CN105741143A - Load characteristic and cluster analysis based electric power commodity pricing model establishment method - Google Patents

Abstract

The invention discloses a load characteristic and cluster analysis based electric power commodity pricing model establishment method. Compared with the prior art, the method overcomes the defect that load characteristics of users cannot be fully reflected in an electric power commodity pricing method. The method comprises the following steps of acquiring and preprocessing original data, obtaining basic load data, and preprocessing the basic load data; clustering the preprocessed load data to obtain a basic load curve; and extracting load characteristic indexes in the basic load curve and constructing an electric power commodity pricing model. The invention provides a massive user load data analysis based modeling method; and through the modeling method, electric power commodities are accurately priced.

Description

A kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof

Technical field

The present invention relates to data analysis technique field, specifically a kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof.

Background technology

At present that the analyzing and processing of Power system load data is very general, powerful data processing function by the application of statistical method and computer, load data is analyzed, processes and matching, to solve manually to calculate process, its reparation being widely used in load data and prediction, in the field such as modeling of load curve.And the relation between the load data of user and electricity rates is considered fully, the price not over electricity commodity that affects of user power utilization cost is fully demonstrated out by part throttle characteristics.In open electricity market, the electricity rates of user and its part throttle characteristics are closely related, how to utilize the power consumer load data of magnanimity, the price of rational electricity commodity, both cost of compensation realized income, realize again the appropriate guiding to user power utilization, have become as urgent need and solve the technical problem that.

Summary of the invention

The invention aims to the defect solving cannot fully to reflect customer charge characteristic in prior art in electricity commodity pricing method, it is provided that a kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof solves the problems referred to above.

To achieve these goals, technical scheme is as follows:

A kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof, comprises the following steps:

Raw data acquisition and pretreatment, obtain basic load data, basic load data carried out pretreatment；

Pretreated load data is clustered, obtains base load curve；

Extract the Load characteristics index in base load curve, construct the pricing model of electricity commodity.

Described raw data acquisition and pretreatment comprise the following steps:

Gathering basic load data, individual consumer is expressed as n, n ∈ N, wherein N represents that several users gather；

Set 96 time collection point i, i ∈ 1,2 ..., 96}；

User n is expressed as at the load data of i-th collection pointAnd y_i ⁿ≥0。

Described pretreated load data carried out cluster comprise the following steps:

Being stored in input matrix by N number of input vector, build the matrix Σ of N × 96, wherein N represents several numbers of users, 96 collection points in 96 expressions one day；

To weights ω_ijInitializing, its span is [0,1] and ω_ijAll differ；

Obtain weight vector collection G, selected vector in N number of input vector from matrix ΣIt is supplied to network input layer, calculates weight vector ω_ijT matrix Σ is also iterated processing until N number of input vector is all selected has processed by (), it is thus achieved that weight vector collection G；It specifically includes following steps:

By vector selected in matrix ΣWith initial weight ω_ijGive network input layer；

The all neurons of computing network output layer are to the distance d of input vector_j, its computing formula is as follows:

$d_j=\underset{i=1}{\overset{h}{\mathrm{\Σ}}}{\left(y_i^n-{\mathrm{\ω}}_{ij}\left(t\right)\right)}^2,$

Wherein t is current update times；

Select competition triumph neuronNeuron i is triumph neuron；

For all neurons, adjusting it and the neuronic weight vector in neighborhood, more new formula is as follows for it:

${\mathrm{\ω}}_{\mathrm{ij}}(t+1)={\mathrm{\ω}}_{\mathrm{ij}}\left(t\right)+\mathrm{\η}\left(t\right)h_{j,i\left(x\right)}\left(t\right)(y_i^n\left(t\right)-{\mathrm{\ω}}_{\mathrm{ij}}\left(t\right)),$

Wherein, η (t) represents learning rate (0 < η (t) < 1) and monotone decreasing, h_j,i(x)T () represents neuronic neighborhood function of winning；

It is iterated processing to the weights after updating and newly selected vector, proceeds the computing network output layer all neurons distance d to input vector_jWith weights ω_ijRenewal, processed until N number of input vector is all selected；

Obtain weight vector collection G{ ω_ij| j=(1,2 ..., m) }, wherein m represents the number of cluster；

By single weight vector ω_ijT (), as with its cluster centre for neuronic all samples of winning, sample is divided into m class, be expressed as G1, G2 ..., Gm}；The daily load curve of cluster centre is this type of base load curve.

The described Load characteristics index extracted in base load curve comprises the following steps:

Calculate the rate of load condensate LF of all types of user daily load curve^G, its computing formula is as follows:

Wherein:Represent the collection point data of this type of base load curve, LF^GRepresenting this classification rate of load condensate, T represents the statistics phase,For the meansigma methods of load,Maximum for load；

Calculate the basic electricity price p of each time period_i；

Build the pricing model based on part throttle characteristics.

Also include repairing the missing values in basic load data；The interim heart rolling average method of k is adopted to carry out the repairing of missing values；IfFor missing data y_iRepairing value, computing formula is as follows:

$y_i^{*}=\frac{1}{k-1}(y_{i-\frac{k-1}{2}}+...y_{i-1}+y_{i+1}+...+y_{i+\frac{k-1}{2}}),$ K is odd number；

$y_i^{*}=\frac{1}{k-1}(\frac{1}{2}{y}_{i-\frac{k}{2}}+y_{i-\frac{k}{2}+1}+{...y}_{i-1}+y_{i+1}+...+y_{i+\frac{k}{2}-1}+\frac{1}{2}{y}_{i+\frac{k}{2}}),$ K is even number；

Wherein, k represents for carrying out the number of samples repaired.

The described basic electricity price p calculating each time period_iComprise the following steps:

The horizontal p of basic average electricity price is determined by cost and expected revenus_ave, it is as follows with the relational expression of tou power price:

Wherein: p_iRepresent each power price gathering the period, y_iAll customer charge sums during for system loading data and i collection point, h is collection point number；

If each power price gathering the period and load data are linear, then

p_i=ay_i+ b, i ∈ 1,2 ..., h}, wherein p_iRepresent each power price gathering the period, y_iRepresenting system loading data, a, b are the parameter of linear equation；

Calculating the ratio delta of time-of-use tariffs, its computing formula is as follows:

$\mathrm{\&delta;}=\frac{p_{\max }}{p_{\min }}=\frac{{\mathrm{ay}}_{\max }+b}{{\mathrm{ay}}_{\min }+b};$

Calculating the parameter a of linear equation, b, its computing formula is as follows:

$a=\frac{p_{ave}{\mathrm{\&Sigma;}}_{i=1}^h{y}_i}{{\mathrm{\&Sigma;}}_{i=1}^h{y}_i^2+\frac{y_{\max }-{\mathrm{\&delta;y}}_{\min }}{\mathrm{\&delta;}-1}{\mathrm{\&Sigma;}}_{i=1}^h{y}_i};$

$b=\frac{y_{\max }-{\mathrm{\&delta;y}}_{\min }}{\mathrm{\&delta;}-1}a;$

Parameter a, b are substituted into p again_iCalculating, by p_iAs each basic electricity price gathering the period.

The described pricing model based on part throttle characteristics that builds comprises the following steps:

Calculate the additional electricity price based on rate of load condensateIts computing formula is as follows:

${\mathrm{\&Delta;p}}_1^G=\&lsqb;c_2-\frac{{\mathrm{LF}}^G-{\mathrm{LF}}^{\min }}{{\mathrm{LF}}^{\max }-{\mathrm{LF}}^{\min }}(c_1+c_2)\&rsqb;*p_{ave},$

Wherein: G represents the classification of user, LF^maxRepresent rate of load condensate maximum in all categories, LF^minRepresent rate of load condensate minimum in all categories, c₁Expression rate of load condensate is LF^maxTime electricity price reward relatively basic electricity price p_aveRatio, c₂Expression rate of load condensate is LF^minTime electricity price punish relatively basic electricity price p_aveRatio；

Standardization load data, to each collection point data all divided by the average load on the same day, its computing formula is as follows:

Wherein y_aveIt it is the meansigma methods of 96 collection point data；

Calculate the additional electricity price of simultaneity factor of single load peak periodIts computing formula is as follows:

${\mathrm{\&Delta;p}}_{2i}^G=\mathrm{\&mu;}max\{min\&lsqb;\left(z_i^G-z_i{d}_1\right)\&rsqb;,d_2\}*p_i$

Wherein:Represent the standardized acquisition point data of certain class users, z_iThe standardized acquisition point data of expression system, d₁, d₂For limit coefficient, d₁=z_max-1, d₂=-2 (z_max-1), z_maxRepresenting the maximum in the data of system standardization collection point, μ is coefficient of deviation；

Calculate the additional electricity price of simultaneity factor of single load valley periodIts computing formula is as follows:

${\mathrm{\&Delta;p}}_{3i}^G=\mathrm{\&mu;}max\{\min \&lsqb;\left(z_i-z_i^G,d_3\right)\&rsqb;,d_4\}*p_i$

Wherein d₃, d₄It is similarly limit coefficient, d₃=1-z_min, d₄=-2 (1-z_min), z_minRepresenting the minima in the data of system standardization collection point, μ is coefficient of deviation；

Calculate all kinds of load electricity price P^G, its computing formula is as follows:

$p^G=\frac{{\mathrm{\&Sigma;}}_{i=1}^h{z}_i^G(p_i+{\mathrm{\&Delta;p}}_{2i}^G*\mathrm{\&Phi;}(i)+{\mathrm{\&Delta;p}}_{3i}^G*\mathrm{\&theta;}(i\left)\right)}{{\mathrm{\&Sigma;}}_{i=1}^h{z}_i^G}+{\mathrm{\&Delta;p}}_1^G,$

Φ (i)=0 when φ (i)=1, i is low-valley interval when wherein i is peak period；θ (i)=1 when θ (i)=0, i is low-valley interval when i is peak period.

Beneficial effect

A kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof of the present invention, compared with prior art provides the modeling method of analysis based on mass users load data, is achieved the accurate price of electricity commodity by its modeling method.By extract base load curve Load characteristics index, realize fully demonstrating the structure of the Power Pricing system of customer charge characteristic, considering on the basis of economic benefit, guide the reasonable scientific utilization of electricity of user simultaneously, can more reasonably embody the cost of user power utilization, it is achieved high efficiency demand management.

Accompanying drawing explanation

Fig. 1 is the method flow diagram of the present invention.

Detailed description of the invention

For making the architectural feature to the present invention and effect of reaching have a better understanding and awareness, coordinate detailed description in order to preferred embodiment and accompanying drawing, illustrate as follows:

As it is shown in figure 1, a kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof of the present invention, comprise the following steps:

The first step, raw data acquisition and pretreatment.From user power utilization information acquisition system, obtain basic load data, basic load data are carried out pretreatment, provide data support for structure electricity commodity pricing model.Data prediction be all by data analysis obtain information thus setting up prerequisite and the basis of the operating process of model, ensureing on the basis that the initial data that adopts is true and reliable and representative, by suitable statistical analysis and information retrieval, the electricity commodity pricing model that can appropriately reflect customer charge characteristic just can be finally given.It specifically includes following steps:

(1) gathering basic load data, individual consumer is expressed as n, n ∈ N, wherein N represents that several users gather.It is considered herein that user face and type are relatively wide, for increasing versatility so that it is can for multiple users, it is therefore desirable to cluster, mark off several classification, then to every class unbundling, form pricing model.During use, different classes of parameter is different, then can produce different electricity prices after input pricing model.

(2) set 96 time collection point i, i ∈ 1,2 ..., 96}.

(3) user n is expressed as at the load data of i-th collection pointAnd

Data above can be obtained information by associated mechanisms, for instance the power consumer load data of certain province can obtain in the user power utilization information acquisition system of the thus grid company measurement centre in province.Between the daily load curve of user, similarity is significantly high, so adopt daily load curve to be analyzed representative at this, it is possible to the part throttle characteristics of reflection user, is easier to the electricity commodity pricing model illustrating thus to build in following steps.And consider from technical standpoint, in this step, it would however also be possible to employ other data.

Although the acquisition rate of user power utilization information acquisition system is significantly high at present, but individual data is really always inevitable.It is considered herein that the disappearance to individual data can destroy data structure so that ensuing data analysis is difficult to, the method therefore also providing at this repairing the missing values in basic load data, adopt the interim heart rolling average method of k to carry out the repairing of missing values.

IfFor missing data y_iRepairing value, computing formula is as follows:

$y_i^{*}=\frac{1}{k-1}(y_{i-\frac{k-1}{2}}+...y_{i-1}+y_{i+1}+...+y_{i+\frac{k-1}{2}}),$ K is odd number；

$y_i^{*}=\frac{1}{k-1}(\frac{1}{2}{y}_{i-\frac{k}{2}}+y_{i-\frac{k}{2}+1}+{...y}_{i-1}+y_{i+1}+...+y_{i+\frac{k}{2}-1}+\frac{1}{2}{y}_{i+\frac{k}{2}}),$ K is even number；

Wherein, k represents for carrying out the number of samples repaired, and it can choose the collection point sum minimum approximate number except 1,2.

Obtain the user's daily load curve for analyzing after repairing, be namely expressed as a time series:

$\left\{\begin{array}{cccc}{y}_1^n,& y_2^n,& ...,& y_{96}^n\end{array}\right\}.$

Second step, clusters pretreated load data, obtains base load curve.It specifically comprises the following steps that

(1) being stored in input matrix by N number of input vector, build the matrix Σ of N × 96, wherein N represents several numbers of users, 96 collection points in 96 expressions one day.At this, time series of multiple user's daily load curves is fused into a matrix, in order to follow-up cluster analysis.

(2) to weights ω_ijInitializing, its span is [0,1] and ω_ijAll differ.

(3) weight vector collection G is obtained.Random selected vector in N number of input vector from matrix ΣIt is supplied to network input layer, calculates weight vector ω_ijT matrix Σ is also iterated processing until N number of input vector is all selected has processed by (), it is thus achieved that weight vector collection G.It specifically comprises the following steps that

A, will selected vector in matrix ΣWith initial weight ω_ijGive network input layer.

B, all neurons of computing network output layer are to the distance d of input vector_j, its computing formula is as follows:

$d_j=\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{\left(y_i^n-{\mathrm{\&omega;}}_{ij}\left(t\right)\right)}^2,$

Wherein t is current update times；

According to the principle that the victor is a king, select competition triumph neuron i (x)=mind_j, neuron i is triumph neuron.

C, for all neurons, adjust it and the neuronic weight vector in neighborhood, more new formula is as follows for it:

${\mathrm{\&omega;}}_{\mathrm{ij}}(t+1)={\mathrm{\&omega;}}_{\mathrm{ij}}\left(t\right)+\mathrm{\&eta;}\left(t\right)h_{j,i\left(x\right)}\left(t\right)(y_i^n\left(t\right)-{\mathrm{\&omega;}}_{\mathrm{ij}}\left(t\right)),$

Wherein, η (t) represents learning rate (0 < η (t) < 1) and monotone decreasing, h_j,i(x)T () represents neuronic neighborhood function of winning, it is possible to adopt GAUSS field function.

D, it is iterated processing to weights and the newly selected vector after updating, proceeds the network output layer all neurons distance d to input vector_jCalculating and weights ω_ijRenewal, processed until N number of input vector is all selected.

E, acquisition weight vector collection G{ ω_ij| j=(1,2 ..., m) }, wherein m represents the number of cluster.From training process, the weight vector of output neuron is close to triumph neuron gradually.Then weight vector collection G{ ω_ij| j=(1,2 ..., m) } it is the description that training sample is concentrated all samples, and single weight vector is considered as with its cluster centre for the neuronic all samples of triumph, using the cluster centre base load curve as this type of.

(4) by single weight vector ω_ijT (), as with its cluster centre for neuronic all samples of winning, sample is divided into m class, be expressed as G1, G2 ..., Gm}；The daily load curve of cluster centre is this type of base load curve.

3rd step, extracts the Load characteristics index in base load curve, constructs the pricing model of electricity commodity.It specifically includes following steps:

(1) the rate of load condensate LF of all types of user daily load curve is calculated^G.The pricing model of this method is based on the part throttle characteristics of user, it is considered to the electricity consumption behavior of the variant feature impact on electricity commodity prices.The content that part throttle characteristics includes is very abundant, and 1989 Nian Yuan Ministry of Energy propose 14 indexs describing part throttle characteristics including load curve, rate of load condensate in " power industry production leadtime index explanation ".At the beginning of former State Power Corporation calendar year 2001, above-mentioned explanation having been carried out additional modifications, added peak-valley ratio index, these indexs are conventional and specification Load characteristics indexes.Wherein rate of load condensate is most important index.In " Power Pricing policy " that the World Bank publishes, the price of electricity commodity is using rate of load condensate as sole criterion.So first calculating the rate of load condensate of user's daily load curve.The definition that rate of load condensate (LoadFactor) is general refers to the ratio of average load at the appointed time and peak load, wherein, maximum load is the peak load of interior record during referring to statistics, average load refers to the meansigma methods of instantaneous load in the statistics phase, i.e. duration of load application ordered series of numbers sequential average, its computing formula is as follows:

Wherein:Represent the collection point data of this type of base load curve, LF^GRepresenting this classification rate of load condensate, T represents the statistics phase,For the meansigma methods of load,Maximum for load.

(2) the basic electricity price p of each time period is calculated_i, it specifically comprises the following steps that

A, determined the horizontal p of basic average electricity price by cost and expected revenus_ave, it is as follows with the relational expression of tou power price:

Wherein: p_iRepresent each power price gathering the period, y_iAll customer charge sums during for system loading data and i collection point, h is collection point number.

B, based on determining this electricity price level with factors such as rational power purchase price, distribution price, sale of electricity cost and operating profits for starting point, and by p_aveCoupling system part throttle characteristics determines p_i.If each power price gathering the period and load data are linear, then

p_i=ay_i+ b, i ∈ 1,2 ..., h}, wherein p_iRepresent each power price gathering the period, y_iRepresenting system loading data, a, b are the parameter of linear equation.

C, calculate time-of-use tariffs ratio delta, its computing formula is as follows:

$\mathrm{\&delta;}=\frac{p_{\max }}{p_{\min }}=\frac{{\mathrm{ay}}_{\max }+b}{{\mathrm{ay}}_{\min }+b}.$

D, calculating the parameter a of linear equation, b, its computing formula is as follows:

$a=\frac{p_{ave}{\mathrm{\&Sigma;}}_{i=1}^h{y}_i}{{\mathrm{\&Sigma;}}_{i=1}^h{y}_i^2+\frac{y_{\max }-{\mathrm{\&delta;y}}_{\min }}{\mathrm{\&delta;}-1}{\mathrm{\&Sigma;}}_{i=1}^h{y}_i};$

$b=\frac{y_{\max }-{\mathrm{\&delta;y}}_{\min }}{\mathrm{\&delta;}-1}a.$

E, parameter a, b are substituted into p again_iCalculating, by p_iAs each basic electricity price gathering the period.

(3) build the pricing model based on part throttle characteristics, namely draw each group of load electricity price p^GComputing formula, then complete the foundation of pricing model.It specifically includes following steps:

A, calculate based on the additional electricity price of rate of load condensateIts computing formula is as follows:

${\mathrm{\&Delta;p}}_1^G=\&lsqb;c_2-\frac{{\mathrm{LF}}^G-{\mathrm{LF}}^{\min }}{{\mathrm{LF}}^{\max }-{\mathrm{LF}}^{\min }}(c_1+c_2)\&rsqb;*p_{ave},$

Wherein: G represents the classification of user, LF^maxRepresent rate of load condensate maximum in all categories, LF^minRepresent rate of load condensate minimum in all categories, c₁Expression rate of load condensate is LF^maxTime electricity price reward relatively basic electricity price p_aveRatio, c₂Expression rate of load condensate is LF^minTime electricity price punish relatively basic electricity price p_aveRatio.

B, standardization load data, to each collection point data all divided by the average load on the same day.At this mainly for eliminating the different dimension impact on data, typically by collection point data divided by maximum, being processed as conveniently distinguishing peak valley, namely standardized data is the peak period more than 1, is the paddy period less than 1, it is considered to the additional electricity price of simultaneity factor.Its computing formula is as follows:

Wherein y_aveIt it is the meansigma methods of 96 collection point data.

C, calculate the additional electricity price of simultaneity factor of single load peak periodIts computing formula is as follows:

${\mathrm{\&Delta;p}}_{2i}^G=\mathrm{\&mu;}max\{min\&lsqb;\left(z_i^G-z_i{d}_1\right)\&rsqb;,d_2\}*p_i$

Wherein:Represent the standardized acquisition point data of certain class users, z_iThe standardized acquisition point data of expression system, d₁, d₂For limit coefficient, d₁=z_max-1, d₂=-2 (z_max-1), z_maxRepresenting the maximum in the data of system standardization collection point, μ is coefficient of deviation.

D, calculate the additional electricity price of simultaneity factor of single load valley periodIts computing formula is as follows:

${\mathrm{\&Delta;p}}_{3i}^G=\mathrm{\&mu;}max\{\min \&lsqb;\left(z_i-z_i^G,d_3\right)\&rsqb;,d_4\}*p_i$

Wherein d₃, d₄It is similarly limit coefficient, d₃=1-z_min, d₄=-2 (1-z_min), z_minRepresenting the minima in the data of system standardization collection point, μ is coefficient of deviation.

E, calculate all kinds of load electricity price P^G, its computing formula is as follows:

$p^G=\frac{{\mathrm{\&Sigma;}}_{i=1}^h{z}_i^G(p_i+{\mathrm{\&Delta;p}}_{2i}^G*\mathrm{\&Phi;}(i)+{\mathrm{\&Delta;p}}_{3i}^G*\mathrm{\&theta;}(i\left)\right)}{{\mathrm{\&Sigma;}}_{i=1}^h{z}_i^G}+{\mathrm{\&Delta;p}}_1^G,$

Φ (i)=0 when Φ (i)=1, i is low-valley interval when wherein i is peak period；θ (i)=1 when θ (i)=0, i is low-valley interval when i is peak period.

The ultimate principle of the present invention, principal character and advantages of the present invention have more than been shown and described.Skilled person will appreciate that of the industry; the present invention is not restricted to the described embodiments; simply principles of the invention described in above-described embodiment and description; the present invention also has various changes and modifications without departing from the spirit and scope of the present invention, and these changes and improvements both fall within claimed the scope of the present invention.The protection domain of application claims is defined by appending claims and equivalent thereof.

Claims (7)

1. the electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof, it is characterised in that comprise the following steps:

11) raw data acquisition and pretreatment, obtains basic load data, basic load data is carried out pretreatment；

12) pretreated load data is clustered, obtain base load curve；

13) extract the Load characteristics index in base load curve, construct the pricing model of electricity commodity.

2. a kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof according to claim 1, it is characterised in that described raw data acquisition and pretreatment comprise the following steps:

21) gathering basic load data, individual consumer is expressed as n, n ∈ N, wherein N represents that several users gather；

22) set 96 time collection point i, i ∈ 1,2 ..., 96}；

23) user n is expressed as at the load data of i-th collection pointAnd

3. a kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof according to claim 1, it is characterised in that described pretreated load data is carried out cluster comprise the following steps:

31) being stored in input matrix by N number of input vector, build the matrix Σ of N × 96, wherein N represents several numbers of users, 96 collection points in 96 expressions one day；

32) to weights ω_ijInitializing, its span is [0,1] and ω_ijAll differ；

33) weight vector collection G is obtained, selected vector in N number of input vector from matrix ΣIt is supplied to network input layer, calculates weight vector ω_ijT matrix Σ is also iterated processing until N number of input vector is all selected has processed by (), it is thus achieved that weight vector collection G；It specifically includes following steps:

331) by vector selected in matrix ΣWith initial weight ω_ijGive network input layer；

332) all neurons of computing network output layer are to the distance d of input vector_j, its computing formula is as follows:

$d_j=\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{(y_i^n-{\mathrm{\&omega;}}_{ij}(t\left)\right)}^2,$

Wherein t is current update times；

Select competition triumph neuron i (x)=mind_j, neuron i is triumph neuron；

333) for all neurons, adjusting it and the neuronic weight vector in neighborhood, more new formula is as follows for it:

${\mathrm{\&omega;}}_{\mathrm{ij}}(t+1)={\mathrm{\&omega;}}_{\mathrm{ij}}\left(t\right)+\mathrm{\&eta;}\left(t\right)h_{j,i\left(x\right)}\left(t\right)(y_i^n\left(t\right)-{\mathrm{\&omega;}}_{\mathrm{ij}}\left(t\right)),$

Wherein, η (t) represents learning rate (0 < η (t) < 1) and monotone decreasing, h_{J, i (x)}T () represents neuronic neighborhood function of winning；

334) it is iterated processing to the weights after updating and newly selected vector, proceeds the computing network output layer all neurons distance d to input vector_jWith weights ω_ijRenewal, processed until N number of input vector is all selected；

335) weight vector collection G{ ω is obtained_ij| j=(1,2 ..., m) }, wherein m represents the number of cluster；

34) by single weight vector ω_ijT (), as with its cluster centre for neuronic all samples of winning, sample is divided into m class, be expressed as G1, G2 ..., Gm}；The daily load curve of cluster centre is this type of base load curve.

4. a kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof according to claim 1, it is characterised in that the described Load characteristics index extracted in base load curve comprises the following steps:

41) the rate of load condensate LF of all types of user daily load curve is calculated^G, its computing formula is as follows:

Wherein:Represent the collection point data of this type of base load curve, LF^GRepresenting this classification rate of load condensate, T represents the statistics phase,For the meansigma methods of load,Maximum for load；

42) the basic electricity price p of each time period is calculated_i；

43) pricing model based on part throttle characteristics is built.

5. a kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof according to claim 2, it is characterised in that: also include repairing the missing values in basic load data；The interim heart rolling average method of k is adopted to carry out the repairing of missing values；IfFor missing data y_iRepairing value, computing formula is as follows:

$y_i^{*}=\frac{1}{k-1}(y_{i-\frac{k-1}{2}}+...y_{i-1}+y_{i+1}+...+y_{i+\frac{k-1}{2}}),$ K is odd number；

$y_i^{*}=\frac{1}{k-1}(\frac{1}{2}{y}_{i-\frac{k}{2}}+y_{i-\frac{k}{2}+1}+...y_{i-1}+y_{i+1}+...+y_{i+\frac{k}{2}-1}+\frac{1}{2}{y}_{i+\frac{k}{2}}),$ K is even number；

Wherein, k represents for carrying out the number of samples repaired.

6. a kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof according to claim 4, it is characterised in that the described basic electricity price p calculating each time period_iComprise the following steps:

61) the horizontal p of basic average electricity price is determined by cost and expected revenus_ave, it is as follows with the relational expression of tou power price:

Wherein: p_iRepresent each power price gathering the period, y_iAll customer charge sums during for system loading data and i collection point, h is collection point number；

62) set each power price gathering the period and load data is linear, then

p_i=ay_i+ b, i ∈ 1,2 ..., h}, wherein p_iRepresent each power price gathering the period, y_iRepresenting system loading data, a, b are the parameter of linear equation；

63) calculating the ratio delta of time-of-use tariffs, its computing formula is as follows:

$\mathrm{\&delta;}=\frac{p_{\max }}{p_{\min }}=\frac{{\mathrm{ay}}_{\max }+b}{{\mathrm{ay}}_{\min }+b};$

64) calculating the parameter a of linear equation, b, its computing formula is as follows:

$a=\frac{p_{ave}{\mathrm{\&Sigma;}}_{i=1}^h{y}_i}{{\mathrm{\&Sigma;}}_{i=1}^h{y}_i^2+\frac{y_{\max }-{\mathrm{\&delta;y}}_{\min }}{\mathrm{\&delta;}-1}{\mathrm{\&Sigma;}}_{i=1}^h{y}_i};$

$b=\frac{y_{\max }-{\mathrm{\&delta;y}}_{\min }}{\mathrm{\&delta;}-1}a;$

65) parameter a, b are substituted into p again_iCalculating, by p_iAs each basic electricity price gathering the period.

7. a kind of electricity commodity pricing model method for building up based on part throttle characteristics and cluster analysis thereof according to claim 4, it is characterised in that the described pricing model based on part throttle characteristics that builds comprises the following steps:

71) the additional electricity price based on rate of load condensate is calculatedIts computing formula is as follows:

${\mathrm{\&Delta;p}}_1^G=\&lsqb;c_2-\frac{{\mathrm{LF}}^G-{\mathrm{LF}}^{\min }}{{\mathrm{LF}}^{\max }-{\mathrm{LF}}^{\min }}(c_1+c_2)\&rsqb;*p_{ave},$

Wherein: G represents the classification of user, LF^maxRepresent rate of load condensate maximum in all categories, LF^minRepresent rate of load condensate minimum in all categories, c₁Expression rate of load condensate is LF^maxTime electricity price reward relatively basic electricity price p_aveRatio, c₂Expression rate of load condensate is LF^minTime electricity price punish relatively basic electricity price p_aveRatio；

72) standardization load data, to each collection point data all divided by the average load on the same day, its computing formula is as follows:

Wherein y_aveIt it is the meansigma methods of 96 collection point data；

73) the additional electricity price of simultaneity factor of single load peak period is calculatedIts computing formula is as follows:

${\mathrm{\&Delta;p}}_{2i}^G=\mathrm{\&mu;}\max \{\min \&lsqb;(z_i^G-z_i,d_1)\&rsqb;,d_2\}*p_i$

Wherein:Represent the standardized acquisition point data of certain class users, z_iThe standardized acquisition point data of expression system, d₁, d₂For limit coefficient, d₁=z_max-1, d₂=-2 (z_max-1), z_maxRepresenting the maximum in the data of system standardization collection point, μ is coefficient of deviation；

74) the additional electricity price of simultaneity factor of single load valley period is calculatedIts computing formula is as follows:

${\mathrm{\&Delta;p}}_{3i}^G=\mathrm{\&mu;}max\{\min \&lsqb;(z_i-z_i^G,d_3)\&rsqb;,d_4\}*p_i$

Wherein d₃, d₄It is similarly limit coefficient, d₃=1-z_min, d₄=-2 (1-z_min), z_minRepresenting the minima in the data of system standardization collection point, μ is coefficient of deviation；

75) all kinds of load electricity price p is calculated^G, its computing formula is as follows:

$p^G=\frac{{\mathrm{\&Sigma;}}_{i=1}^h{z}_i^G(p_i+{\mathrm{\&Delta;p}}_{2i}^G*\mathrm{\&Phi;}(i)+{\mathrm{\&Delta;p}}_{3i}^G*\mathrm{\&theta;}(i\left)\right)}{{\mathrm{\&Sigma;}}_{i=1}^h{z}_i^G}+{\mathrm{\&Delta;p}}_1^G,$

Φ (i)=0 when Φ (i)=1, i is low-valley interval when wherein i is peak period；θ (i)=1 when θ (i)=0, i is low-valley interval when i is peak period.