CN106920014A - A kind of short-term load forecasting method and device - Google Patents
A kind of short-term load forecasting method and device Download PDFInfo
- Publication number
- CN106920014A CN106920014A CN201710156989.9A CN201710156989A CN106920014A CN 106920014 A CN106920014 A CN 106920014A CN 201710156989 A CN201710156989 A CN 201710156989A CN 106920014 A CN106920014 A CN 106920014A
- Authority
- CN
- China
- Prior art keywords
- data
- load
- historical
- load forecasting
- abnormal
- 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
- G06Q10/04—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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention discloses a kind of short-term load forecasting method and device, the method includes：Acquisition includes the historical data of environmental data and corresponding load data, and the relation set up between environmental data and load data using nonparametric PDF estimation；Based on the adaptive forgetting factor in Relation acquisition historical data, the time-varying Cook distance statistics amounts based on adaptive forgetting factor are built；The load data in historical data is detected using time-varying Cook distance statistics amount, if detecting abnormal data, abnormal data is repaired based on abnormal data corresponding environmental data using nonparametric PDF estimation；Abnormal data and other historical datas to repairing completion are trained and obtain corresponding load forecasting model, and load forecasting model is optimized with MCSO algorithms, short-term load forecasting is realized with using the load forecasting model, so as to substantially increase the precision of prediction of load forecasting model and short-term load forecasting.
Description
Technical field
The present invention relates to Techniques for Prediction of Electric Loads field, more specifically to a kind of short-term load forecasting method and
Device.
Background technology
Load prediction is one of necessary means of Power Market Development, can accurately predict the electricity at a certain specific moment
Power load value, has important economic implications and social effect to power system.Short-term load forecasting be to certain in a few days each when
The load of (such as daily load curve is made up of 24,48,96,144 or 288 points) is carved, the purpose is to arrange a day startup-shutdown
Group plan and generation schedule.
In the last few years, artificial intelligence and machine learning obtains fast development, scholars in short-term load forecasting research field
A series of researchs are carried out to load prediction by mathematical modeling, it is proposed that many methods for solving the problems, such as load prediction, typically
It is required to use the load data produced in history when above-mentioned mathematical modeling is carried out, but, if in these load datas
In the presence of abnormal data, influence will be produced on mathematical modeling, and then cause the precision of prediction of correspondence forecast model to substantially reduce,
It is corresponding, precision relatively low situation occurs when realizing the prediction of load using this forecast model.
In sum, how a kind of technical scheme of precision of prediction short-term load forecasting higher is provided, is current ability
Field technique personnel's problem demanding prompt solution.
The content of the invention
It is an object of the invention to provide a kind of short-term load forecasting method and device, to reach precision of prediction higher.
To achieve these goals, the present invention provides following technical scheme：
A kind of short-term load forecasting method, including：
Acquisition includes the historical data of environmental data and corresponding load data, and utilizes nonparametric PDF estimation
The relation set up between the environmental data and the load data；
Based on the adaptive forgetting factor in historical data described in the Relation acquisition, build and forgotten based on the self adaptation
The time-varying Cook distance statistics amounts of the factor；
The load data in the historical data is detected using the time-varying Cook distance statistics amount, if detection
Go out abnormal data, be then based on the corresponding environmental data of the abnormal data to described different using nonparametric PDF estimation
Regular data is repaired；
The abnormal data and other described historical datas to repairing completion are trained and obtain corresponding load prediction
Model, and the load forecasting model is optimized with MCSO algorithms, born in short-term with using load forecasting model realization
Lotus is predicted.
Preferably, obtaining includes the historical data of environmental data and corresponding load data, including：
Acquisition includes the historical data of environmental data and corresponding load data, wherein the environmental data includes humidity data
And temperature data.
Preferably, based on the adaptive forgetting factor in historical data described in the Relation acquisition, including：
The adaptive forgetting factor in historical data described in the Relation acquisition is based on using recurrent least square method.
Preferably, the abnormal data and other described historical datas of repairing completion are trained and obtain corresponding negative
Lotus forecast model, including：
The abnormal data and other described historical datas of repairing completion are trained and obtain corresponding using ORELM
Load forecasting model.
A kind of short-term load forecasting device, including：
Processing module, is used for：Acquisition includes the historical data of environmental data and corresponding load data, and general using nonparametric
The relation that rate estimation of density function is set up between the environmental data and the load data；
Module is built, is used for：Based on the adaptive forgetting factor in historical data described in the Relation acquisition, structure is based on
The time-varying Cook distance statistics amounts of the adaptive forgetting factor；
Repair module, is used for：The load data in the historical data is entered using the time-varying Cook distance statistics amount
Row detection, if detecting abnormal data, is based on the abnormal data corresponding using nonparametric PDF estimation
Environmental data is repaired to the abnormal data；
Training module, is used for：The abnormal data and other described historical datas to repairing completion are trained and obtain
Corresponding load forecasting model, and the load forecasting model is optimized with MCSO algorithms, to utilize the load prediction
Model realization short-term load forecasting.
Preferably, the processing module includes：
Acquiring unit, is used for：Acquisition includes the historical data of environmental data and corresponding load data, wherein the environment number
According to including humidity data and temperature data.
Preferably, the structure module includes：
Construction unit, is used for：Using recurrent least square method based on adaptive in historical data described in the Relation acquisition
Answer forgetting factor.
Preferably, the training module includes：
Training unit, is used for：The abnormal data and other described historical datas of repairing completion are carried out using ORELM
Training obtains corresponding load forecasting model.
The invention provides a kind of short-term load forecasting method and device, wherein the method includes：Acquisition includes environment number
According to and corresponding load data historical data, and set up the environmental data and described using nonparametric PDF estimation
Relation between load data；Based on the adaptive forgetting factor in historical data described in the Relation acquisition, build and be based on institute
State the time-varying Cook distance statistics amounts of adaptive forgetting factor；Using the time-varying Cook distance statistics amount to the historical data
In load data detected, if detecting abnormal data, be based on using nonparametric PDF estimation described
The corresponding environmental data of abnormal data is repaired to the abnormal data；To repairing the abnormal data and other institutes that complete
State historical data and be trained and obtain corresponding load forecasting model, and the load forecasting model is carried out with MCSO algorithms
Optimization, short-term load forecasting is realized with using the load forecasting model.Above-mentioned technical proposal provided in an embodiment of the present invention, in profit
Carried out with the historical data for getting before the foundation of load forecasting model, it is necessary to using the time-varying based on adaptive forgetting factor
Whether there is abnormal data in Cook distance statistics amount detection load data, and be based on using nonparametric PDF estimation
The corresponding environmental data of abnormal data for detecting is repaired to abnormal data, with what is completed using reparation after the completion of reparation
Abnormal data and other whole historical data training load forecast models, and load forecasting model is carried out with MCSO algorithms excellent
Change, short-term load forecasting is realized using the load forecasting model, so as to avoid abnormal data from producing corresponding load forecasting model
Raw influence, substantially increases the precision of prediction of load forecasting model, further increases and is realized using the load forecasting model
The precision of prediction of short-term load forecasting.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
The accompanying drawing to be used needed for having technology description is briefly described, it should be apparent that, drawings in the following description are only this
Inventive embodiment, for those of ordinary skill in the art, on the premise of not paying creative work, can also basis
The accompanying drawing of offer obtains other accompanying drawings.
Fig. 1 is a kind of flow chart of short-term load forecasting method provided in an embodiment of the present invention；
Fig. 2 is the first result of different Forecasting Methodologies in a kind of short-term load forecasting method provided in an embodiment of the present invention
Schematic diagram；
Fig. 3 is second result of different Forecasting Methodologies in a kind of short-term load forecasting method provided in an embodiment of the present invention
Schematic diagram；
Fig. 4 is the change feelings of prediction degree/day and humidity in a kind of short-term load forecasting method provided in an embodiment of the present invention
The schematic diagram of condition；
Fig. 5 is that the prediction knot before and after cumulative effect is considered in a kind of short-term load forecasting method provided in an embodiment of the present invention
The schematic diagram of fruit contrast；
Fig. 6 is a kind of structural representation of short-term load forecasting device provided in an embodiment of the present invention.
Specific embodiment
Below in conjunction with 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
Site preparation is described, it is clear that described embodiment is only a part of embodiment of the 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 creative work is not made
Embodiment, belongs to the scope of protection of the invention.
Fig. 1 is referred to, it illustrates a kind of flow chart of short-term load forecasting method provided in an embodiment of the present invention, can be with
Comprise the following steps：
S11：Acquisition includes the historical data of environmental data and corresponding load data, and utilizes nonparametric probability density function
The relation that estimation is set up between environmental data and load data.
It should be noted that historical data is the data for having produced in history, and to the data sampling of historical data frequently
Rate can be 30 minutes, historical data include environmental data and load data, wherein environmental data and load data be it is corresponding,
Namely environmental data at a time is corresponding with the load data at the moment.Using nonparametric PDF estimation
(non-parameter probability density function estimation, NPDF) sets up environmental data and bears
Relation between lotus data, namely the corresponding relation between environmental data and load data is got, by the relation set up,
The load data at the moment can be learnt by the environmental data at a certain moment；Can be realized for ring by the foundation of the relation
The labor of influence of the border data to load data, to ensure the smooth realization of subsequent step.Specifically, due to load
It is different with the situation of change of environmental factor in Various Seasonal with seasonal characteristic, i.e. load, therefore step S11 obtaining
After getting historical data, the generation time that historical data is based on the historical data can also be divided into Various Seasonal, Jin Erjian
The relation of environmental data and load data in vertical Various Seasonal, hence for load point to be detected, it is possible to use its correspondence season
Save the relation set up and judge whether the load point is exceptional value, namely judge whether the load data of the load point is abnormal number
According to.
S12：Based on the adaptive forgetting factor in Relation acquisition historical data, build based on adaptive forgetting factor when
Become Cook distance statistics amounts.
Adaptive forgetting factor in historical data can be got based on above-mentioned relation, and then forgotten based on the self adaptation
The factor builds corresponding time-varying Cook distance statistics amount (time-varying Cook distance, TVCD), wherein introducing certainly
Adapting to forgetting factor can carry out self-adoptive trace to real-time load data, so that it is determined that going out load abnormal in load data
Data (i.e. abnormal data), and then for the follow-up reparation to the data.
S13：The load data in historical data is detected using time-varying Cook distance statistics amount, if detected different
Regular data, then repaiied based on the corresponding environmental data of abnormal data using nonparametric PDF estimation to abnormal data
It is multiple.
The load data in historical data is detected using time-varying Cook distance statistics amount, if not detecting exception
Data, then can directly carry out the training of load forecasting model using historical data, otherwise, then need the abnormal number to detecting
According to being repaired, abnormal data is repaired using nonparametric PDF estimation in the application, specifically, can be with
The stochastic model of real-time load data and environmental data is built using nonparametric PDF estimation, is asked with using the model
Obtaining environmental data corresponding with abnormal data has the load data of corresponding relation as the replacement data of abnormal data, so as to realize
The reparation of abnormal data.In simple terms, step S11 to step S13 is by taking into full account history real time environmental data to load number
According to influence, it is right using nonparametric PDF estimation, it is contemplated that the time series characteristic of seasonal move and random fluctuation
Abnormal load data carries out self adaptation reparation.It is further to note that the abnormal data in the application can refer to that appearance is different
Often or with the strong data for influenceing, can mainly include two class data：(1) number of power information acquisition system collection mistake
According to；(2) load part Outlier Data, i.e. noise sample.
S14：Abnormal data and other historical datas to repairing completion are trained and obtain corresponding load forecasting model,
And load forecasting model is optimized with MCSO algorithms, realize short-term load forecasting with using the load forecasting model.
Historical data to including the abnormal data repaired is trained and obtains corresponding load forecasting model, and transports
It is pre- to load with MCSO (modified crisscross optimization algorithm, improved intersection in length and breadth) algorithm
The model parameter for surveying model is optimized, so as to need the environmental data of prediction time as the input of the model, you can
To the load data of the prediction time of prediction, so as to realize the prediction of prediction time load data.
Above-mentioned technical proposal provided in an embodiment of the present invention, load forecasting model is being carried out using the historical data for getting
Foundation before, it is necessary to using whether there is in the time-varying Cook distance statistics amount detection load data based on adaptive forgetting factor
Abnormal data, and the corresponding environmental data of abnormal data for detecting is based on to exception using nonparametric PDF estimation
Data are repaired, with pre- using the abnormal data and other whole historical data training loads of repairing completion after the completion of reparation
Model is surveyed, short-term load forecasting is realized using the load forecasting model, so as to avoid abnormal data to corresponding load prediction mould
The influence that type is produced, substantially increases the precision of prediction of load forecasting model, further increases using the load forecasting model
Realize the precision of prediction of short-term load forecasting.The model parameter of load forecasting model is optimized using MCSO algorithms in addition,
The precision of prediction of load forecasting model can further be improved.
A kind of short-term load forecasting method provided in an embodiment of the present invention, acquisition includes environmental data and corresponding load data
Historical data, can include：
Acquisition includes the historical data of environmental data and corresponding load data, and wherein environmental data includes humidity data and temperature
Degrees of data.
In technical scheme provided in an embodiment of the present invention, environmental data can include temperature and humidity, certainly can be with root
According to setting is actually needed, within protection scope of the present invention.From temperature and humidity as environmental data, energy in the application
Enough quick obtainings for realizing required data, and consider temperature Thermal incubation effect and humidity Thermal incubation effect pair simultaneously in this application
The influence of electric load, further increases the precision of prediction of short-term load forecasting.
A kind of short-term load forecasting method provided in an embodiment of the present invention, based on the self adaptation in Relation acquisition historical data
Forgetting factor, can include：
The adaptive forgetting factor in Relation acquisition historical data is based on using recurrent least square method.
Wherein, the adaptive forgetting factor in Relation acquisition historical data, Ye Ji are based on using recurrent least square method
Corresponding recurrence least square model is set up using recurrent least square method during opening relationships, and then by being got in the model
Adaptive forgetting factor, so as to realize using acquisition of the recurrent least square method to adaptive forgetting factor.Specifically, can be with
Including：
The relation of Various Seasonal temperature, humidity and load data is set up using nonparametric PDF estimation, is introduced
Real-time load data respectively with recurrence least square (recursive least square, RLS) model of temperature and humidity：
Wherein, ytIt is real-time load data；For mode input becomes
Amount；βtIt is the parameter of model；TtAnd HtTemperature variable and humidity variables are represented respectively；ηtIt is independent identically distributed stochastic variable sequence
Row, and E (ηt)=σ2＜ ∞ wherein E () are represented and are asked expectation, σ2Represent variance；Q represents the preceding q moment of prediction time
Load value, temperature and humidity as model input variable.
The iterative process such as following formula (2) and (3) that parameter vector is estimated：
Wherein,It is βtParameter Estimation；It is gain matrix predicted value；It is model predictive error.
From formula (2) and (3),Recurrence value it is relevant with λ, select appropriate λ to have the adaptive process of model
Important influence, λ is the adaptive forgetting factor of acquisition.
Based on the adaptive forgetting factor in Relation acquisition historical data, the time-varying based on adaptive forgetting factor is built
Cook distance statistics amount is simultaneously detected to the load data in historical data using time-varying Cook distance statistics amount, can wrapped
Include：
In above-mentioned recursive procedure, the time-varying Cook distance statistics amounts based on adaptive forgetting factor are given：
In formula,WithWhen respectively adaptive forgetting factor is not applied in recursive procedureWithValue；M
It is vectorLength.During t λ=1, parameter matrix fromArriveChange influenceed by the load value of t,
Illustrate that historical load has strong influence power；It is σ2Consistent Estimation, have
Cook is apart from CtAdaptive forgetting factor λ is set come self adaptation using nonlinear measurementt, by minimum threshold λminCome
Improve the robustness of RLS algorithm.For big-sample data, Cook distances are combined estimation λ with chi square distribution probabilityt, i.e.,
In formula：P is chi square distribution probability function, ΛtFor the result that Cook distances are calculated using chi square distribution；For
Obey the chi square distribution of free degree m.
Formula (5) by Cook range conversions to interval [0,1], by setting the threshold value of adaptive forgetting factor, adaptive
The negative effect of weakening abnormal data (exceptional value or influential cases) in recursive process is answered, and then ensures the stabilization of adaptive process
Property.
Adaptive forgetting factor expression formula based on Cook distances is：
λt=λmin+(1-λmin)Λt (6)
In formula：λminIt is the minimum value of adaptive forgetting factor.
Time-varying Cook distance statistics amounts based on adaptive forgetting factor are drawn by formula (4), is distributed by Cook distances and F
Probability estimated, under given confidence level α, if the Cook distances of t meet inequality (7), i.e.,
Ct＞ F (α, p, n-p) (7)
Then can be determined that the moment load is abnormal data (exceptional value or influential cases).
In formula, F (α, p, n-p) is to obey the free degree for the F of p and n-p is distributed；N is load sample number；P is mould
Type input variable number.In this case it is necessary to by automatically adjusting adaptive forgetting factor λtTo redefine model ginseng
Number, and need to be repaired by NPDF for abnormal data.
Abnormal data is repaiied based on abnormal data corresponding environmental data using nonparametric PDF estimation
It is multiple, can include：
According to the load abnormal data (exceptional value or influential cases) for detecting, abnormal data is repaiied using NPDF
It is multiple, it is ensured that the continuity and integrality of data.Present problem is converted under conditions of given temperature and humidity, the estimation of load
Value.
The loading condiction probabilistic model of given temperature, humidity：
ft(y | T)=κTft(y,T) (10)
ft(y | H)=κHft(y,H) (11)
Wherein, κT、κHIt is constant, is tried to achieve by below equation：
Wherein, y represents load value, ymax、yminRespectively load maximum and minimum value, T and H are respectively temperature and humidity
Variable.
By the temperature and humidity under preset time, the desired value of load is tried to achieve：
The estimate of load：
Wherein,The average value of the total number of days daily load of modeling is participated in for each season.In the application, t abnormal load
Estimate be the average value of t load and Temperature estimate value, load and humidity estimate.
A kind of short-term load forecasting method provided in an embodiment of the present invention, to repairing the abnormal data and other history that complete
Data are trained and obtain corresponding load forecasting model, can include：
The abnormal data and other historical datas of repairing completion are trained and obtain corresponding load prediction using ORELM
Model.
Specifically, the step can include：
Give N number of training datasetWherein xi=[xi1,xi2,…,xir]TIt is input vector, r represents input
The number of variable, i represents hidden layer node number, oiIt is corresponding desired output.For includingIndividual hidden layer, activation
Function can be expressed as the extreme learning machine Mathematical Modeling of g (x)：
Wherein wi=[wi1,wi1,…,w1r T] it is the weights for connecting i-th hidden layer neuron and input layer,It is i-th deviation of implicit node.βi=[βi1,βi2,…,βir]TTo connect the i-th hidden layer neuron
With the weights of output layer.
Formula (20) can be write a Chinese character in simplified form with the form of matrix：
H β=O (21)
Wherein,
In formula, H is that hidden layer exports layer matrix.
Export weights solution be ensure loss function obtain minimum value, introduce adjustment factor γ come weigh training error and
Output weights, using the sparse characteristic of 1- norms, are improved, the damage after improvement on the basis of classical ORELM loss functions
Lose function as follows：
The optimal solution in (24) is solved using augmentation Lagrangian.Build Lagrange's equation：
In formula, π is Lagrange factor, and μ is punishment, this takes μ=2N/ | | O | |1,||·||1Represent 1- norms, e
It is training error.
Obtain recursive function：
According to recurrence, (e can be obtainedk,βk) relation it is as follows：
βk+1=(HTH+2/γμI)-1HT(O-ek+πk/μ) (27)
Wherein, U=O-H βk+1+πk/ μ is auxiliary parameter.
Training error e is introduced, using error feedback come adaptive updates output layer weights β, being formed has abnormal robustness
Forecast model.The effect of 1- norms is to reduce the interference of exceptional value the characteristics of openness using its.
A kind of short-term load forecasting method provided in an embodiment of the present invention, is carried out with MCSO algorithms to load forecasting model
Optimization, can include：
It is assumed that parent particle X (i) and X (j) carry out lateral cross in d dimensions, computing formula is as follows：
Wherein, r1And r2To be uniformly distributed in the random number of [0,1], c1And c2To be uniformly distributed in the random number of [- 1,1],
MShc(i, d) and MShc(j, d) is that X (i, d) and X (j, d) intersects the filial generation for producing.
The population that updates after lateral cross as crossed longitudinally parent population will be performed, it is assumed that particle X (i) is in d1
And d2Dimension carries out crossed longitudinally, and computing formula is as follows：
Wherein, r is the random number for being uniformly distributed in [0,1]；D is the dimension sum of particle；M is population quantity.Will be laterally
Intersection is at war with the golden mean of the Confucian school solution of crossed longitudinally generation with parent particle, obtains optimal population.
By the variable τ in Population adaptation angle value2Carry out adjust automatically pvc, wherein τ2Expression formula it is as follows：
Wherein fiIt is the fitness value of particle i, favgIt is the average value of now all fitness, n is particle number；
Finally, crossed longitudinally Probability p can be obtained using formula (32)vcLinear representation：
Wherein, pvcmaxAnd pvcminIt is crossed longitudinally Probability pvcMaximum and minimum value,Represent the τ of t2Value；With
This obtains load prediction optimal models.
To consider hidden layer numberInfluence with adjustment factor γ values to model accuracy, due to horizontal and vertical intersection
There is the result that probability influences whether optimization.In order to seek to optimize, in most cases can be by lateral cross Probability phcIf
It is set to 1, but crossed longitudinally Probability pvcIt is difficult to set in advance, because pvcValue sometimes influence whether the overall situation of MCSO algorithms
Search.
Load data is predicted using robust extreme learning machine (extreme learning machine, ORELM) algorithm
And model parameter is optimized using MCSO (modified crisscross optimization algorithm) algorithm；
Precision of prediction can be improved, and there is self-adapting detecting and repair ability to abnormal load.
It is further to note that the above-mentioned technical proposal that the present invention is provided can be as follows with concrete application：
Using the real-time load data in certain electric power enterprise in December, -2010 in January, 2006, real time temperature and real-time humidity,
The real-time load data of -2009 years 2006 is trained to load forecasting model, by setting up load forecasting model come pre-
The Real-time Load value of 2010 is surveyed, the actual Real-time Load value of 2010 is verified to load forecasting model.
Following confidence level α=0.0 of TVCD parameter settings；5 adaptive forgetting factor minimum value λmin=0.6,0.7,；
MSCO parameter settings are as follows：Population quantity is 20；Maximum iteration is 500；phc=1；pvcmaxAnd pvcminRespectively
It is taken as 0.8 and 0.2；ParameterAccording to obtain come optimal value take respectively 5 and 2 multiple.The optimal ginseng of different Forecasting Methodologies
Number is as shown in table 1：
The various method optimized parameters of table 1
Found by table 1, load forecasting model optimized parameter in technical scheme (TVCD-NPDF-ORELM) disclosed in the present application
Hidden layer neutral net number is minimum, so can greatly improve predicted time.
λ is verified with the Real-time Load of 2010min=0.6, to the influence of load prediction precision when 0.7,0.8, and use
Mean absolute percentage error (mean absolute percentage error, MAPE) evaluation index is smart to characterize prediction
Degree, i.e.,
In formula：yi、tiIt is the actual value and predicted value of i moment electric loads；N is prediction number of days.
The result of contrast such as following table：
The difference of table 2 λminPredict the outcome
As shown in Table 2, λminThe number of abnormal (or strong influence) load value is detected when=0.8 at most (27), is passed through
Precision of prediction highest (MAPE values are 1.13%) after NPDF reparations, embodies technical scheme disclosed in the present application for exception
The repair ability of (or strong influence) load value.
In order to verify the precision of TVCD-NPDF-ORELM load forecasting methods proposed by the present invention, by it with ELM algorithms,
RELM algorithms, predicting the outcome for WRELM algorithms are compared, and consider that abnormal load repairs both front and back situation, as a result such as table 3
It is shown：
Various method precision of predictions compare before and after the abnormal load reparation of table 3
Known by table 3, every kind of method precision of prediction before and after abnormal load reparation is all significantly improved, and especially carries herein
The TVCD-NPDF-ORELM method precision for going out obtains larger raising, and this is also demonstrated
The characteristics of ORELM algorithms introduced above have antijamming capability to load exceptional value.
In order to verify the precision of TVCD-NPDF-ORELM load forecasting methods proposed by the present invention, it is considered to extreme weather because
The influence of element, and it is compared with ELM algorithms, RELM algorithms, predicting the outcome for WRELM algorithms：
The prediction under a certain extreme weather of summer in 2010 day (48 load points) is chosen to be analyzed.The prediction day highest
Temperature, mean daily temperature are higher by 9.7 DEG C and 8.3 DEG C respectively than the same period in former years, and relative humidity is higher by 23.2%, the Er Qiefa of the same period
It is existing, 10.5 DEG C have been higher by than the prediction day equal temperature of previous balance, belong to typical extreme weather phenomenon, four kinds of methods predict the outcome
As shown in Figures 2 and 3.
Obviously, other three kinds smaller, corresponding MAPE of robust TVCD-NPDF-ORELM Forecasting Methodologies predicated error fluctuation ratio
It is 0.92% to be worth, and illustrates the suddenly change of climatic factor, and robust TVCD-NPDF-ORELM methods show very strong anti-interference energy
Power, adaptive correction model parameter tackles this burst weather phenomenon, so as to improve precision of prediction.
Load to a certain week of Summer High Temperature in 2010 is predicted analysis, the Zhou Tianqi present continuous high temperature with
Dampness, the duration is 3 days (144 load points), and Fig. 4 is the situation of change for predicting degree/day and humidity, and Fig. 5 show
Consider the contrast that predicts the outcome before and after cumulative effect.
Found by Fig. 4, the temperature of 3 prediction days of selection is first to raise, and is slowly declined after reaching 31.9 DEG C of maximum temperature,
And decline process has fluctuation again.When not considering meteorologic factor cumulative effect, due to the lasting rising of temperature, cause high in electricity consumption
Precision of prediction during the peak phase is not high, and with the rising of temperature, predicted load is more relatively low than actual value (Fig. 5), and MAPE values are
2.96%.But, it is considered to after meteorologic factor cumulative effect, forecast model input variable is modified using NPDF so that pre-
Survey precision to significantly improve, MAPE values are 1.25%.
The embodiment of the present invention additionally provides a kind of short-term load forecasting device, as shown in fig. 6, can include：
Processing module 11, is used for：Acquisition includes the historical data of environmental data and corresponding load data, and utilizes nonparametric
The relation that PDF estimation is set up between environmental data and load data；
Module 12 is built, is used for：Based on the adaptive forgetting factor in Relation acquisition historical data, build and be based on self adaptation
The time-varying Cook distance statistics amounts of forgetting factor；
Repair module 13, is used for：The load data in historical data is detected using time-varying Cook distance statistics amount,
If detecting abnormal data, the corresponding environmental data of abnormal data is based on to different using nonparametric PDF estimation
Regular data is repaired；
Training module 14, is used for：Abnormal data and other historical datas to repairing completion are trained and obtain corresponding
Load forecasting model, and being optimized to load forecasting model with MCSO algorithms, it is short to be realized using the load forecasting model
When load prediction.
A kind of short-term load forecasting device provided in an embodiment of the present invention, processing module can include：
Acquiring unit, is used for：Acquisition includes the historical data of environmental data and corresponding load data, wherein environmental data bag
Include humidity data and temperature data.
A kind of short-term load forecasting device provided in an embodiment of the present invention, building module can include：
Construction unit, is used for：Using recurrent least square method be based on Relation acquisition historical data in self adaptation forget because
Son.
A kind of short-term load forecasting device provided in an embodiment of the present invention, training module can include：
Training unit, is used for：The abnormal data and other historical datas of repairing completion are trained and obtained using ORELM
Corresponding load forecasting model.
The explanation of relevant portion refers to of the invention real in a kind of short-term load forecasting device provided in an embodiment of the present invention
The detailed description of corresponding part in a kind of short-term load forecasting method of example offer is provided, be will not be repeated here.
The foregoing description of the disclosed embodiments, enables those skilled in the art to realize or uses the present invention.To this
Various modifications of a little embodiments will be apparent for a person skilled in the art, and generic principles defined herein can
Without departing from the spirit or scope of the present invention, to realize in other embodiments.Therefore, the present invention will not be limited
It is formed on the embodiments shown herein, but wants most wide consistent with principles disclosed herein and features of novelty of load
Scope.
Claims (8)
1. a kind of short-term load forecasting method, it is characterised in that including：
Acquisition includes the historical data of environmental data and corresponding load data, and is set up using nonparametric PDF estimation
Relation between the environmental data and the load data；
Based on the adaptive forgetting factor in historical data described in the Relation acquisition, build and be based on the adaptive forgetting factor
Time-varying Cook distance statistics amounts；
The load data in the historical data is detected using the time-varying Cook distance statistics amount, if detected different
Regular data, then be based on the corresponding environmental data of the abnormal data to the abnormal number using nonparametric PDF estimation
According to being repaired；
The abnormal data and other described historical datas to repairing completion are trained and obtain corresponding load forecasting model,
And the load forecasting model is optimized with MCSO algorithms, realize that short-term load (duty) is pre- with using the load forecasting model
Survey.
2. method according to claim 1, it is characterised in that acquisition includes the history of environmental data and corresponding load data
Data, including：
Acquisition includes the historical data of environmental data and corresponding load data, wherein the environmental data includes humidity data and temperature
Degrees of data.
3. method according to claim 2, it is characterised in that adaptive in based on historical data described in the Relation acquisition
Forgetting factor is answered, including：
The adaptive forgetting factor in historical data described in the Relation acquisition is based on using recurrent least square method.
4. method according to claim 1, it is characterised in that gone through described in the abnormal data and other that complete to repairing
History data are trained and obtain corresponding load forecasting model, including：
The abnormal data and other described historical datas of repairing completion are trained and obtain corresponding load using ORELM
Forecast model.
5. a kind of short-term load forecasting device, it is characterised in that including：
Processing module, is used for：Acquisition includes the historical data of environmental data and corresponding load data, and close using nonparametric probability
The relation that degree Function Estimation is set up between the environmental data and the load data；
Module is built, is used for：Based on the adaptive forgetting factor in historical data described in the Relation acquisition, build based on described
The time-varying Cook distance statistics amounts of adaptive forgetting factor；
Repair module, is used for：The load data in the historical data is examined using the time-varying Cook distance statistics amount
Survey, if detecting abnormal data, the corresponding environment of the abnormal data is based on using nonparametric PDF estimation
Data are repaired to the abnormal data；
Training module, is used for：The abnormal data and other described historical datas to repairing completion are trained and obtain correspondence
Load forecasting model, and the load forecasting model is optimized with MCSO algorithms, to utilize the load forecasting model
Realize short-term load forecasting.
6. device according to claim 5, it is characterised in that the processing module includes：
Acquiring unit, is used for：Acquisition includes the historical data of environmental data and corresponding load data, wherein the environmental data bag
Include humidity data and temperature data.
7. device according to claim 6, it is characterised in that the structure module includes：
Construction unit, is used for：The self adaptation being based on using recurrent least square method in historical data described in the Relation acquisition is lost
Forget the factor.
8. device according to claim 5, it is characterised in that the training module includes：
Training unit, is used for：The abnormal data and other described historical datas of repairing completion are trained using ORELM
Obtain corresponding load forecasting model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710156989.9A CN106920014A (en) | 2017-03-16 | 2017-03-16 | A kind of short-term load forecasting method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710156989.9A CN106920014A (en) | 2017-03-16 | 2017-03-16 | A kind of short-term load forecasting method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106920014A true CN106920014A (en) | 2017-07-04 |
Family
ID=59461954
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710156989.9A Pending CN106920014A (en) | 2017-03-16 | 2017-03-16 | A kind of short-term load forecasting method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106920014A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108767883A (en) * | 2018-06-27 | 2018-11-06 | 深圳库博能源科技有限公司 | A kind of response processing method of Demand-side |
| 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 |
| CN109408772A (en) * | 2018-10-11 | 2019-03-01 | 四川长虹电器股份有限公司 | To the restoration methods of the abnormal data in continuity data |
| CN111222769A (en) * | 2019-12-30 | 2020-06-02 | 河南拓普计算机网络工程有限公司 | Annual report data quality evaluation method and device, electronic equipment and storage medium |
| CN111583065A (en) * | 2020-05-12 | 2020-08-25 | 广东电网有限责任公司计量中心 | Power load data prediction method and device |
| CN112747413A (en) * | 2019-10-31 | 2021-05-04 | 北京国双科技有限公司 | Air conditioning system load prediction method and device |
| US20210209467A1 (en) * | 2018-09-25 | 2021-07-08 | Ennew Digital Technology Co., Ltd. | Method and device for predicting thermal load of electrical system |
| CN115907168A (en) * | 2022-11-28 | 2023-04-04 | 浙江浙能能源服务有限公司 | Abnormal data processing system for power load prediction |
| CN116467667A (en) * | 2023-06-20 | 2023-07-21 | 图观(天津)数字科技有限公司 | Power failure monitoring and early warning method and system based on data fusion |
| CN117435873A (en) * | 2023-12-21 | 2024-01-23 | 山东中都机器有限公司 | Data management method based on intelligent spraying dust fall |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631483A (en) * | 2016-03-08 | 2016-06-01 | 国家电网公司 | Method and device for predicting short-term power load |
-
2017
- 2017-03-16 CN CN201710156989.9A patent/CN106920014A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631483A (en) * | 2016-03-08 | 2016-06-01 | 国家电网公司 | Method and device for predicting short-term power load |
Non-Patent Citations (2)
| Title |
|---|
| 孟安波等: ""基于纵横交叉算法优化神经网络的负荷预测模型"", 《电力系统保护与控制》 * |
| 彭显刚等: ""考虑负荷自适应检测和修复的鲁棒极限学习机短期负荷预测方法"", 《中国电机工程学报》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108767883A (en) * | 2018-06-27 | 2018-11-06 | 深圳库博能源科技有限公司 | A kind of response processing method of Demand-side |
| 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 |
| US20210209467A1 (en) * | 2018-09-25 | 2021-07-08 | Ennew Digital Technology Co., Ltd. | Method and device for predicting thermal load of electrical system |
| CN109408772A (en) * | 2018-10-11 | 2019-03-01 | 四川长虹电器股份有限公司 | To the restoration methods of the abnormal data in continuity data |
| CN112747413B (en) * | 2019-10-31 | 2022-06-21 | 北京国双科技有限公司 | Air conditioning system load prediction method and device |
| CN112747413A (en) * | 2019-10-31 | 2021-05-04 | 北京国双科技有限公司 | Air conditioning system load prediction method and device |
| CN111222769A (en) * | 2019-12-30 | 2020-06-02 | 河南拓普计算机网络工程有限公司 | Annual report data quality evaluation method and device, electronic equipment and storage medium |
| CN111583065A (en) * | 2020-05-12 | 2020-08-25 | 广东电网有限责任公司计量中心 | Power load data prediction method and device |
| CN111583065B (en) * | 2020-05-12 | 2023-08-22 | 广东电网有限责任公司计量中心 | Power load data prediction method and device |
| CN115907168A (en) * | 2022-11-28 | 2023-04-04 | 浙江浙能能源服务有限公司 | Abnormal data processing system for power load prediction |
| CN116467667A (en) * | 2023-06-20 | 2023-07-21 | 图观(天津)数字科技有限公司 | Power failure monitoring and early warning method and system based on data fusion |
| CN116467667B (en) * | 2023-06-20 | 2023-08-22 | 图观(天津)数字科技有限公司 | Power failure monitoring and early warning method and system based on data fusion |
| CN117435873A (en) * | 2023-12-21 | 2024-01-23 | 山东中都机器有限公司 | Data management method based on intelligent spraying dust fall |
| CN117435873B (en) * | 2023-12-21 | 2024-03-08 | 山东中都机器有限公司 | Data management method based on intelligent spraying dust fall |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106920014A (en) | A kind of short-term load forecasting method and device | |
| Gao et al. | Day-ahead power forecasting in a large-scale photovoltaic plant based on weather classification using LSTM | |
| Han et al. | Mid-to-long term wind and photovoltaic power generation prediction based on copula function and long short term memory network | |
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| US11409347B2 (en) | Method, system and storage medium for predicting power load probability density based on deep learning | |
| CN108280551A (en) | A kind of photovoltaic power generation power prediction method using shot and long term memory network | |
| CN112733462B (en) | Ultra-short-term wind power plant power prediction method combining meteorological factors | |
| CN104636985A (en) | Method for predicting radio disturbance of electric transmission line by using improved BP (back propagation) neural network | |
| CN103489038A (en) | Photovoltaic ultra-short-term power prediction method based on LM-BP neural network | |
| CN105913151A (en) | Photovoltaic power station power generation amount predication method based on adaptive mutation particle swarm and BP network | |
| CN109284870A (en) | Short-term method for forecasting photovoltaic power generation quantity based on shot and long term Memory Neural Networks | |
| Niu et al. | Uncertainty modeling for chaotic time series based on optimal multi-input multi-output architecture: Application to offshore wind speed | |
| CN110738344B (en) | Distributed reactive power optimization method and device for load prediction of power system | |
| CN109146121A (en) | The power predicating method stopped in the case of limited production based on PSO-BP model | |
| CN109948845A (en) | A kind of distribution network load shot and long term Memory Neural Networks prediction technique | |
| CN103105246A (en) | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm | |
| CN109615124B (en) | SCADA master station load prediction method based on deep learning | |
| CN112881601B (en) | Moisture detecting system based on cloud platform | |
| CN108038568A (en) | A kind of changeable weight combination Short-Term Load Forecasting of Electric Power System based on particle cluster algorithm | |
| CN113219871B (en) | Curing room environmental parameter detecting system | |
| CN104598765A (en) | Building energy consumption prediction method based on elastic adaptive neural network | |
| Zhou et al. | Deep learning-based rolling horizon unit commitment under hybrid uncertainties | |
| CN109034500A (en) | A kind of mid-term electric load forecasting method of multiple timings collaboration | |
| CN114861980A (en) | BP-LSTM model-based carbon prediction method | |
| CN109447843B (en) | Photovoltaic output prediction method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170704 |
|
| RJ01 | Rejection of invention patent application after publication |