CN105096159A - Method and device for predicting regional electricity sales - Google Patents

Abstract

The invention discloses a method and a device for predicting regional electricity sales. The method comprises steps of performing determination of time domain and frequency domain characteristics on history electricity sale curves of regions to be predicted, performing clustering on each region to be predicted, wherein, in one cluster group, the characteristics of the history electricity sales curves of all region to be predicted are identical under the time domain and the frequency domain, and the characteristic is used as the characteristic label of the cluster group, and, for every cluster group, choosing a preset prediction algorithm corresponding to the characteristic label of the cluster group to perform electricity sales prediction on all regions to be predicted in the cluster group. The invention performs cluster partitioning on various regions to be predicted, adopts matched identical prediction algorithm to perform uniform prediction on the regions having same history sales curve characteristics, and improves the accuracy of the regional electricity sales.

Description

A kind of region electricity sales amount Forecasting Methodology and device

Technical field

The application relates to electric power data processing technology field, more particularly, relates to a kind of region electricity sales amount Forecasting Methodology and device.

Background technology

Along with deepening continuously of intelligent grid construction and power system reform, monthly electricity sales amount prediction formulates monthly production and management mode for grid company, instruct the reasonable operation of generating plant and transmission and distribution network, formulate ordered electric scheme, promote the development and construction of electricity market all tool be of great significance.

Present inventor is by studying existing electricity sales amount Forecasting Methodology, find that it exists following defect: existing electricity sales amount Forecasting Methodology is generally all use same prediction algorithm for each electricity consumption region, it also reckons without the different qualities of each department power consumption curve, only utilize the electricity sales amount of a kind of prediction algorithm to many areas to predict, cause precision of prediction not high.

Summary of the invention

In view of this, this application provides a kind of region electricity sales amount Forecasting Methodology and device, all using same prediction algorithm for solving existing electricity sales amount Forecasting Methodology for different electricity consumption region, causing the problem that precision of prediction is not high.

To achieve these goals, the existing scheme proposed is as follows:

A kind of region electricity sales amount Forecasting Methodology, comprising:

Obtain the history electricity sales amount curve in each region to be predicted;

Determine the feature of history electricity sales amount curve under time domain and frequency domain in each region to be predicted;

According to the feature of history electricity sales amount curve under time domain and frequency domain in each region to be predicted, cluster is carried out to each region to be predicted, in same phylogenetic group, the feature of history electricity sales amount curve under time domain with frequency domain in each region to be predicted is identical, and this feature is as the feature tag of phylogenetic group;

For each phylogenetic group, the preset prediction algorithm corresponding with feature tag that is this phylogenetic group is selected to carry out electricity sales amount prediction to the region to be predicted of each in this phylogenetic group.

Preferably, in the feature of history electricity sales amount curve under time domain and frequency domain according to each region to be predicted, when carrying out cluster to each region to be predicted, vision clustering algorithm is adopted to carry out cluster.

Preferably, the described feature of history electricity sales amount curve under time domain and frequency domain determining each region to be predicted, comprising:

Determine that the time domain of the history electricity sales amount curve in each region to be predicted under time domain is with making a gesture of measuring and time domain chain rate amount;

Determine the frequency domain extreme difference point amplitude of the history electricity sales amount curve in each region to be predicted under frequency domain and Fei Shi index;

Time domain is utilized to determine whether the history electricity sales amount curve in corresponding region to be predicted possesses chronicity with making a gesture of measuring;

Whether the history electricity sales amount curve in corresponding region to be predicted possesses stationarity to utilize time domain chain rate amount to determine;

Whether the history electricity sales amount curve of corresponding estimation range possesses perturbation to utilize frequency domain extreme difference point amplitude to determine;

Whether the history electricity sales amount curve of corresponding estimation range possesses periodically to utilize Fei Shi index to determine.

Preferably, the feature tag of each phylogenetic group of cluster formation comprises:

Fisrt feature label: possess chronicity, stationarity and periodicity simultaneously, do not possess perturbation;

Second feature label: possess chronicity, periodicity and perturbation simultaneously, do not possess stationarity;

Third feature label: possess stationarity, periodicity and perturbation simultaneously, do not possess chronicity;

Fourth feature label: simultaneously possess chronicity and perturbation, does not possess stationarity and periodicity;

Fifth feature label: possess stationarity and periodicity simultaneously, do not possess chronicity and perturbation.

Preferably, the preset prediction algorithm corresponding with each feature tag comprises:

The BP neural network algorithm corresponding with described fisrt feature label;

The L corresponding with described second feature label _1/2openness regression algorithm;

The support vector machines regression algorithm corresponding with described third feature label;

The L corresponding with described fourth feature label _1/2openness regression algorithm;

The difference ARMA model ARIMA algorithm corresponding with described fifth feature label.

A kind of region electricity sales amount prediction unit, comprising:

Curve acquisition unit, for obtaining the history electricity sales amount curve in each region to be predicted;

Characteristics determining unit, for determining the feature of history electricity sales amount curve under time domain and frequency domain in each region to be predicted;

Clustering processing unit, for the feature of history electricity sales amount curve under time domain and frequency domain according to each region to be predicted, cluster is carried out to each region to be predicted, in same phylogenetic group, the feature of history electricity sales amount curve under time domain with frequency domain in each region to be predicted is identical, and this feature is as the feature tag of phylogenetic group;

Algorithmic match unit, for for each phylogenetic group, selects the preset prediction algorithm corresponding with feature tag that is this phylogenetic group to carry out electricity sales amount prediction to the region to be predicted of each in this phylogenetic group.

Preferably, described clustering processing unit comprises:

First clustering processing subelement, for adopting vision clustering algorithm, according to the feature of history electricity sales amount curve under time domain and frequency domain in each region to be predicted, carries out cluster to each region to be predicted.

Preferably, described characteristics determining unit comprises:

Temporal signatures determining unit, for determining that the time domain of the history electricity sales amount curve in each region to be predicted under time domain is with making a gesture of measuring and time domain chain rate amount;

Frequency domain character determining unit, for determining the frequency domain extreme difference point amplitude of the history electricity sales amount curve in each region to be predicted under frequency domain and Fei Shi index;

With making a gesture of measuring, chronicity determining unit, determines whether the history electricity sales amount curve in corresponding region to be predicted possesses chronicity for utilizing time domain;

Stationarity determining unit, determines for utilizing time domain chain rate amount whether the history electricity sales amount curve in corresponding region to be predicted possesses stationarity;

Perturbation determining unit, determines for utilizing frequency domain extreme difference point amplitude whether the history electricity sales amount curve of corresponding estimation range possesses perturbation;

Periodically determining unit, determines for utilizing Fei Shi index whether the history electricity sales amount curve of corresponding estimation range possesses periodically.

Preferably, the feature tag of each phylogenetic group of cluster formation comprises:

Fisrt feature label: possess chronicity, stationarity and periodicity simultaneously, do not possess perturbation;

Second feature label: possess chronicity, periodicity and perturbation simultaneously, do not possess stationarity;

Third feature label: possess stationarity, periodicity and perturbation simultaneously, do not possess chronicity;

Fourth feature label: simultaneously possess chronicity and perturbation, does not possess stationarity and periodicity;

Fifth feature label: possess stationarity and periodicity simultaneously, do not possess chronicity and perturbation.

Preferably, the preset prediction algorithm corresponding with each feature tag comprises:

The BP neural network algorithm corresponding with described fisrt feature label;

The L corresponding with described second feature label _1/2openness regression algorithm;

The support vector machines regression algorithm corresponding with described third feature label;

The L corresponding with described fourth feature label _1/2openness regression algorithm;

The difference ARMA model ARIMA algorithm corresponding with described fifth feature label.

As can be seen from above-mentioned technical scheme, the region electricity sales amount Forecasting Methodology that the embodiment of the present application provides, history electricity sales amount curve for regional to be predicted carries out the determination of time domain and frequency domain character, and accordingly cluster is carried out to each region to be predicted, in same phylogenetic group, the feature of history electricity sales amount curve under time domain with frequency domain in each region to be predicted is identical, and this feature is as the feature tag of phylogenetic group, last for each phylogenetic group, the preset prediction algorithm corresponding with feature tag that is this phylogenetic group is selected to carry out electricity sales amount prediction to the region to be predicted of each in this phylogenetic group.The application's method, by carrying out clustering to difference region to be predicted, adopts the identical prediction algorithm of coupling to carry out unifying prediction for the region that history electricity sales amount curvilinear characteristic is identical, thus improves the degree of accuracy to the prediction of region electricity sales amount.

Accompanying drawing explanation

In order to be illustrated more clearly in the embodiment of the present application or technical scheme of the prior art, be briefly described to the accompanying drawing used required in embodiment or description of the prior art below, apparently, accompanying drawing in the following describes is only the embodiment of the application, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to the accompanying drawing provided.

Fig. 1 is the structural representation of BP neural network;

Fig. 2 is a kind of region electricity sales amount Forecasting Methodology process flow diagram disclosed in the embodiment of the present application;

Fig. 3 is a kind of method flow diagram determining history electricity sales amount curve feature under time domain and frequency domain in region to be predicted disclosed in the embodiment of the present application;

Fig. 4 is a kind of 27 company's cluster result schematic diagram disclosed in the embodiment of the present application;

Fig. 5 is a kind of region electricity sales amount prediction unit structural representation disclosed in the embodiment of the present application.

Embodiment

Before introducing the application's scheme, some nouns used herein and phrase make an explanation by first-selection:

1, cluster: so-called cluster, refers to and divides into groups (Data Placement) to the data with common trend or structure.Data item is grouped into multiple class, and the data differences between class should be large as far as possible, and the data differences in class should be little as far as possible.Namely " minimize the similarity between class, maximize similar interior similarity ".Cluster analysis can set up the concept of macroscopic view, finds the distribution pattern of data.

2, vision clustering algorithm: in vision clustering algorithm, each data point is considered a luminous point in space, so data set pair should constitute the piece image in space.When this image of obfuscation, first each little luminous point is changed to a small light spot.Jin mono-Bu Di Steamed sticks with paste, and making small light spot molten is gradually large spot.When resolution is fully low, whole figure just becomes a hot spot.If each hot spot to be regarded as a class, then the process of above-mentioned obfuscation just forms clustering tree step by step, and node represents the class of different scale cluster, and the class of class representated by son's node of father's node representative is fused to be formed.

To a given data set X={x _i∈ R ^d: i=1, L, N}, by each data point x _iregard a little luminous point as, mathematically this luminous point is by Dirac generalized function δ (x-x _i) represent.So data set X forms piece image p (x) in space, wherein:

$p\left(x\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\mathrm{\δ}(x-x_i)---\left(1\right)$

According to the Scale-space theory of vision front-end system, the multi-scale Representation of image p (x) is p (x) and Gaussian function $g(x,\mathrm{\σ})={\left(3{\mathrm{\π\σ}}^2\right)}^{-\frac{d}{2}}{e}^{-\frac{x^{T}x}{2{\mathrm{\σ}}^2}}$ Convolution, namely

$p(x,\mathrm{\σ})=p\left(x\right)*g(x,\mathrm{\σ})={\left(2{\mathrm{\π\σ}}^2\right)}^{-\frac{d}{2}}\underset{i=1}{\overset{N}{\Σ}}{e}^{-\frac{{(x-x_i)}^T(x-x_i)}{2{\mathrm{\σ}}^2}}---\left(2\right)$

Wherein, the window width σ of Gaussian function is called scale parameter.Giving under dimensioning σ, the center that we define a hot spot is the maximum point of p (x, σ) about x, and corresponding to a center x ^*hot spot be then defined as x ^*about gradient system basin of attraction, be designated as B (x ^*), namely

$B\left(x^{*}\right)=\{x_0\∈R^d:\underset{t\→\mathrm{\∞}}{\lim }x(t,x_0)=x^{*}\}---\left(3\right)$

Here x (t, x ₀) be the solution of gradient system initial-value problem:

$\left\{\begin{array}{c}\frac{dx}{dt}={\▿}_{x}p(x,\mathrm{\σ})\\ x(0,x_0)=x_0\end{array}\right.---\left(4\right)$

Therefore, under dimensioning, 1 x is verified ₀whether belong to a hot spot B (x ^*) come by numerical solution equation (4).

3, BP neural network algorithm: BP neural network is one of important models of artificial neural network.BP neural network is a kind of Multi-layered Feedforward Networks of one way propagation, and it comprises input layer, hidden layer and output layer, is a kind of model that application is more at present.This algorithm adopts the reverse propagation mode of learning of error in Hierarchical network structure, and learning process is made up of forward-propagating and the reverse propagation of error.

The structure of BP neural network as shown in Figure 1, the main thought of algorithm is that learning process is divided into two stages: the first stage is forward-propagating process, input information successively calculates the real output value of each unit through hidden layer from input layer, the neuronic state of every one deck only has an impact to the neuronic state of lower one deck; Subordinate phase is back-propagation process, the output valve obtaining expecting if fail at output layer, then step-by-step recursion calculates the actual difference exported between desired output, makes error signal tend to minimum according to one deck weights before this error correction.It is by continuously computational grid weights and change of error and approach target gradually on the direction declined relative to error function slope.

4, L _1/2openness recurrence

Data-oriented collection: x _i∈ R ^m, y _i∈ R, the essence of regression problem is from Learning machine F, find optimum function f ^*approach x with best ground, the unknown relation between y, method popular at present adopts L _pthe method of regularization framework is as follows:

$f^{*}=\underset{f\∈F}{\arg min}\{\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}l\left(f\right(x_i),y_i)+\mathrm{\λ}|\left|f\right|{|}_p\}---\left(5\right)$

Wherein l (. .) be loss function, when predicted value and actual value close to time, loss function is minimum.λ is the regularization parameter of control machine complexity.|| f|| _pfor certain norm of separating, represent and certain solution is expected (as slickness, openness etc.).When usually solving regression problem, often can run into the problem of over-fitting, regularization framework has taken into full account this situation, by increasing || f|| _pcome to increase certain constraint to solution, reduce the risk of over-fitting.

When time, i.e. L _1/2openness, ensure that solvability on the one hand, ensure the openness of understanding on the other hand.

5, SVM regression algorithm

It is the application of support vector in function regression field that SVM returns.The sample point that SVM returns only has a class, and sought optimal hyperlane makes all sample points minimum from lineoid " total departure ".At this moment sample point is all between two boundary lines, asks optimum regression lineoid to be equivalent to equally and asks largest interval.

Data-oriented collection D: x _i∈ R ^m, y _i∈ R, the essence of regression problem finds function f (x), to infer the y value that any one pattern x is corresponding.

6, difference ARMA model ARIMA algorithm

Time series be arrange in chronological order, in time change and the data sequence that is mutually related.Analysis time sequence method composition data analyze key areas, i.e. a time series analysis.

Conventional model AR (p), MA (q), ARMA (p, q) be all for stationary time series, for nonstationary time series, need to adopt method of difference, eliminate its tendency and seasonality, the sequence after making it convert is stationary sequence.

1 jump divides: $\▿X_t=X_t-X_{t-1}=(1-B)X_t$

2 jumps divide: ${\▿}^2{X}_t=X_t-2X_{t-1}+X_{t-2}={(1-B)}^2{X}_t$

D jump divides: ${\▿}^d{X}_t={(1-B)}^d{X}_t$

If { X _t, t=0, ± 1, ± 2 ... non-stationary series, if there is positive integer d, make: and { W _t, t=0, ± 1, ± 2 ... ARMA (p, q) sequence, then claim X _taRIMA (p, d, q) sequence, at this moment, X _tmeet

Wherein:

θ(B)＝1-θ ₁B-θ ₂B ²-…θ _qB ^q

Below in conjunction with the accompanying drawing in the embodiment of the present application, be clearly and completely described the technical scheme in the embodiment of the present application, obviously, described embodiment is only some embodiments of the present application, instead of whole embodiments.Based on the embodiment in the application, those of ordinary skill in the art are not making the every other embodiment obtained under creative work prerequisite, all belong to the scope of the application's protection.

See Fig. 2, Fig. 2 a kind of region electricity sales amount Forecasting Methodology process flow diagram disclosed in the embodiment of the present application.

As shown in Figure 2, the method comprises:

Step S200, obtain the history electricity sales amount curve in each region to be predicted;

Particularly, history electricity sales amount curve can be drawn by before current date year, 2 years or electricity sales amount At All Other Times in section and draw.

Step S210, determine the feature of history electricity sales amount curve under time domain and frequency domain in each region to be predicted;

Time series analysis is horizontal ordinate with time, is that ordinate is to analyze the relation of electricity sales amount and time with electricity sales amount.Frequency-domain analysis is horizontal ordinate with frequency, is that ordinate is to analyze the feature in each frequency with amplitude.In general, time series analysis comparison image is with directly perceived, and frequency-domain analysis is then more terse at data Angle, and more can portray curvilinear characteristic.No matter being temporal aspect or frequency domain character, is all that history electricity sales amount curve by treating estimation range carries out Fourier transform and obtains.

Step S220, the feature of history electricity sales amount curve under time domain and frequency domain according to each region to be predicted, carry out cluster to each region to be predicted;

This step cluster obtains multiple phylogenetic group, and in same phylogenetic group, the feature of history electricity sales amount curve under time domain with frequency domain in each region to be predicted is identical, and this feature is as the feature tag of phylogenetic group.

Step S230, for each phylogenetic group, the preset prediction algorithm corresponding with feature tag that is this phylogenetic group is selected to carry out electricity sales amount prediction to the region to be predicted of each in this phylogenetic group.

Due to each phylogenetic group of cluster gained, the history electricity sales amount curve in each region to be predicted in phylogenetic group possesses identical time domain and frequency domain character, therefore for the region to be predicted of this kind of feature, adopts the algorithm of same coupling to carry out electricity sales amount prediction.

The region electricity sales amount Forecasting Methodology that the embodiment of the present application provides, history electricity sales amount curve for regional to be predicted carries out the determination of time domain and frequency domain character, and accordingly cluster is carried out to each region to be predicted, in same phylogenetic group, the feature of history electricity sales amount curve under time domain with frequency domain in each region to be predicted is identical, and this feature is as the feature tag of phylogenetic group, last for each phylogenetic group, select the preset prediction algorithm corresponding with feature tag that is this phylogenetic group to carry out electricity sales amount prediction to the region to be predicted of each in this phylogenetic group.The application's method, by carrying out clustering to difference region to be predicted, adopts the identical prediction algorithm of coupling to carry out unifying prediction for the region that history electricity sales amount curvilinear characteristic is identical, thus improves the degree of accuracy to the prediction of region electricity sales amount.

Optionally, when carrying out clustering processing, vision clustering algorithm can be adopted to carry out cluster.

In another embodiment of the application, disclose a kind of method determining the feature of history electricity sales amount curve under time domain and frequency domain in region to be predicted, as shown in Figure 3, the method comprises:

Step S300, determine that the time domain of the history electricity sales amount curve in each region to be predicted under time domain is with making a gesture of measuring and time domain chain rate amount;

Step S310, determine the frequency domain extreme difference point amplitude of the history electricity sales amount curve in each region to be predicted under frequency domain and Fei Shi index;

History electricity sales amount curve, after Fourier transform, obtains the Fourier coefficient of this curve under frequency (being also amplitude).

In addition, include Fei Shi index in consideration, obtained the seasonal effect in time series periodogram of electricity sales amount by Fourier transform, and utilize fisherg statistic sense cycle figure peak value.

Step S320, time domain is utilized to determine whether the history electricity sales amount curve in corresponding region to be predicted possesses chronicity with making a gesture of measuring;

Time domain then means that curve chronicity is obvious greatly with making a gesture of measuring, and is thus defined as possessing chronicity, otherwise, determine that it does not possess chronicity.

Step S330, whether the history electricity sales amount curve in corresponding region to be predicted possesses stationarity to utilize time domain chain rate amount to determine;

Time domain chain rate amount is little, means that curve stationarity is obvious, is thus defined as possessing stationarity, otherwise, determine that it does not possess stationarity.

Step S340, whether the history electricity sales amount curve of corresponding estimation range possesses perturbation to utilize frequency domain extreme difference point amplitude to determine;

Frequency domain extreme difference point amplitude means that greatly frequency domain character is large, represents curve perturbation obvious simultaneously, is thus defined as possessing perturbation, otherwise, determine that it does not possess perturbation.

Step S350, whether the history electricity sales amount curve of corresponding estimation range possesses periodically to utilize Fei Shi index to determine.

Fei Shi index means that greatly curve cycle is obvious, is thus defined as possessing periodically, otherwise, determine that it does not possess periodically.

According to above-mentioned electricity sales amount curvilinear characteristic, the feature tag of each phylogenetic group that cluster is formed can be determined.

Combined by above-mentioned several electricity sales amount curvilinear characteristic, obtain following several feature tag:

Fisrt feature label: possess chronicity, stationarity and periodicity simultaneously, do not possess perturbation;

Second feature label: possess chronicity, periodicity and perturbation simultaneously, do not possess stationarity;

Third feature label: possess stationarity, periodicity and perturbation simultaneously, do not possess chronicity;

Fourth feature label: simultaneously possess chronicity and perturbation, does not possess stationarity and periodicity;

Fifth feature label: possess stationarity and periodicity simultaneously, do not possess chronicity and perturbation.

According to the difference of feature tag, we preset prediction algorithm corresponding with it.Through the research of inventor to various prediction algorithm, find to set according to following corresponding relation, can be more accurate predict each region electricity sales amount.

Concrete corresponding relation is:

The BP neural network algorithm corresponding with described fisrt feature label;

The L corresponding with described second feature label _1/2openness regression algorithm;

The support vector machines regression algorithm corresponding with described third feature label;

The L corresponding with described fourth feature label _1/2openness regression algorithm;

The difference ARMA model ARIMA algorithm corresponding with described fifth feature label.

For the introduction of above-mentioned various prediction algorithm, can with reference to the introduction of relevant portion above.

Next, in order to verify the validity of the application's scheme further, be described by instantiation.

Extract electricity sales amount data from the marketing information system of State Grid Corporation of China, and make electricity sales amount curve.Belonging to the data extracted, the time is 01 month in Dec, 2014 in 2010, and regional extent is 27 province (city) companies of State Grid Corporation of China and subordinate.

From the temporal signatures of electricity sales amount curve, have chosen monthly on year-on-year basis with 24 data of chain rate, if the time domain in table 1 is with making a gesture of measuring and time domain chain rate figureofmerit, embody the monthly variation tendency of electricity sales amount.

From the frequency domain character of electricity sales amount curve, original electricity sales amount curve, after Fourier transform, obtains the Fourier coefficient of this curve under frequency.By comparing the electricity sales amount Fourier coefficient size of 27 province (city) companies, the feature difference of electricity sales amount under frequency domain can be embodied.Specifically choose amplitude and the Fei Shi index of 7 extreme difference amplitude Frequency points, as shown in table 1, following table 1 is the feature clustering input pointer of 27 province (city) companies.

Table 1

After utilizing vision clustering algorithm to carry out cluster, obtain region clustering result as shown in Figure 4.The feature of 5 clusters is as shown in table 2 below:

Classification/feature	With making a gesture of measuring	Chain rate amount	Fei Shi index	Frequency domain extreme difference point amplitude
					Cluster 1	Larger	Less	Larger
Cluster 2	Larger		Larger	Larger
					Cluster 3		Less	Larger	Larger
Cluster 4	Larger			Larger

Classification/feature	With making a gesture of measuring	Chain rate amount	Fei Shi index	Frequency domain extreme difference point amplitude
					Cluster 5		Less	Larger

Table 2

From the feature of 5 clusters, or obvious with making a gesture of measuring larger meaning chronicity, chain rate amount is less means that stationarity is obvious, and Fei Shi index means that more greatly periodically obviously frequency domain extreme difference point amplitude means that greatly perturbation is obvious.

In conjunction with the chronicity of electricity sales amount curve in cluster structures, stationarity, periodicity and perturbation, according to the self-characteristic of different Forecasting Methodology, extract following prediction algorithm and the curvilinear characteristic coupling table of comparisons:

Classification/feature	Chronicity	Stationarity	Periodically	Perturbation	Be suitable for prediction algorithm
						Cluster 1	Obviously	Obviously	Obviously		BP neural network
Cluster 2	Obviously		Obviously	Obviously	L 1/2Openness recurrence
						Cluster 3		Obviously	Obviously	Obviously	SVM returns
Cluster 4	Obviously			Obviously	L 1/2Openness recurrence
						Cluster 5		Obviously	Obviously		BP neural network

Table 3

According to above-mentioned prediction algorithm and the curvilinear characteristic coupling table of comparisons, carry out the monthly prediction of the electricity sales amount of (city) company of each province.The data 01 month in Dec, 2013 in 2010 utilizing 27 province (city) companies are training data, and year Dec in January, 2014 to 2014 is test data, obtains following precision of prediction:

Table 4

Can be found out by upper table 4, when using the scheme of the application to predict zones of different, predicated error is very little, further demonstrates the validity of the application's scheme.

Be described the region electricity sales amount prediction unit that the embodiment of the present application provides below, region described below electricity sales amount prediction unit can mutual corresponding reference with above-described region electricity sales amount Forecasting Methodology.

See Fig. 5, Fig. 5 a kind of region electricity sales amount prediction unit structural representation disclosed in the embodiment of the present application.

As shown in Figure 5, this device comprises:

Curve acquisition unit 51, for obtaining the history electricity sales amount curve in each region to be predicted;

Characteristics determining unit 52, for determining the feature of history electricity sales amount curve under time domain and frequency domain in each region to be predicted;

Clustering processing unit 53, for the feature of history electricity sales amount curve under time domain and frequency domain according to each region to be predicted, cluster is carried out to each region to be predicted, in same phylogenetic group, the feature of history electricity sales amount curve under time domain with frequency domain in each region to be predicted is identical, and this feature is as the feature tag of phylogenetic group;

Algorithmic match unit 54, for for each phylogenetic group, selects the preset prediction algorithm corresponding with feature tag that is this phylogenetic group to carry out electricity sales amount prediction to the region to be predicted of each in this phylogenetic group.

Optionally, described clustering processing unit comprises:

First clustering processing subelement, for adopting vision clustering algorithm, according to the feature of history electricity sales amount curve under time domain and frequency domain in each region to be predicted, carries out cluster to each region to be predicted.

Optionally, described characteristics determining unit comprises:

Temporal signatures determining unit, for determining that the time domain of the history electricity sales amount curve in each region to be predicted under time domain is with making a gesture of measuring and time domain chain rate amount;

Frequency domain character determining unit, for determining the frequency domain extreme difference point amplitude of the history electricity sales amount curve in each region to be predicted under frequency domain and Fei Shi index;

With making a gesture of measuring, chronicity determining unit, determines whether the history electricity sales amount curve in corresponding region to be predicted possesses chronicity for utilizing time domain;

Stationarity determining unit, determines for utilizing time domain chain rate amount whether the history electricity sales amount curve in corresponding region to be predicted possesses stationarity;

Perturbation determining unit, determines for utilizing frequency domain extreme difference point amplitude whether the history electricity sales amount curve of corresponding estimation range possesses perturbation;

Periodically determining unit, determines for utilizing Fei Shi index whether the history electricity sales amount curve of corresponding estimation range possesses periodically.

Optionally, the feature tag of each phylogenetic group of cluster formation comprises:

Fisrt feature label: possess chronicity, stationarity and periodicity simultaneously, do not possess perturbation;

Second feature label: possess chronicity, periodicity and perturbation simultaneously, do not possess stationarity;

Third feature label: possess stationarity, periodicity and perturbation simultaneously, do not possess chronicity;

Fourth feature label: simultaneously possess chronicity and perturbation, does not possess stationarity and periodicity;

Fifth feature label: possess stationarity and periodicity simultaneously, do not possess chronicity and perturbation.

Optionally, the preset prediction algorithm corresponding with each feature tag comprises:

The BP neural network algorithm corresponding with described fisrt feature label;

The L corresponding with described second feature label _1/2openness regression algorithm;

The support vector machines regression algorithm corresponding with described third feature label;

The L corresponding with described fourth feature label _1/2openness regression algorithm;

The difference ARMA model ARIMA algorithm corresponding with described fifth feature label.

The region electricity sales amount prediction unit that the embodiment of the present application provides, history electricity sales amount curve for regional to be predicted carries out the determination of time domain and frequency domain character, and accordingly cluster is carried out to each region to be predicted, in same phylogenetic group, the feature of history electricity sales amount curve under time domain with frequency domain in each region to be predicted is identical, and this feature is as the feature tag of phylogenetic group, last for each phylogenetic group, select the preset prediction algorithm corresponding with feature tag that is this phylogenetic group to carry out electricity sales amount prediction to the region to be predicted of each in this phylogenetic group.The application's device, by carrying out clustering to difference region to be predicted, adopts the identical prediction algorithm of coupling to carry out unifying prediction for the region that history electricity sales amount curvilinear characteristic is identical, thus improves the degree of accuracy to the prediction of region electricity sales amount.

Finally, also it should be noted that, in this article, the such as relational terms of first and second grades and so on is only used for an entity or operation to separate with another entity or operational zone, and not necessarily requires or imply the relation that there is any this reality between these entities or operation or sequentially.And, term " comprises ", " comprising " or its any other variant are intended to contain comprising of nonexcludability, thus make to comprise the process of a series of key element, method, article or equipment and not only comprise those key elements, but also comprise other key elements clearly do not listed, or also comprise by the intrinsic key element of this process, method, article or equipment.When not more restrictions, the key element limited by statement " comprising ... ", and be not precluded within process, method, article or the equipment comprising described key element and also there is other identical element.

In this instructions, each embodiment adopts the mode of going forward one by one to describe, and what each embodiment stressed is the difference with other embodiments, between each embodiment identical similar portion mutually see.

To the above-mentioned explanation of the disclosed embodiments, professional and technical personnel in the field are realized or uses the application.To be apparent for those skilled in the art to the multiple amendment of these embodiments, General Principle as defined herein when not departing from the spirit or scope of the application, can realize in other embodiments.Therefore, the application can not be restricted to these embodiments shown in this article, but will meet the widest scope consistent with principle disclosed herein and features of novelty.

Claims (10)

1. a region electricity sales amount Forecasting Methodology, is characterized in that, comprising:

Obtain the history electricity sales amount curve in each region to be predicted;

Determine the feature of history electricity sales amount curve under time domain and frequency domain in each region to be predicted;

According to the feature of history electricity sales amount curve under time domain and frequency domain in each region to be predicted, cluster is carried out to each region to be predicted, in same phylogenetic group, the feature of history electricity sales amount curve under time domain with frequency domain in each region to be predicted is identical, and this feature is as the feature tag of phylogenetic group;

For each phylogenetic group, the preset prediction algorithm corresponding with feature tag that is this phylogenetic group is selected to carry out electricity sales amount prediction to the region to be predicted of each in this phylogenetic group.

2. method according to claim 1, is characterized in that, in the feature of history electricity sales amount curve under time domain and frequency domain according to each region to be predicted, when carrying out cluster to each region to be predicted, adopts vision clustering algorithm to carry out cluster.

3. method according to claim 1, is characterized in that, the described feature of history electricity sales amount curve under time domain and frequency domain determining each region to be predicted, comprising:

Determine that the time domain of the history electricity sales amount curve in each region to be predicted under time domain is with making a gesture of measuring and time domain chain rate amount;

Determine the frequency domain extreme difference point amplitude of the history electricity sales amount curve in each region to be predicted under frequency domain and Fei Shi index;

Time domain is utilized to determine whether the history electricity sales amount curve in corresponding region to be predicted possesses chronicity with making a gesture of measuring;

Whether the history electricity sales amount curve in corresponding region to be predicted possesses stationarity to utilize time domain chain rate amount to determine;

Whether the history electricity sales amount curve of corresponding estimation range possesses perturbation to utilize frequency domain extreme difference point amplitude to determine;

Whether the history electricity sales amount curve of corresponding estimation range possesses periodically to utilize Fei Shi index to determine.

4. method according to claim 3, is characterized in that, the feature tag of each phylogenetic group that cluster is formed comprises:

Fisrt feature label: possess chronicity, stationarity and periodicity simultaneously, do not possess perturbation;

Second feature label: possess chronicity, periodicity and perturbation simultaneously, do not possess stationarity;

Third feature label: possess stationarity, periodicity and perturbation simultaneously, do not possess chronicity;

Fourth feature label: simultaneously possess chronicity and perturbation, does not possess stationarity and periodicity;

Fifth feature label: possess stationarity and periodicity simultaneously, do not possess chronicity and perturbation.

5. method according to claim 4, is characterized in that, the preset prediction algorithm corresponding with each feature tag comprises:

The BP neural network algorithm corresponding with described fisrt feature label;

The L corresponding with described second feature label _1/2openness regression algorithm;

The support vector machines regression algorithm corresponding with described third feature label;

The L corresponding with described fourth feature label _1/2openness regression algorithm;

The difference ARMA model ARIMA algorithm corresponding with described fifth feature label.

6. a region electricity sales amount prediction unit, is characterized in that, comprising:

Curve acquisition unit, for obtaining the history electricity sales amount curve in each region to be predicted;

Characteristics determining unit, for determining the feature of history electricity sales amount curve under time domain and frequency domain in each region to be predicted;

Clustering processing unit, for the feature of history electricity sales amount curve under time domain and frequency domain according to each region to be predicted, cluster is carried out to each region to be predicted, in same phylogenetic group, the feature of history electricity sales amount curve under time domain with frequency domain in each region to be predicted is identical, and this feature is as the feature tag of phylogenetic group;

Algorithmic match unit, for for each phylogenetic group, selects the preset prediction algorithm corresponding with feature tag that is this phylogenetic group to carry out electricity sales amount prediction to the region to be predicted of each in this phylogenetic group.

7. device according to claim 6, is characterized in that, described clustering processing unit comprises:

First clustering processing subelement, for adopting vision clustering algorithm, according to the feature of history electricity sales amount curve under time domain and frequency domain in each region to be predicted, carries out cluster to each region to be predicted.

8. device according to claim 6, is characterized in that, described characteristics determining unit comprises:

Temporal signatures determining unit, for determining that the time domain of the history electricity sales amount curve in each region to be predicted under time domain is with making a gesture of measuring and time domain chain rate amount;

Frequency domain character determining unit, for determining the frequency domain extreme difference point amplitude of the history electricity sales amount curve in each region to be predicted under frequency domain and Fei Shi index;

With making a gesture of measuring, chronicity determining unit, determines whether the history electricity sales amount curve in corresponding region to be predicted possesses chronicity for utilizing time domain;

Stationarity determining unit, determines for utilizing time domain chain rate amount whether the history electricity sales amount curve in corresponding region to be predicted possesses stationarity;

Perturbation determining unit, determines for utilizing frequency domain extreme difference point amplitude whether the history electricity sales amount curve of corresponding estimation range possesses perturbation;

Periodically determining unit, determines for utilizing Fei Shi index whether the history electricity sales amount curve of corresponding estimation range possesses periodically.

9. device according to claim 8, is characterized in that, the feature tag of each phylogenetic group that cluster is formed comprises:

Fisrt feature label: possess chronicity, stationarity and periodicity simultaneously, do not possess perturbation;

Second feature label: possess chronicity, periodicity and perturbation simultaneously, do not possess stationarity;

Third feature label: possess stationarity, periodicity and perturbation simultaneously, do not possess chronicity;

Fourth feature label: simultaneously possess chronicity and perturbation, does not possess stationarity and periodicity;

Fifth feature label: possess stationarity and periodicity simultaneously, do not possess chronicity and perturbation.

10. device according to claim 9, is characterized in that, the preset prediction algorithm corresponding with each feature tag comprises:

The BP neural network algorithm corresponding with described fisrt feature label;

The L corresponding with described second feature label _1/2openness regression algorithm;

The support vector machines regression algorithm corresponding with described third feature label;

The L corresponding with described fourth feature label _1/2openness regression algorithm;

The difference ARMA model ARIMA algorithm corresponding with described fifth feature label.