater than zero;In comparison, spring and autumn distribution situation dispersion, with the spring
Season error distribution curve the most " short and stout ".
2.2 forecast error sampling
Based on systematic sampling, the relative error probability density in similar day error sample set under each season of matching is bent
Line is sampled.
Using systematic sampling, forecast error is sampled, the principle of systematic sampling is as follows:
When the sample that the totally middle extraction capacity for N for the capacity is n, if being totally divided into uniform stem portion, then press
According to the rule pre-establishing, extract an individual from each section, obtain required sample.It is characterized by:
1) overall capacity N is larger;
2) pitch requirements equal (generally K=N/n) of overall segmentation;
3) in each sample segments using stochastic sampling it is ensured that the probability that is pumped to of each sample is equal.
In sampling, the error information in similar day error sample set was divided by the present invention according to the period, by one day
24 hours be divided into 6 periods it is ensured that sampling uniformly carry out in day part.It should be noted that because big error goes out
Existing probability is low, and in order to avoid, sampling resultses are excessive to cause error overcompensation phenomenon, chooses in each matched curve centered on peak value,
Probability be 85% part as actual sampling interval.
2.3 are compensated based on the forecast error of undulatory property analysis
By analyze sample in history day forecast error fluctuation direction and amplitude regularity, permissible by predicting means
The variation tendency of anticipation error simultaneously produces corresponding offset.
In undulatory property analysis, choose n long-term relative error as the criterion of recent relative error analysis.Definition
The long-term horizontal σ of variancelMarginal value k with fitting a straight line slope absolute valuel:
In formula, k1And k2It is respectively the upper and lower marginal value in unilateral confidence interval being determined by model of fit and confidence level.
The relative error obtaining front 3 data points of future position, as the sample value of recent relative error, calculates its variances sigmas
And the slope absolute value k of fitting a straight lines, in conjunction with the two just can approximate analyses last-period forecast relative error undulatory property, can obtain in advance
The estimated value of measuring point relative error.Specific analytical method is as shown in table 4.
The undulatory property analysis method of table 4 relative error
In order to avoid undulatory property analysis introduces new error, set up error compensation principle.Base in relative error estimated value
On plinth, according to future position error deltaiWith previous moment error deltai-1With the threshold values error delta dividing big error and little error0Relatively large
Little, determine the compensation way of relative error, as shown in table 5.
Table 5 relative error compensation way
Compensate error amount size based on undulatory property analysis gained the sampling resultses of probabilistic model are ranked up, choose and ripple
Dynamic property analyzes the immediate sampling resultses of offset as the correction value of relative error, realizes the error correction to prediction, thus
The precision of short-term forecast can be improved.
3 Short-term Load Forecasting Models based on error correction
The thought of error correction is introduced in short-term load forecasting research the present invention, by being simulated life to forecast error
Become error fit value, and it is overlapped with prediction load, finally give the predicted load after error correction.
The present invention is primarily based on SPSS and carries out regression analyses to the main weather factor of impact this area load variations, is dividing
Consider the Seasonal Characteristics of meteorological condition change in analysis, carry out load prediction work, concrete thought such as Fig. 2 institute based on this
Show.
Wherein, in the load prediction link setting up history error sample set, the present invention have selected that structure is simple, has relatively
Good self-learning capability and the wavelet neural network prediction algorithm of very fast convergence rate, wavelet neural network Algorithm for Training step is such as
Under:
Step 1:Netinit.When starting training, first wavelet neural network should be initialized, that is, initialize
The parameters of wavelet function.
Step 2:Sample classification, including carrying out the training sample of network training and the test sample of measuring accuracy.
Step 3:Prediction output.Training sample is inputted network, the mistake between comparing cell prediction output and desired output
Difference e.
Step 4:Update weights.After obtaining the error e of previous step, by the parameters of wavelet function and intermediate layer
Weights are further updated, so that predict the outcome more meeting expection.
Step 5:Judge whether prediction output is close enough with expected value, if being not over, return to step 3.
The flow chart of wavelet algorithm program is as shown in Figure 3:
Embodiment two
In order to verify the suitability of the load prediction principle containing simulation relative error, example adopts the load data in somewhere
It is predicted with meteorological data, and emulated using MATLAB programming.
The present embodiment choose each one day of this area's four seasons spring, summer, autumn and winter in 2014 as prediction day, be denoted as respectively A1, A2,
A3、A4.For a certain prediction day, determine its affiliated season, with real time meteorological data for selecting the foundation of similar day, and then set up
Similar day forecast error sample set, carries out uniform sampling based on systematic sampling, carries out sampling resultses according to undulatory property analysis
Sequence, Fig. 4 be A1~A4 undulatory property analyze gained compensate error result with sequence after final error of fitting.
Using matching relative error, wavelet neural network is predicted the outcome and be modified, the load obtaining containing error of fitting is pre-
Measured value.Fig. 5 is finally predicting the outcome of A1~A4, and this figure shows:After being combined with error of fitting, the prediction in each season predicts the outcome
Precision of prediction all can be improved to a certain extent.Can be seen that relative error matching distribution sample value in conjunction with Fig. 4, Fig. 5 permissible
Effectively eliminate the extreme value occurring in undulatory property analysis, it is to avoid error dissipates and overcompensation phenomenon, and undulatory property analysis possesses
Ability according to a preliminary estimate to the offset of relative error, is ranked up to sampling error based on it, can by predicted load with
Matching relative error magnitudes accurately correspond to, and then reach the compensation to error, improve the effect of precision of prediction.
Understand, it is related to load data to meteorological factor in Various Seasonal that the present invention is primarily based on SPSS in conjunction with the embodiments
Property is analyzed, and carries out the selection of similar day based on this, and the historical forecast error using similar day sets up the general of relative error
Rate density function, is sampled to historical forecast error sample set by systematic sampling, and by sampling resultses according to undulatory property
Analysis is ranked up, and finally gives the match value of load prediction relative error, and it can be reached correction with predictive value is superimposed
Predictive value, improves the effect of precision of prediction.Demonstrate practicality and the effectiveness of institute of the present invention extracting method by embodiment.
Claims (10)
1. the short-term load forecasting method based on meteorological similar day and error correction it is characterised in that:Its step is:
Initially set up multiple linear regression model, by historical data:Load data, meteorological data, carry out the pretreatment of data, then
According to multiple linear regression model, stepwise regression analysis showed is carried out to historical data, select impact load in each season the most notable
Meteorological factor, determine the weight of each factor and in this, as the foundation selecting meteorological similar day;Set up history by season type
Prediction error data sample set, for a certain prediction day, extracts the error information Sample Establishing set of its similar day, and presses season
Probability density distribution matching is carried out to it;The fluctuating error situation of future position is analyzed, obtains forecast error offset, choosing
Take the error sample value closest to forecast error offset as the error fit value in this moment, and be added on predictive value, make
For final load prediction results.
2. the short-term load forecasting method based on meteorological similar day and error correction according to claim 1, its feature exists
In:Described multiple linear regression model is to be set up based on principle of least square method, and the analytical parameters of its model are that reaction returns
Regression effect test value in the coefficient of determination of effect, variance analyses, its significance test value, the test value of regression coefficient.
3. the short-term load forecasting method based on meteorological similar day and error correction according to claim 1, its feature exists
In:Described load data is daily peak load, day minimum load, per day load;Described meteorological data is max. daily temperature, day
Minimum temperature, mean daily temperature, day relative humidity, daily rainfall.
4. the short-term load forecasting method based on meteorological similar day and error correction according to claim 1, its feature exists
In:The determination method of described forecast error offset is to choose multiple long-term relative erroies as near in history day forecast error
The criterion of phase relative error analysis, then calculates long-term variance level and fitting a straight line slope absolute value, near in conjunction with the two
Like the undulatory property analyzing last-period forecast relative error, obtain the estimated value of future position relative error;Finally according to prediction point tolerance
With the relative size of previous moment error, the threshold values error dividing big error and little error, determine relative error offset.
5. the short-term load forecasting method based on meteorological similar day and error correction according to claim 1, its feature exists
In:Described error sample value is sampled in the error information Sample Establishing set of similar day using systematic sampling;It is
The step of system sampling method is as follows:
When the sample that the totally middle extraction capacity for N for the capacity is n, if being totally divided into uniform stem portion, then according in advance
The rule first formulated, extracts an individual from each section, obtains required sample;
Wherein,
1) overall capacity N is larger;
2) pitch requirements equal (generally K=N/n) of overall segmentation;
3) in each sample segments using stochastic sampling it is ensured that the probability that is pumped to of each sample is equal.
6. the short-term load forecasting method based on meteorological similar day and error correction according to claim 1, its feature exists
In:Described error fit value is according to forecast error offset size, error sample value to be ranked up, and chooses and forecast error
Offset immediate error sample value is as error fit value.
7. the short-term load forecasting method based on meteorological similar day and error correction according to claim 1 and 2, its feature
It is:Described multiple linear regression model is
Y=y '+μ=b0+b1X1+b2X2+...+biXi+...+bkXk(1) .
8. the short-term load forecasting method based on meteorological similar day and error correction according to claim 4, its feature exists
In:The analysis method of the undulatory property of described relative error is to see table:
Wherein, σlFor long-term variance level, σsFor in the recent period relative to the variance of calculation error, klFor fitting a straight line slope absolute value
Marginal value, ksThe slope absolute value of the fitting a straight line calculating for recent relative error.
9. the short-term load forecasting method based on meteorological similar day and error correction according to claim 4, its feature exists
In:The determination method of described error compensation value see table:
Wherein, δiFor predicting point tolerance, δi-1For future position previous moment error, δ0Little with future position for dividing the big error of future position
The threshold values error of error.
10. the short-term load forecasting method based on meteorological similar day and error correction according to claim 5, its feature
It is:Described error sample value, in sampling, the error information in similar day error sample set was divided according to the period, will
24 hours of one day are divided into 6 periods it is ensured that sampling is uniformly carried out in day part;
Wherein, choose centered on peak value in each matched curve, probability be 85% part as actual sampling interval.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611079983.8A CN106408223A (en) | 2016-11-30 | 2016-11-30 | Short-term load prediction based on meteorological similar day and error correction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611079983.8A CN106408223A (en) | 2016-11-30 | 2016-11-30 | Short-term load prediction based on meteorological similar day and error correction |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106408223A true CN106408223A (en) | 2017-02-15 |
Family
ID=58084353
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611079983.8A Pending CN106408223A (en) | 2016-11-30 | 2016-11-30 | Short-term load prediction based on meteorological similar day and error correction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106408223A (en) |
Cited By (29)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107423836A (en) * | 2017-04-07 | 2017-12-01 | 山东大学 | Short-term load forecasting method based on sendible temperature |
CN107464015A (en) * | 2017-07-21 | 2017-12-12 | 南京林洋电力科技有限公司 | A kind of short term and power predicating method that possess fusing mechanism based on electricity consumption behavior |
CN107506843A (en) * | 2017-07-03 | 2017-12-22 | 国网上海市电力公司 | A kind of short-term load forecasting method and device |
CN107748933A (en) * | 2017-10-23 | 2018-03-02 | 成都信息工程大学 | Meteorological element message data error correcting method, mist, sunrise, sea of clouds, rime Forecasting Methodology |
CN108229742A (en) * | 2018-01-04 | 2018-06-29 | 国网浙江省电力公司电力科学研究院 | A kind of load forecasting method based on meteorological data and data trend |
CN108734341A (en) * | 2018-04-27 | 2018-11-02 | 广东电网有限责任公司 | A kind of self-organizing Electric Load Forecasting reporting method based on time series arma modeling |
CN109063983A (en) * | 2018-07-18 | 2018-12-21 | 北京航空航天大学 | A kind of natural calamity loss real time evaluating method based on social media data |
CN109146063A (en) * | 2018-08-27 | 2019-01-04 | 广东工业大学 | A kind of more segmentation short-term load forecasting methods based on vital point segmentation |
CN109299814A (en) * | 2018-08-30 | 2019-02-01 | 国网江苏电力设计咨询有限公司 | A kind of meteorological effect quantity division prediction technique |
CN109325622A (en) * | 2018-09-26 | 2019-02-12 | 巢湖学院 | A kind of method of Load Prediction In Power Systems |
CN109389238A (en) * | 2017-08-14 | 2019-02-26 | 中国电力科学研究院 | A kind of short-term load forecasting method and device based on ridge regression |
CN109726862A (en) * | 2018-12-24 | 2019-05-07 | 深圳供电局有限公司 | A kind of user's daily electricity mode prediction method |
CN109858700A (en) * | 2019-02-01 | 2019-06-07 | 华北水利水电大学 | BP neural network heating system energy consumption prediction technique based on similar screening sample |
CN110033134A (en) * | 2019-04-09 | 2019-07-19 | 国网安徽省电力有限公司 | A kind of short-term load forecasting algorithm of segmentation day by day considering meteorologic factor |
CN110097205A (en) * | 2019-03-15 | 2019-08-06 | 天津大学 | A kind of building load prediction weather forecast data preprocessing method |
CN110163429A (en) * | 2019-05-10 | 2019-08-23 | 湖南大学 | A kind of short-term load forecasting method based on similar day optimal screening |
CN110265165A (en) * | 2019-06-18 | 2019-09-20 | 中广核核电运营有限公司 | Nuclear power vessel temp adjusting method, device, computer equipment and storage medium |
CN110598896A (en) * | 2019-07-26 | 2019-12-20 | 陕西省水利电力勘测设计研究院 | Photovoltaic power prediction method based on prediction error correction |
CN110688620A (en) * | 2019-09-11 | 2020-01-14 | 新奥数能科技有限公司 | Short-term load prediction method and device |
CN111144650A (en) * | 2019-12-26 | 2020-05-12 | 南京工程学院 | Power load prediction method, device, computer readable storage medium and equipment |
CN111967655A (en) * | 2020-07-28 | 2020-11-20 | 中国南方电网有限责任公司 | Short-term load prediction method and system |
CN112258337A (en) * | 2020-09-14 | 2021-01-22 | 陕西讯格信息科技有限公司 | Self-complementing and self-correcting base station energy consumption model prediction method |
CN112699600A (en) * | 2020-12-23 | 2021-04-23 | 中国大唐集团科学技术研究院有限公司火力发电技术研究院 | Thermal power operating parameter and NOxAnalysis method for bias between emission concentrations |
WO2021109515A1 (en) * | 2019-12-03 | 2021-06-10 | 江苏智臻能源科技有限公司 | Short-term load prediction method based on association analysis and kalman filtering method |
CN113095542A (en) * | 2021-03-01 | 2021-07-09 | 华中科技大学 | Photovoltaic output power prediction error fitting method and system based on DPMM |
CN113456031A (en) * | 2021-08-09 | 2021-10-01 | 首都医科大学附属北京天坛医院 | Training device and prediction device of brain state prediction model and electronic equipment |
WO2021213192A1 (en) * | 2020-04-22 | 2021-10-28 | 国网江苏省电力有限公司苏州供电分公司 | Load prediction method and load prediction system employing general distribution |
CN115309052A (en) * | 2022-08-19 | 2022-11-08 | 北京全应科技有限公司 | Online correction method for time sequence prediction result of industrial real-time data |
CN116826745A (en) * | 2023-08-30 | 2023-09-29 | 山东海兴电力科技有限公司 | Layered and partitioned short-term load prediction method and system in power system background |
-
2016
- 2016-11-30 CN CN201611079983.8A patent/CN106408223A/en active Pending
Cited By (41)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107423836B (en) * | 2017-04-07 | 2020-04-28 | 山东大学 | Short-term load prediction method based on body sensing temperature |
CN107423836A (en) * | 2017-04-07 | 2017-12-01 | 山东大学 | Short-term load forecasting method based on sendible temperature |
CN107506843A (en) * | 2017-07-03 | 2017-12-22 | 国网上海市电力公司 | A kind of short-term load forecasting method and device |
CN107464015B (en) * | 2017-07-21 | 2021-06-11 | 南京林洋电力科技有限公司 | Short-term load with fusing mechanism and electric quantity prediction method based on electricity utilization behavior |
CN107464015A (en) * | 2017-07-21 | 2017-12-12 | 南京林洋电力科技有限公司 | A kind of short term and power predicating method that possess fusing mechanism based on electricity consumption behavior |
CN109389238B (en) * | 2017-08-14 | 2022-08-09 | 中国电力科学研究院 | Ridge regression-based short-term load prediction method and device |
CN109389238A (en) * | 2017-08-14 | 2019-02-26 | 中国电力科学研究院 | A kind of short-term load forecasting method and device based on ridge regression |
CN107748933A (en) * | 2017-10-23 | 2018-03-02 | 成都信息工程大学 | Meteorological element message data error correcting method, mist, sunrise, sea of clouds, rime Forecasting Methodology |
CN107748933B (en) * | 2017-10-23 | 2020-07-14 | 成都信息工程大学 | Meteorological element message data error correction method and fog, sunrise, cloud sea and rime prediction method |
CN108229742B (en) * | 2018-01-04 | 2021-10-22 | 国网浙江省电力公司电力科学研究院 | Load prediction method based on meteorological data and data trend |
CN108229742A (en) * | 2018-01-04 | 2018-06-29 | 国网浙江省电力公司电力科学研究院 | A kind of load forecasting method based on meteorological data and data trend |
CN108734341A (en) * | 2018-04-27 | 2018-11-02 | 广东电网有限责任公司 | A kind of self-organizing Electric Load Forecasting reporting method based on time series arma modeling |
CN109063983B (en) * | 2018-07-18 | 2022-06-21 | 北京航空航天大学 | Natural disaster damage real-time evaluation method based on social media data |
CN109063983A (en) * | 2018-07-18 | 2018-12-21 | 北京航空航天大学 | A kind of natural calamity loss real time evaluating method based on social media data |
CN109146063A (en) * | 2018-08-27 | 2019-01-04 | 广东工业大学 | A kind of more segmentation short-term load forecasting methods based on vital point segmentation |
CN109299814B (en) * | 2018-08-30 | 2024-02-02 | 国网江苏电力设计咨询有限公司 | Meteorological influence electric quantity decomposition prediction method |
CN109299814A (en) * | 2018-08-30 | 2019-02-01 | 国网江苏电力设计咨询有限公司 | A kind of meteorological effect quantity division prediction technique |
CN109325622A (en) * | 2018-09-26 | 2019-02-12 | 巢湖学院 | A kind of method of Load Prediction In Power Systems |
CN109726862A (en) * | 2018-12-24 | 2019-05-07 | 深圳供电局有限公司 | A kind of user's daily electricity mode prediction method |
CN109858700A (en) * | 2019-02-01 | 2019-06-07 | 华北水利水电大学 | BP neural network heating system energy consumption prediction technique based on similar screening sample |
CN110097205A (en) * | 2019-03-15 | 2019-08-06 | 天津大学 | A kind of building load prediction weather forecast data preprocessing method |
CN110033134A (en) * | 2019-04-09 | 2019-07-19 | 国网安徽省电力有限公司 | A kind of short-term load forecasting algorithm of segmentation day by day considering meteorologic factor |
CN110163429A (en) * | 2019-05-10 | 2019-08-23 | 湖南大学 | A kind of short-term load forecasting method based on similar day optimal screening |
CN110163429B (en) * | 2019-05-10 | 2023-06-09 | 湖南大学 | Short-term load prediction method based on similarity day optimization screening |
CN110265165A (en) * | 2019-06-18 | 2019-09-20 | 中广核核电运营有限公司 | Nuclear power vessel temp adjusting method, device, computer equipment and storage medium |
CN110598896A (en) * | 2019-07-26 | 2019-12-20 | 陕西省水利电力勘测设计研究院 | Photovoltaic power prediction method based on prediction error correction |
CN110688620A (en) * | 2019-09-11 | 2020-01-14 | 新奥数能科技有限公司 | Short-term load prediction method and device |
WO2021109515A1 (en) * | 2019-12-03 | 2021-06-10 | 江苏智臻能源科技有限公司 | Short-term load prediction method based on association analysis and kalman filtering method |
CN111144650A (en) * | 2019-12-26 | 2020-05-12 | 南京工程学院 | Power load prediction method, device, computer readable storage medium and equipment |
WO2021213192A1 (en) * | 2020-04-22 | 2021-10-28 | 国网江苏省电力有限公司苏州供电分公司 | Load prediction method and load prediction system employing general distribution |
CN111967655A (en) * | 2020-07-28 | 2020-11-20 | 中国南方电网有限责任公司 | Short-term load prediction method and system |
CN112258337A (en) * | 2020-09-14 | 2021-01-22 | 陕西讯格信息科技有限公司 | Self-complementing and self-correcting base station energy consumption model prediction method |
CN112258337B (en) * | 2020-09-14 | 2024-03-12 | 陕西讯格信息科技有限公司 | Self-complement correction base station energy consumption model prediction method |
CN112699600A (en) * | 2020-12-23 | 2021-04-23 | 中国大唐集团科学技术研究院有限公司火力发电技术研究院 | Thermal power operating parameter and NOxAnalysis method for bias between emission concentrations |
CN113095542A (en) * | 2021-03-01 | 2021-07-09 | 华中科技大学 | Photovoltaic output power prediction error fitting method and system based on DPMM |
CN113095542B (en) * | 2021-03-01 | 2023-11-14 | 华中科技大学 | Fitting method and system for photovoltaic output power prediction error based on DPMM |
CN113456031A (en) * | 2021-08-09 | 2021-10-01 | 首都医科大学附属北京天坛医院 | Training device and prediction device of brain state prediction model and electronic equipment |
CN115309052A (en) * | 2022-08-19 | 2022-11-08 | 北京全应科技有限公司 | Online correction method for time sequence prediction result of industrial real-time data |
CN115309052B (en) * | 2022-08-19 | 2023-04-28 | 北京全应科技有限公司 | Online correction method for industrial real-time data time sequence prediction result |
CN116826745A (en) * | 2023-08-30 | 2023-09-29 | 山东海兴电力科技有限公司 | Layered and partitioned short-term load prediction method and system in power system background |
CN116826745B (en) * | 2023-08-30 | 2024-02-09 | 山东海兴电力科技有限公司 | Layered and partitioned short-term load prediction method and system in power system background |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106408223A (en) | Short-term load prediction based on meteorological similar day and error correction | |
CN108921339B (en) | Quantile regression-based photovoltaic power interval prediction method for genetic support vector machine | |
CN110232203B (en) | Knowledge distillation optimization RNN short-term power failure prediction method, storage medium and equipment | |
CN110222882A (en) | A kind of prediction technique and device of electric system Mid-long Term Load | |
CN106779223A (en) | A kind of photovoltaic system electricity generation power real-time predicting method and device | |
CN113496311A (en) | Photovoltaic power station generated power prediction method and system | |
CN112257941A (en) | Photovoltaic power station short-term power prediction method based on improved Bi-LSTM | |
CN107944594B (en) | Short-term load prediction method based on spearman grade and RKELM microgrid | |
CN111985701B (en) | Power consumption prediction method based on power supply enterprise big data model base | |
CN106779478A (en) | A kind of load scheduling Valuation Method | |
Pierro et al. | Photovoltaic generation forecast for power transmission scheduling: A real case study | |
CN115545333A (en) | Method for predicting load curve of multi-load daily-type power distribution network | |
CN109858668B (en) | Coordination prediction method for power load region in thunder and lightning climate | |
CN117013527A (en) | Distributed photovoltaic power generation power prediction method | |
CN108694475B (en) | Short-time-scale photovoltaic cell power generation capacity prediction method based on hybrid model | |
CN110489893B (en) | Variable weight-based bus load prediction method and system | |
CN112183877A (en) | Photovoltaic power station fault intelligent diagnosis method based on transfer learning | |
CN112508254A (en) | Method for determining investment prediction data of transformer substation engineering project | |
Gunasekaran et al. | Solar irradiation forecasting using genetic algorithms | |
CN115860797A (en) | Electric quantity demand prediction method suitable for new electricity price reform situation | |
CN112116127B (en) | Photovoltaic power prediction method based on association of meteorological process and power fluctuation | |
CN114529071A (en) | Method for predicting power consumption of transformer area | |
Hoffmann | Temporal aggregation methods for energy system modeling | |
Coelho et al. | A general variable neighborhood search heuristic for short term load forecasting in smart grids environment | |
Zhu et al. | Day-ahead campus load interval forecast based on similar day and kernel function estimation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170215 |