CN103996076B - The change method for early warning and system of electricity needs - Google Patents
The change method for early warning and system of electricity needs Download PDFInfo
- Publication number
- CN103996076B CN103996076B CN201410206780.5A CN201410206780A CN103996076B CN 103996076 B CN103996076 B CN 103996076B CN 201410206780 A CN201410206780 A CN 201410206780A CN 103996076 B CN103996076 B CN 103996076B
- Authority
- CN
- China
- Prior art keywords
- data
- variable
- mrow
- reference index
- default
- 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.)
- Active
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of change method for early warning of electricity needs, including：Stationary test is carried out to the achievement data in default alternative index storehouse and default reference index data；According to stationary test result, if same order list is not present in achievement data and reference index data whole, Rejection index data；If achievement data and reference index data same order list are whole, whether judge index data exist with reference index data is assisted whole property, if it is not, Rejection index data, if so, the then fluctuation relation between judge index data and reference index data；Achievement data, reference index data and fluctuation relation are inputted to neural network prediction model, obtain electric power demand forecasting value；The history electricity needs actual value in predetermined period is obtained, further according to electric power demand forecasting value, consumer confidence index is calculated, regenerates early warning information.Present invention system also corresponding to offer, can rapidly obtain the consumer confidence index of reflection electricity needs change, and the change to electricity needs sends early warning information.
Description
Technical field
The present invention relates to electricity needs technical field, more particularly to a kind of change method for early warning of electricity needs, and
A kind of change early warning system of electricity needs.
Background technology
The production and supply of electric power are that China's economic development, social progress and living standards of the people improve essential bar
Part, in recent years, with the continuous development of national economy, electric power also constantly increases as important energy safeguard.Power industry meeting
According to the good power generation total amount of the prior overall arrangement of the power demand of prediction, for the scope in the whole nation, the supply and demand of electric power now
Sufficiently effective guarantee can not still be obtained.The energy imbalance of region power supply and demand development is in China's long-term existence, whenever the summer in winter
Can also often occur the phenomenon of power cuts to limit consumption during two season peaks of power consumption, the work and life to resident bring great inconvenience.Cause
This, which is needed badly, can the technology for making early warning in advance to Regional Electric Market demand, be electric power enterprise department electric power totalizing method,
Electricity needs planning etc. provides warning information.
The content of the invention
Based on this, the present invention provides a kind of the change method for early warning and system of electricity needs, can rapidly obtain reflection electricity
The consumer confidence index of power changes in demand, the change to electricity needs send early warning information.
A kind of change method for early warning of electricity needs, comprises the following steps：
Stationary test is carried out to each achievement data in default alternative index storehouse and default reference index data；
According to the stationary test result, if same order list is not present with the reference index data in the achievement data
It is whole, then reject the achievement data from the alternative index storehouse；
If the achievement data and the reference index data same order list are whole, judge the achievement data whether with it is described
Reference index data, which exist, assists whole property, if it is not, the achievement data is rejected from the alternative index storehouse, if so, then judging institute
State the fluctuation relation between achievement data and the reference index data；
By between the achievement data, the reference index data and the achievement data and the reference index data
Fluctuation relation is inputted to default neural network prediction model, obtains electric power demand forecasting value；
The history electricity needs actual value in predetermined period is obtained, further according to the electric power demand forecasting value, utilizes following formula
Calculate consumer confidence index：
Early warning information is generated according to the consumer confidence index.
A kind of change early warning system of electricity needs, including：
Stationary test module, for each achievement data in default alternative index storehouse and default reference index
Data carry out stationary test；
Module is rejected, for according to the stationary test result, if the achievement data and the reference index data
It is whole in the absence of same order list, then reject the achievement data from the alternative index storehouse；
Mould preparation block is assisted, if whole for the achievement data and the reference index data same order list, judges the index
Whether data exist with the reference index data is assisted whole property, if it is not, the achievement data is rejected from the alternative index storehouse,
If so, then judge the fluctuation relation between the achievement data and the reference index data；
Neural network module, for by the achievement data, the reference index data and the achievement data with it is described
Fluctuation relation between reference index data is inputted to default neural network prediction model, obtains electric power demand forecasting value；
Consumer confidence index module, for obtaining the history electricity needs actual value in predetermined period, needed further according to the electric power
Predicted value is sought, consumer confidence index is calculated using following formula：
Generation module, for generating early warning information according to the consumer confidence index.
The change method for early warning and system of above-mentioned electricity needs, first to each achievement data and base in alternative index storehouse
Quasi- achievement data carries out stationary test, and the data with reference index data not same order are deleted according to stationary test result；Again
Whether judge index data exist with reference index data is assisted whole property, is effectively avoided shadowing property, is then judged the index number
According to the fluctuation relation between the reference index data, input to neural network prediction model, obtain electric power demand forecasting value；
Consumer confidence index is finally calculated, early warning information is generated according to consumer confidence index, early warning is made in the prosperous change to electricity needs.
Brief description of the drawings
Fig. 1 is the schematic flow sheet of the change method for early warning of electricity needs of the present invention in one embodiment.
Fig. 2 is the structural representation of multilayer perceptron.
Fig. 3 is the structural representation of more time quantum neutral nets.
Fig. 4 is the structural representation of concurrent integration neutral net.
Fig. 5 is the structural representation of the change early warning system of electricity needs of the present invention in one embodiment.
Embodiment
The present invention is described in further detail with reference to embodiment and accompanying drawing, but embodiments of the present invention are not limited to
This.
As shown in figure 1, be a kind of schematic flow sheet of change method for early warning of electricity needs of the present invention in one embodiment,
Comprise the following steps：
S11, stationarity is carried out to each achievement data in default alternative index storehouse and default reference index data
Examine；
Reference index data can be to need the electricity sales amount of prewarning area, and the electricity sales amount in region is affected by many factors, such as with
Family power off time, transformer station's quantity, supply station quantity, absolute Business Process System amount, main transformer fault rate, raw coal output, area production
The many factors such as total value, money supply and investment in fixed assets；Default alternative index storehouse includes a variety of influence electricity sales amounts
Achievement data, achievement data is more, then the electric power demand forecasting value finally obtained is then more accurate；
For example it may be selected from the different field such as employment, finance, finance, consumption, price, stock, trade and may be to electricity
Power demand produces all types of indexs influenceed to a certain degree, collects relevant historical data and establishes alternative index storehouse, not only covers
Demand, supply, fund, tax revenue and the several principal elements of price, also include such as weather factor.Index number in alternative index storehouse
According to being contemplated that following condition：
(1) it is comprehensive on field.Selected index should cover as far as possible can reflect the prosperous fluctuation of electric power market demand
Various aspects, so as to ensure follow-up model prediction it can be made correctly judge and analysis.
(2) statistical adequacy.Selected index statistically should be complete as far as possible and with necessarily reliable
Property, selected achievement data should be able to be easy to get in time, and correct without prolonged interruption and frequently.Pay attention in selective goal sequence
The long span of achievement data should be ensured, while to widen coverage rate, monthly data is chosen and be preferred.
(3) correspondence fluctuated with boom.I.e. selected index should have certain representativeness, can in time, accurately
Reflection electric power market demand fluctuation status, while also tackle its following developing state and make prediction.
In default alternative index storehouse, the association countless ties between different indexs, or mutually restrict, or reciprocal causation.It is right
Before extensive sample is analyzed and predicted, it is necessary to alternative index storehouse is screened, ensures the steady of index time series
Long-run equilibrium relation of the property (statistical law of sequence data does not change with time and changed) between index, avoids non-stationary sequence
Appearance caused by row are possible a series of problems, such as " shadowing property " and " pseudo- conclusion ", reduce the complexity of forecast model.
The present embodiment carries out stationary test first, if a time series yt, to pass through d (d>0) secondary difference could become
It is a non-stationary series into stable time series, and during this differential of sequence d-1 times, then then claim time series yt
It is d ranks singly whole time series, is designated as yt~I (d).
Specifically, each achievement data in default alternative index storehouse and default reference index data are carried out
The step of stationary test, may include：
The achievement data and the reference index data are sequentially arranged, obtain time series；
Unit root test is carried out to the time series by following regression equations：
In formula, β is default constant term, and ρ is default coefficient correlation, and η t are default linear trend function, and l is default
Lag order, εtFor default residual error item,For default hysteresis item, ytFor the time series, Δ ytFor yt's
Difference；To ensure residual error item εtIt is white noise, therefore hysteresis item is added in regression equationΔyt-iLag item
Number l selection criterion is to eliminate εtUntill interior auto-correlation；
The condition of stationary test is as follows：
Wherein, t=(ρ ' -1)/se (ρ ')；ρ ' is ρ least-squares estimation value, and se (ρ ') is ρ ' unbiased estimator；It is former
Assuming that H0A unit root at least be present for the time series；Alternative hvpothesis H1Unit root is not present for time series；
If | t | more than default critical value, receive null hypothesis H0, there is unit root in the time series, when described
Between sequence be jiggly, then again to Δ ytStationary test is carried out, until it is determined that ytFor the single whole sequence of several ranks；
If | t | less than the critical value, refusal null hypothesis H0, receive alternative hvpothesis H1, unit is not present in time series
Root, i.e., described time series is stable, determines that the time series is singly whole for N ranks；N is current time for carrying out unit root test
Number.
S12, according to the stationary test result, if same order is not present with the reference index data in the achievement data
It is single whole, then reject the achievement data from the alternative index storehouse；
, will be with reference index according to reference index data stationarity assay and the stationary test result of achievement data
Data are not that the whole achievement data of same order list is all deleted from alternative index storehouse.
If S13, the achievement data and the reference index data same order list are whole, judge the achievement data whether with
The reference index data, which exist, assists whole property, if it is not, the achievement data is rejected from the alternative index storehouse, if so, then sentencing
Disconnected fluctuation relation between the achievement data and the reference index data；
If two or more sequences are non-stationary series, and their linear combination is stable, then just
Think that there is long-term balanced relation between these sequences；For the time series of non-stationary, only when with whole relation is assisted,
Shadowing property can effectively be avoided.
Specifically, described judge that the step of whether achievement data assists whole property with reference index data presence is：
All time serieses are formed into the rank column vector of k × 1：Yt=(y1t, y2t..., ykt) ', YtGather for vector, y1t,
y2t,…,yktFor 1~k of sequence number time series；
Judge whether the vector set meets following two conditions：
yit~I (d), i=1,2 ..., k, YtIn each time series the whole exponent number of list all equal to the reference index
The exponent number d, y of datait~I (d) represents time series yitIt is d ranks singly whole time series；
In the presence of the rank column vector β of a k × 1=(β1, β2..., βk) ' so that
(β≠0)；β is to assist whole vector, β element β1,β2,…,βkTo assist whole parameter, b is the exponent number of the default whole parameter of association, and b is less than
d；
If satisfied, then there is the whole property of association in the achievement data with the reference index data.
Next, Granger causality test is carried out with reference index respectively to each index filtered out, further
Mark off the leading indicators and lagging indicator of electricity needs fluctuation.
Granger causality between test variable, it that is to say by the hypothesis testing in regression analysis, examine past
Variable x can make the explanation of much degree to present y, and its explanation degree is improved after x lagged value is introduced, can also
Say, if predictions of the variable x to variable y is helpful, when the coefficient correlation statistical check of two variables is notable, it is possible to be considered from
Granger causality be present to variable y in variable x.
The fluctuation relation judged between the achievement data and the reference index data includes：
Test variable x whether be variable y granger cause, wherein, variable x and variable y are the achievement data and institute
State reference index data：
To variable y all hysteresis item yt-1,yt-2,…,yt-mReturned, obtain constrained regression equation, calculate it is described by
Residual sum of squares (RSS) RSS in constrained regression formulaR；
The hysteresis item that the variable x is added in the constrained regression equation obtains no constrained regression formula, calculates the nothing
The residual sum of squares (RSS) RSS of constrained regression formulaUR；
Examine the F values in following formula：
Wherein, m is the number that variable y lags item, and n is sample size, and q is the number that variable x lags item, and j is the nothing
Number of parameters to be estimated in constrained regression formula, it obeys the F distributions that the free degree is q and (n-j)；
If the F values calculated are more than default critical value, refuse null hypothesis, variable x hysteresis item belongs to the no constraint
Regression equation, variable x are variable y granger causes；Wherein, the hysteresis item that the null hypothesis is variable x is not belonging to described by about
Beam regression equation；
Test variable y whether be variable x granger cause；
If variable x is variable y granger cause, the achievement data is the leading finger of the reference index data
Mark, if variable y is variable x granger cause, the achievement data is the lagging indicator of the reference index data.
Next it may further determine that index issue：
Pass through impulse response function (IRF), it may be determined that the dynamic change of model when system is impacted, utilize time sequence
Row model is analyzed influence relation, is investigated disturbance term is how will to influence to be delivered to each variable, that is, is analyzed to error term
Plus the influence after standard unit's error to endogenous variable currency and its future value.To lag the vector auto regression of 2 ranks
Model (VAR, using the form of multiple equations simultaneousnesses, using each endogenous variable in system as all endogenous variables in system
The function of lagged value build model) exemplified by：
In above formula, ai、bi、ciAnd diAll it is parameter, εt=(ε1t,ε2t)TFor Disturbance, it is assumed that εtIt is that there is following property
The white noise variable of matter：
Hypothesized model is and the y since 0 phase-1=y-2=x-1=x-2=0, it is assumed that given random since the 0th phase
Disturbance term ε10=1, ε20=0, ε1t=ε2t=0, (t=1,2 ...), i.e., a pulse is given in the 0th phase, variable y is then discussedtWith
xtResponse：During t=0, y0=1, x0=0；During t=1, y1=a1, x1=c1；During t=2,x2=c1a1+
c2；By that analogy, variable y and variable x receptance function as caused by variable y pulse be can obtain.IRF can significantly be caught
To the time series variable response to stochastic error disturbance over time, time series change can be dynamically depicted
Dynamic path change between amount.
S14, by the achievement data, the reference index data and the achievement data and the reference index data it
Between fluctuation relation input to default neural network prediction model, obtain electric power demand forecasting value；
According to default neural network prediction model, by the achievement data, the reference index data and the index
Fluctuation relation input between data and the reference index data, obtains the output result of neural network prediction model, obtains
Electric power demand forecasting value.
Neural network structure may include：
Multilayer perceptron：Such as the network diagram that Fig. 2 is multilayer perceptron (MLP), the structure is a complete forward edge
The network connect, generally it is made up of input layer, hidden layer and output layer.Input layer receives the information from outside input, the number of neuron
Mesh depends on the quantity of input variable.Hidden layer can be counted as a feature extractor, integrate input layer information, and generate new
Function e-learning, the number of neuron depends on the circumstances, and the number of the hidden layer of neuron is chosen for input layer and defeated
Go out the neuron number purpose arithmetic mean of instantaneous value of layer.Output layer generates prediction result and the error of propagation parameter estimation.Neuron
Output layer number depend on need that how many future time predicted.In the present embodiment, the number of output neuron can be 1,
The characteristics of MLP is hidden layer by the feature for integrating the information of input layer to generate new, i.e., simultaneously using at hand all available
Information produces the characteristics of new (being played a role by hidden neuron).
More time quantum neutral nets：, will in the structure such as the network diagram that Fig. 3 is more time quantum neutral nets
It resolves into several sub-blocks, and each piece of task is each to be trained.In the present embodiment, input variable have two even it is more
Individual different source.More time quantum neutral nets (MTUNN) handle the input variable of a variety of separate sources, every kind of source respectively
Input variable be an ensemble, each self-training of each ensemble is then extracted into feature.After this, training result is whole
It is combined, to obtain final prediction.Fig. 3 show MTUNN structures.MTUNN concrete function is as follows：Input layer receives outer
Portion inputs information, and the second layer carries out classification processing respectively by the division of different field to data, and this layer is named as functional layer.
One important feature of functional layer is weight portion connection.Third layer integrates the information from different sub-blocks, is named as integrating
Layer.In addition, this layer of weight is fully connected to exchange information.It is defeated that output layer generation network is fully connected in conformable layer neuron
Go out.Training sample and test sample MTUNN are completely the same.
Concurrent integration neutral net：Such as the network diagram that Fig. 4 is concurrent integration neutral net；Concurrent integration neutral net
(MTUNN) division input information is divided into the respective study of several sub-blocks progress, for example the information of acquisition meteorological data or electric power need
Internal information is sought, it is even more more, then the learning outcome of each sub-block is integrated in a network.Then several networks are exported and carried out
It is integrated finally to be predicted with producing.Concurrent integration neutral net (PENN) is designed by above-mentioned idea, as shown in Figure 4.PENN bags
Two are contained even more than independent submodel：Power model or climate model etc..Each submodel is one independent
MLP structures.This structure, which may not be needed substantial amounts of historical sample, can extract data characteristics, and forecast model is by time sequence
The influence of row historical trend is reduced, and can be suitable for up-to-date information, prediction accuracy lifting.Can between the data of different models
Caused can associate influences to be eliminated by the mode that data are decomposed.
History electricity needs actual value in S15, acquisition predetermined period, further according to the electric power demand forecasting value, is utilized
Following formula calculates consumer confidence index：
S16, according to the consumer confidence index generate early warning information；
Consumer confidence index is cut off value with 50, higher than 50, it is very fast to show that electricity needs increases, in prosperous state；Less than 50,
Show electricity needs growth slowdown, in depression gaseity；Exponential number deviation 50.0 is more remote, represents that amplitude of variation is bigger；Cause
This, when the consumer confidence index is less than 50, can generate warning information according to consumer confidence index value；Also the boom that each region can be calculated refers to
Number, to default total electricity sales amount, is allocated in the different consumer confidence index ratio of different zones.
The present invention also provides a kind of change early warning system of electricity needs, as shown in figure 5, including：
Stationary test module 21, for referring to each achievement data in default alternative index storehouse and default benchmark
Mark data and carry out stationary test；
Reference index data can be to need the electricity sales amount of prewarning area, and the electricity sales amount in region is affected by many factors, such as with
Family power off time, transformer station's quantity, supply station quantity, absolute Business Process System amount, main transformer fault rate, raw coal output, area production
The many factors such as total value, money supply and investment in fixed assets；Default alternative index storehouse includes a variety of influence electricity sales amounts
Achievement data, achievement data is more, then the electric power demand forecasting value finally obtained is then more accurate；
For example it may be selected from the different field such as employment, finance, finance, consumption, price, stock, trade and may be to electricity
Power demand produces all types of indexs influenceed to a certain degree, collects relevant historical data and establishes alternative index storehouse, not only covers
Demand, supply, fund, tax revenue and the several principal elements of price, also include such as weather factor.Index number in alternative index storehouse
According to being contemplated that following condition：
(1) it is comprehensive on field.Selected index should cover as far as possible can reflect the prosperous fluctuation of electric power market demand
Various aspects, so as to ensure follow-up model prediction it can be made correctly judge and analysis.
(2) statistical adequacy.Selected index statistically should be complete as far as possible and with necessarily reliable
Property, selected achievement data should be able to be easy to get in time, and correct without prolonged interruption and frequently.Pay attention in selective goal sequence
The long span of achievement data should be ensured, while to widen coverage rate, monthly data is chosen and be preferred.
(3) correspondence fluctuated with boom.I.e. selected index should have certain representativeness, can in time, accurately
Reflection electric power market demand fluctuation status, while also tackle its following developing state and make prediction.
In default alternative index storehouse, the association countless ties between different indexs, or mutually restrict, or reciprocal causation.It is right
Before extensive sample is analyzed and predicted, it is necessary to alternative index storehouse is screened, ensures the steady of index time series
Long-run equilibrium relation of the property (statistical law of sequence data does not change with time and changed) between index, avoids non-stationary sequence
Appearance caused by row are possible a series of problems, such as " shadowing property " and " pseudo- conclusion ", reduce the complexity of forecast model.
The present embodiment carries out stationary test first, if a time series yt, to pass through d (d>0) secondary difference could become
It is a non-stationary series into stable time series, and during this differential of sequence d-1 times, then then claim time series yt
It is d ranks singly whole time series, is designated as yt~I (d).
Specifically, the stationary test module 21 is additionally operable to；
The achievement data and the reference index data are sequentially arranged, obtain time series；
Unit root test is carried out to the time series by following regression equations：
In formula, β is default constant term, and ρ is default coefficient correlation, and η t are default linear trend function, and l is default
Lag order, εtFor default residual error item,For default hysteresis item, ytFor the time series, Δ ytFor yt's
Difference；To ensure residual error item εtIt is white noise, therefore hysteresis item is added in regression equationΔyt-iLag item
Number l selection criterion is to eliminate εtUntill interior auto-correlation；
The condition of stationary test is as follows：
Wherein, t=(ρ ' -1)/se (ρ ')；ρ ' is ρ least-squares estimation value, and se (ρ ') is ρ ' unbiased estimator；It is former
Assuming that H0A unit root at least be present for the time series；Alternative hvpothesis H1Unit root is not present for time series；
If | t | more than default critical value, receive null hypothesis H0, there is unit root in the time series, when described
Between sequence be jiggly, then again to Δ ytCarry out unit root test；
If | t | less than the critical value, refusal null hypothesis H0, receive alternative hvpothesis H1, unit is not present in time series
Root, i.e., described time series is stable, determines that the time series is singly whole for N ranks；N is current time for carrying out unit root test
Number.
Module 22 is rejected, for according to the stationary test result, if the achievement data and the reference index number
According to whole in the absence of same order list, then the achievement data is rejected from the alternative index storehouse；
, will be with reference index according to reference index data stationarity assay and the stationary test result of achievement data
Data are not that the whole achievement data of same order list is all deleted from alternative index storehouse.
Mould preparation block 23 is assisted, if whole for the achievement data and the reference index data same order list, judges the finger
Whether mark data exist with the reference index data is assisted whole property, if it is not, rejecting the index number from the alternative index storehouse
According to if so, then judging the fluctuation relation between the achievement data and the reference index data；
If two or more sequences are non-stationary series, and their linear combination is stable, then just
Think that there is long-term balanced relation between these sequences；For the time series of non-stationary, only when with whole relation is assisted,
Shadowing property can effectively be avoided.
Specifically, association's mould preparation block 23 is additionally operable to：
All time serieses are formed into the rank column vector of k × 1：Yt=(y1t, y2t..., ykt) ', YtGather for vector, y1t,
y2t,…,yktFor 1~k of sequence number time series；
Judge whether the vector set meets following two conditions：
yit~I (d), i=1,2 ..., k, YtIn the whole exponent number of list of each time series be equal to the reference index data
Exponent number d, yit~I (d) represents time series ytIt is d ranks singly whole time series；
In the presence of the rank column vector β of a k × 1=(β1, β2..., βk) ' so that
(β≠0)；β is to assist whole vector, β element β1,β2,…,βkTo assist whole parameter, b is the exponent number of the default whole parameter of association, and b is less than
d；
If satisfied, then there is the whole property of association in the achievement data with the reference index data.
Further drawn next, carrying out Granger causality test with reference index respectively to each index filtered out
Separate the leading indicators and lagging indicator of electricity needs fluctuation.
Granger causality between test variable, it that is to say by the hypothesis testing in regression analysis, examine past
Variable x can make the explanation of much degree to present y, and its explanation degree is improved after x lagged value is introduced, can also
Say, if predictions of the variable x to variable y is helpful, when the coefficient correlation statistical check of two variables is notable, it is possible to be considered from
Granger causality be present to variable y in variable x.
Association's mould preparation block 23 is additionally operable to：
Test variable x whether be variable y granger cause, wherein, variable x and variable y are the achievement data and institute
State reference index data：
To variable y all hysteresis item yt-1,yt-2,…,yt-mReturned, obtain constrained regression equation, calculate it is described by
Residual sum of squares (RSS) RSS in constrained regression formulaR；
The hysteresis item that the variable x is added in the constrained regression equation obtains no constrained regression formula, calculates the nothing
The residual sum of squares (RSS) RSS of constrained regression formulaUR；
Examine the F values in following formula：
Wherein, m is the number that variable y lags item, and n is sample size, and q is the number that variable x lags item, and j is the nothing
Number of parameters to be estimated in constrained regression formula, it obeys the F distributions that the free degree is q and (n-j)；
If the F values calculated are more than default critical value, refuse null hypothesis, variable x hysteresis item belongs to the no constraint
Regression equation, variable x are variable y granger causes；
Test variable y whether be variable x granger cause；
If variable x is variable y granger cause, the achievement data is the leading finger of the reference index data
Mark, if variable y is variable x granger cause, the achievement data is the lagging indicator of the reference index data.
Neural network module 24, for by the achievement data, the reference index data and the achievement data and institute
State the fluctuation relation between reference index data to input to default neural network prediction model, obtain electric power demand forecasting value；
According to default neural network prediction model, by the achievement data, the reference index data and the index
Fluctuation relation input between data and the reference index data, obtains the output result of neural network prediction model, obtains
Electric power demand forecasting value.
Consumer confidence index module 25, for obtaining the history electricity needs actual value in predetermined period, further according to the electric power
Requirement forecasting value, consumer confidence index is calculated using following formula：
Generation module 26, for generating early warning information according to the consumer confidence index.
Consumer confidence index is cut off value with 50, higher than 50, it is very fast to show that electricity needs increases, in prosperous state；Less than 50,
Show electricity needs growth slowdown, in depression gaseity；Exponential number deviation 50.0 is more remote, represents that amplitude of variation is bigger；Cause
This, when the consumer confidence index is less than 50, can generate warning information according to consumer confidence index value；Also the boom that each region can be calculated refers to
Number, to default total electricity sales amount, is allocated in the different consumer confidence index ratio of different zones.
The change method for early warning and system of electricity needs of the present invention, first to each achievement data in alternative index storehouse and
Reference index data carry out stationary test, and the data with reference index data not same order are deleted according to stationary test result；
Whether judge index data exist with reference index data again is assisted whole property, is effectively avoided shadowing property, is then judged the index
Fluctuation relation between data and the reference index data, input to neural network prediction model, obtain electric power demand forecasting
Value；Consumer confidence index is calculated, early warning information is generated according to consumer confidence index, early warning is made in the prosperous change to electricity needs.
Embodiment described above only expresses the several embodiments of the present invention, and its description is more specific and detailed, but simultaneously
Therefore the limitation to the scope of the claims of the present invention can not be interpreted as.It should be pointed out that for one of ordinary skill in the art
For, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to the guarantor of the present invention
Protect scope.Therefore, the protection domain of patent of the present invention should be determined by the appended claims.
Claims (6)
1. the change method for early warning of a kind of electricity needs, it is characterised in that comprise the following steps：
Stationary test is carried out to each achievement data in default alternative index storehouse and default reference index data；
According to the stationary test result, if the achievement data is whole in the absence of same order list with the reference index data,
The achievement data is rejected from the alternative index storehouse；
If the achievement data and the reference index data same order list are whole, judge the achievement data whether with the benchmark
Achievement data, which exists, assists whole property, if it is not, the achievement data is rejected from the alternative index storehouse, if so, then judging the finger
Mark the fluctuation relation between data and the reference index data；
By the fluctuation between the achievement data, the reference index data and the achievement data and the reference index data
Relation is inputted to default neural network prediction model, obtains electric power demand forecasting value；The neural network prediction model includes
One kind in multilayer perceptron model, more time quantum neural network models, concurrent integration neural network model；
The history electricity needs actual value in predetermined period is obtained, further according to the electric power demand forecasting value, is calculated using following formula
Consumer confidence index：
Early warning information is generated according to the consumer confidence index；
Fluctuation relation between the achievement data and the reference index data is that the achievement data is the reference index
The leading indicators of data, or the achievement data are the lagging indicators of the reference index data；
Also include：
When the consumer confidence index is less than 50, the warning information is generated；
According to the consumer confidence index in each region of calculating, to default total electricity sales amount, in the different consumer confidence index ratio of different zones
It is allocated；
The fluctuation relation judged between the achievement data and the reference index data includes：
Test variable x whether be variable y granger cause, wherein, variable x and variable y are the achievement data and the base
Quasi- achievement data：
To variable y all hysteresis item yt-1,yt-2,…,yt-mReturned, obtain constrained regression equation, calculated described constrained
Residual sum of squares (RSS) RSS in regression equationR；
The hysteresis item that the variable x is added in the constrained regression equation obtains no constrained regression formula, calculates the no constraint
The residual sum of squares (RSS) RSS of regression equationUR；
Examine the F values in following formula：
<mrow>
<mi>F</mi>
<mo>=</mo>
<mfrac>
<mrow>
<mo>(</mo>
<msub>
<mi>RSS</mi>
<mi>R</mi>
</msub>
<mo>-</mo>
<msub>
<mi>RSS</mi>
<mrow>
<mi>U</mi>
<mi>R</mi>
</mrow>
</msub>
<mo>)</mo>
<mo>/</mo>
<mi>q</mi>
</mrow>
<mrow>
<msub>
<mi>RSS</mi>
<mrow>
<mi>U</mi>
<mi>R</mi>
</mrow>
</msub>
<mo>/</mo>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>-</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>;</mo>
</mrow>
Wherein, m is the number that variable y lags item, and n is sample size, and q is the number that variable x lags item, and j is the no constraint
Number of parameters to be estimated in regression equation, obey the F distributions that the free degree is q and (n-j)；
If the F values calculated are more than default critical value, refuse null hypothesis, variable x hysteresis item belongs to the no constrained regression
Formula, variable x are variable y granger causes；Wherein, the null hypothesis is that variable x hysteresis item is not belonging to described constrained time
Return formula；
Test variable y whether be variable x granger cause；
If variable x is variable y granger cause, the achievement data is the leading indicators of the reference index data, if
Variable y is variable x granger cause, then the achievement data is the lagging indicator of the reference index data.
2. the change method for early warning of electricity needs according to claim 1, it is characterised in that described alternatively to refer to default
The step of each achievement data and default reference index data in mark storehouse carry out stationary test includes：
The achievement data and the reference index data are sequentially arranged, obtain time series；
Unit root test is carried out to the time series by following regression equations：
<mrow>
<msub>
<mi>&Delta;y</mi>
<mi>t</mi>
</msub>
<mo>=</mo>
<mi>&beta;</mi>
<mo>+</mo>
<mi>&eta;</mi>
<mi>t</mi>
<mo>+</mo>
<msub>
<mi>&rho;y</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>+</mo>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>l</mi>
</munderover>
<msub>
<mi>&alpha;</mi>
<mi>i</mi>
</msub>
<msub>
<mi>&Delta;y</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mi>&epsiv;</mi>
<mi>t</mi>
</msub>
<mo>;</mo>
</mrow>
In formula, β is default constant term, and ρ is default coefficient correlation, and η t are default linear trend function, and l is default stagnant
Exponent number afterwards, εtFor default residual error item,For default hysteresis item, ytFor the time series, Δ ytFor ytDifference
Point；
The condition of stationary test is as follows：
Wherein, t=(ρ ' -1)/se (ρ '), ρ ' is ρ least-squares estimation value, and se (ρ ') is ρ ' unbiased estimator；Null hypothesis
H0A unit root at least be present for time series；Alternative hvpothesis H1Unit root is not present for time series；
If | t | more than default critical value, receive null hypothesis H0, unit root, the time series be present in the time series
It is jiggly, then again to Δ ytCarry out unit root test；
If | t | less than the critical value, refusal null hypothesis H0, receive alternative hvpothesis H1, unit root is not present in time series,
I.e. described time series is stable, determines that the time series is singly whole for N ranks；N is the current number for carrying out unit root test.
3. the change method for early warning of electricity needs according to claim 2, it is characterised in that described to judge the index number
According to whether with the reference index data exist assist whole property the step of be：
All time serieses are formed into the rank column vector of k × 1：Yt=(y1t, y2t..., ykt) ', YtGather for vector, y1t,y2t,…,
yktFor 1~k of sequence number time series；
Judge whether the vector set meets following two conditions：
yit~I (d), i=1,2 ..., k, YtIn each time series the whole exponent number of list be equal to the reference index data rank
Number d, yit~I (d) represents time series ytIt is d ranks singly whole time series；
In the presence of the rank column vector β of a k × 1=(β1,β2,…,βk) so that(β≠0)；β
To assist whole vector, β element β1,β2,…,βkTo assist whole parameter, b is the exponent number of the default whole parameter of association, and b is less than d；
If satisfied, then there is the whole property of association in the achievement data with the reference index data.
A kind of 4. change early warning system of electricity needs, it is characterised in that including：
Stationary test module, for each achievement data in default alternative index storehouse and default reference index data
Carry out stationary test；
Module is rejected, for according to the stationary test result, if the achievement data is not deposited with the reference index data
It is whole in same order list, then reject the achievement data from the alternative index storehouse；
Mould preparation block is assisted, if whole for the achievement data and the reference index data same order list, judges the achievement data
Whether exist with the reference index data and assist whole property, if it is not, the achievement data is rejected from the alternative index storehouse, if
It is then to judge the fluctuation relation between the achievement data and the reference index data；
Neural network module, for by the achievement data, the reference index data and the achievement data and the benchmark
Fluctuation relation between achievement data is inputted to default neural network prediction model, obtains electric power demand forecasting value；The god
Include multilayer perceptron model, more time quantum neural network models, concurrent integration neural network model through Network Prediction Model
In one kind；
Consumer confidence index module, it is pre- further according to the electricity needs for obtaining the history electricity needs actual value in predetermined period
Measured value, consumer confidence index is calculated using following formula：
Generation module, for generating early warning information according to the consumer confidence index；
Fluctuation relation between the achievement data and the reference index data is that the achievement data is the reference index
The leading indicators of data, or the achievement data are the lagging indicators of the reference index data；
The generation module is additionally operable to, when the consumer confidence index is less than 50, generate the warning information；According to each area of calculating
The consumer confidence index in domain, to default total electricity sales amount, it is allocated in the different consumer confidence index ratio of different zones；
Association's mould preparation block is additionally operable to：
Test variable x whether be variable y granger cause, wherein, variable x and variable y are the achievement data and the base
Quasi- achievement data：
To variable y all hysteresis item yt-1,yt-2,…,yt-mReturned, obtain constrained regression equation, calculated described constrained
Residual sum of squares (RSS) RSS in regression equationR；
The hysteresis item that the variable x is added in the constrained regression equation obtains no constrained regression formula, calculates the no constraint
The residual sum of squares (RSS) RSS of regression equationUR；
F values in inspection：
<mrow>
<mi>F</mi>
<mo>=</mo>
<mfrac>
<mrow>
<mo>(</mo>
<msub>
<mi>RSS</mi>
<mi>R</mi>
</msub>
<mo>-</mo>
<msub>
<mi>RSS</mi>
<mrow>
<mi>U</mi>
<mi>R</mi>
</mrow>
</msub>
<mo>)</mo>
<mo>/</mo>
<mi>q</mi>
</mrow>
<mrow>
<msub>
<mi>RSS</mi>
<mrow>
<mi>U</mi>
<mi>R</mi>
</mrow>
</msub>
<mo>/</mo>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>-</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>;</mo>
</mrow>
Wherein, m is the number that variable y lags item, and n is sample size, and q is the number that variable x lags item, and j is the no constraint
Number of parameters to be estimated in regression equation, obey the F distributions that the free degree is q and (n-j)；
If the F values calculated are more than default critical value, refuse null hypothesis, variable x hysteresis item belongs to the no constrained regression
Formula, variable x are variable y granger causes；
Test variable y whether be variable x granger cause；
If variable x is variable y granger cause, the achievement data is the leading indicators of the reference index data, if
Variable y is variable x granger cause, then the achievement data is the lagging indicator of the reference index data.
5. the change early warning system of electricity needs according to claim 4, it is characterised in that the stationary test module
It is additionally operable to；
The achievement data and the reference index data are sequentially arranged, obtain time series；
Unit root test is carried out to the time series by following regression equations：
<mrow>
<msub>
<mi>&Delta;y</mi>
<mi>t</mi>
</msub>
<mo>=</mo>
<mi>&beta;</mi>
<mo>+</mo>
<mi>&eta;</mi>
<mi>t</mi>
<mo>+</mo>
<msub>
<mi>&rho;y</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>+</mo>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>l</mi>
</munderover>
<msub>
<mi>&alpha;</mi>
<mi>i</mi>
</msub>
<msub>
<mi>&Delta;y</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mi>i</mi>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mi>&epsiv;</mi>
<mi>t</mi>
</msub>
</mrow>
3
In formula, β is default constant term, and ρ is default coefficient correlation, and η t are default linear trend function, and l is default stagnant
Exponent number afterwards, εtFor default residual error item,For default hysteresis item, ytFor the time series, Δ ytFor ytDifference
Point；
The condition of stationary test is as follows：
Wherein, t=(ρ ' -1)/se (ρ ')；ρ ' is ρ least-squares estimation value, and se (ρ ') is ρ ' unbiased estimator；Null hypothesis
H0A unit root at least be present for the time series；Alternative hvpothesis H1Unit root is not present for time series；
If | t | more than default critical value, receive null hypothesis H0, unit root, the time series be present in the time series
It is jiggly, then again to Δ ytCarry out unit root test；
If | t | less than the critical value, refusal null hypothesis H0, receive alternative hvpothesis H1, unit root is not present in time series,
I.e. described time series is stable, determines that the time series is singly whole for N ranks；N is the current number for carrying out unit root test.
6. the change early warning system of electricity needs according to claim 5, it is characterised in that association's mould preparation block is also used
In：
All time serieses are formed into the rank column vector of k × 1：Yt=(y1t, y2t..., ykt) ', YtGather for vector, y1t,y2t,…,
yktFor 1~k of sequence number time series；
Judge whether the vector set meets following two conditions：
yit~I (d), i=1,2 ..., k, YtIn each time series the whole exponent number of list be equal to the reference index data rank
Number d；
In the presence of the rank column vector β of a k × 1=(β1, β2..., βk) ' so that(β
≠0)；β is to assist whole vector, β element β1,β2,…,βkTo assist whole parameter, b is the exponent number of the default whole parameter of association, and b is less than d；
If satisfied, then there is the whole property of association in the achievement data with the reference index data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410206780.5A CN103996076B (en) | 2014-05-15 | 2014-05-15 | The change method for early warning and system of electricity needs |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410206780.5A CN103996076B (en) | 2014-05-15 | 2014-05-15 | The change method for early warning and system of electricity needs |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103996076A CN103996076A (en) | 2014-08-20 |
| CN103996076B true CN103996076B (en) | 2017-11-21 |
Family
ID=51310236
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410206780.5A Active CN103996076B (en) | 2014-05-15 | 2014-05-15 | The change method for early warning and system of electricity needs |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103996076B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104483958B (en) * | 2014-10-31 | 2017-05-31 | 中国石油大学(北京) | A kind of complicated refinery process adaptive data-driven method for diagnosing faults and device |
| CN105930982B (en) * | 2016-05-11 | 2018-01-19 | 中国南方电网有限责任公司电网技术研究中心 | The measuring method and system of electric power consumer confidence index |
| CN106022525B (en) * | 2016-05-24 | 2018-03-30 | 中国南方电网有限责任公司电网技术研究中心 | Power planning scheme based on consumer confidence index index determines method and device |
| CN111612276A (en) * | 2020-05-29 | 2020-09-01 | 云南电网有限责任公司 | Power load characteristic index prediction method and device |
| CN111897852A (en) * | 2020-07-01 | 2020-11-06 | 中国建设银行股份有限公司 | Abnormal data determination method and device, electronic equipment and readable storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930155A (en) * | 2012-10-30 | 2013-02-13 | 国网能源研究院 | Method and device for acquiring early-warming parameters of power demands |
-
2014
- 2014-05-15 CN CN201410206780.5A patent/CN103996076B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930155A (en) * | 2012-10-30 | 2013-02-13 | 国网能源研究院 | Method and device for acquiring early-warming parameters of power demands |
Non-Patent Citations (5)
| Title |
|---|
| 中国电力行业周期波动特征及电力需求影响因素分析--基于景气分析及误差修正模型的研究;刘畅等;《资源科学》;20110420;第33卷(第1期);全文 * |
| 中国能源需求建模与实证分析;苏泽雄;《中国优秀硕士论文全文数据库（电子期刊）社会科学I辑》;20030615(第2期);第6-52页 * |
| 中国能源需求的建模与实证分析;韩君;《中国优秀硕士学位论文全文数据库（电子期刊）经济与管理科学辑》;20071115(第5期);全文 * |
| 基于协整理论和误差修正模型的电网投资需求预测研究;赵会茹等;《电网技术》;20110930;第35卷(第9期);全文 * |
| 基于经济先行指标的省级电力市场需求分析;崔巍等;《水电能源科学》;20120531;第30卷(第5期);第191-195页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103996076A (en) | 2014-08-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110097297B (en) | Multi-dimensional electricity stealing situation intelligent sensing method, system, equipment and medium | |
| CN103996076B (en) | The change method for early warning and system of electricity needs | |
| CN108593990B (en) | Electricity stealing detection method based on electricity consumption behavior mode of electric energy user and application | |
| Ma et al. | A method for multiple periodic factor prediction problems using complex fuzzy sets | |
| Costantini et al. | A hierarchical procedure for the combination of forecasts | |
| Fay et al. | On the influence of weather forecast errors in short-term load forecasting models | |
| CN108647891A (en) | Data exception classification, Reasons method and device | |
| Kapoor et al. | Weather forecasting using sliding window algorithm | |
| CN107992968A (en) | Electric energy meter measurement error Forecasting Methodology based on integrated techniques of teime series analysis | |
| CN105117810A (en) | Residential electricity consumption mid-term load prediction method under multistep electricity price mechanism | |
| CN102930155B (en) | Obtain the method and device of the early-warning parameters of electricity needs | |
| CN106251008A (en) | A kind of photovoltaic power Forecasting Methodology chosen based on combining weights similar day | |
| CN104881718B (en) | Region electric power consumer confidence index construction method based on multiple dimensioned economic leading indicators | |
| CN105867341B (en) | A kind of the online equipment health status self checking method and system of tobacco processing equipment | |
| CN107133825A (en) | The Forecasting Methodology and device of a kind of businessman's volume of the flow of passengers | |
| Voronin et al. | A hybrid electricity price forecasting model for the Nordic electricity spot market | |
| CN103729683A (en) | Measurable asset life evaluation method | |
| Oludolapo et al. | Comparing performance of MLP and RBF neural network models for predicting South Africa's energy consumption | |
| Capuno et al. | Very Short‐Term Load Forecasting Using Hybrid Algebraic Prediction and Support Vector Regression | |
| CN107992991A (en) | Annual electricity sales amount Forecasting Methodology based on external environmental factor and Co-integration Theory | |
| CN106874676A (en) | A kind of electric power meter state evaluating method | |
| CN105913366A (en) | Industrial electric power big data-based regional industry business climate index building method | |
| Rezitis et al. | Energy consumption and economic growth in South and Southeast Asian countries: Evidence from a dynamic panel data approach | |
| CN108491955A (en) | A kind of monthly industrial power predicating method of integrated index screening and temperature correction | |
| CN106952026A (en) | A kind of social stability methods of risk assessment to large complicated engineering project |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |