CN105260944A - Method for calculating statistical line loss based on LSSVM (Least Square Support Vector Machine) algorithm and association rule mining - Google Patents

Abstract

The invention relates to a method for calculating the statistical line loss based on an LSSVM (Least Square Support Vector Machine) algorithm and association rule mining, which comprises the steps of processing and simplifying a meter reading mode of a low-voltage transformer area; calculating the power sale quantity of an interval, in which the power supply quantity and the power sale quantity are different in period, of adjacent months; correcting a power sale quantity calculation result by using the economic development level; adding consideration for occurrence of random events, and correcting the calculation result again; calculating the power sale quantity through the above steps, and calculating the statistical line loss of the month by using a formula; and carrying out rationality verification on the acquired power sale quantity and the acquired statistical line loss respectively. The method provided by the invention gives sufficient consideration to meter reading modes and meter reading periods of various types of loads, solves a problem that the power supply quantity and the power sale quantity are different in period, corrects wrong statistical line loss, enables a calculation result of the statistical line loss to be closer to an actual condition, is high in rationality and conducive to improving the line loss management level, and provides a reliable basis for economical operations of a power system.

Description

A kind of statistical line losses computing method based on LSSVM algorithm and association rule mining

Technical field

The present invention relates to a kind of electric system statistical line losses computing method, especially a kind of statistical line losses computing method based on LSSVM algorithm and association rule mining.

Background technology

Line loss is the important means of examination power grid enterprises manager pay level, directly can reflect the height of a level such as Study on Power Grid Planning, operation management.The line loss electricity produced in equipment in the given period of being calculated by gate energy meter reader and each link of marketing, is called statistical line losses.

Boundary of administration loss rate refers in the process of operation of power networks and marketing management, the percentage of the kwh loss caused due to administrative reason and statistics delivery; Also be the important evidence formulated wastage reducing and energy saving plan, determine score loss rate index.Obtain boundary of administration loss rate accurately, must under statistical line losses calculates rational prerequisite.

The statistical system of current China utilizes supply and distribution surplus method to calculate line loss per unit.Due in electricity sales amount, dissimilar user checks meter time difference, must cause like this in the electricity sales amount calculating this month, some electricity was user's upper month, also some this month voltameter calculation is in lower monthly power demand, because may there be the appearance of random occurrence the change of temperature between the moon and not corresponding period, the line loss per unit that this method is obtained does not conform to actual pole, even occur that electricity sales amount is greater than the situation of delivery, line loss Changing Pattern can not be reflected strictly according to the facts, have impact on line loss statistical accuracy and rationality, thus affect the work of electric system Controlling line loss aspect.

In sum, be necessary to invent a kind of brand-new statistical line losses computing method, to solve the problems that existing statistical line losses work exists.

Summary of the invention

The object of the invention is to provide that a kind of accuracy is high, Consideration is comprehensive, practicality is high based on the statistical line losses computing method of LSSVM algorithm and association rule mining.

For achieving the above object, the step of computing method of the present invention is as follows:

(1) carry out processing to the meter reading method of low-voltage platform area mesolow user and simplify, determining meter reading method and the cycle of power load;

(2) calculate adjacent month supply and distribution not the same period interval electricity sales amount;

(3) local economic development horizontal corrected Calculation electricity sales amount result is utilized;

(4) consideration that random occurrence is occurred is added, corrected Calculation result again;

(5) calculate electricity sales amount, utilize formulae discovery this month statistical line losses;

(6) electricity sales amount obtained step (3) and step (5) and statistical line losses electricity carry out mark soundness verification.

Further, the detailed process of described step (1) is as follows:

(1-1) determine that equivalence is checked meter example day

Equivalence check meter client electricity that example refers to all residents day all equivalence check meter in this sky, equivalence example of checking meter is determined by following formula day,

Wherein e _ibe electricity of checking meter on the i-thth, n is number of days of monthly checking meter, E _jfor every monthly power demand, be the proportion that electricity of checking meter on the i-thth accounts for when monthly power demand;

(1-2) determine that equivalence is checked meter interval

In conjunction with the data fusion method of optimal weighting, the meter reading data equivalence of every day is considered as a sensor, check meter example day result of calculation for reference value with equivalence, obtaining the weight of meter reading data every day, select relative variance reckling, is optimal estimation;

Weight asks for formula: $\mathrm{\ω}=\frac{1}{{\mathrm{\σ}}_i^2}/\underset{i=1}{\overset{n}{\Σ}}\frac{1}{{\mathrm{\σ}}_i^2}$

Wherein, σ _ibe that meter reading data on the i-thth and equivalence are checked meter the difference of example day data, n for monthly checking meter number of days,

And then obtain optimal estimation:

$\hat{X}=\underset{i=1}{\overset{n}{\Σ}}\[\frac{1}{{\mathrm{\σ}}_i^2}/\underset{i=1}{\overset{n}{\Σ}}\frac{1}{{\mathrm{\σ}}_i^2}\]X_i$

Wherein, X _ibe meter reading data on the i-thth, for optimal estimation;

By what obtain carry out variance with each day electricity to calculate and contrast, be on duty for the dayly decided to be equivalence by minimum for variance and check meter interval from date, be namely equivalent to all low-voltage customers and be unified in this date and check meter, equivalence electricity of checking meter is check meter electricity sum in all low-voltage platform areas.

Further, the detailed process of described step (2) is as follows:

(2-1) sample data is determined

The daily load choosing somewhere continuous a period of time and corresponding day weather data, take weather data as sample training, weather data comprises the lowest temperature, the highest temperature, medial humidity, quantity of precipitation;

(2-2) electricity sales amount is calculated based on LSSVM

Overload Class is different, its cycle of checking meter accordingly is also different, Overload Class is divided into residential electricity consumption, commercial power, commercial power, farming power four class by this method, utilizes LSSVM algorithm to calculate each type load respectively and to check meter cycle and total electricity sales amount of calendar month not corresponding period, do difference to it;

For residential electricity consumption load, be analyzed as follows, other classification loads as commercial power, commercial power, farming power computation process in like manner.

Then corresponding actual monthly power demand is: $L_{T^{\′}}=L_{T_a}+(L_{t_a}-L_{t_a^{\′}}),$

In formula: for electricity sales amount of checking meter by this moon, for interval t _awith t' _aelectricity sales amount difference; t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month, t' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date, represent t _atotal electricity sales amount of segment, represent t' _atotal electricity sales amount of segment.

To difference carry out revising further and can obtain more accurate monthly power demand.

Further, the detailed process of described step (3) is as follows:

(3-1) choose monthly electricity sales amount and carry out Multiple Non Linear Regression with corresponding each monthly economic factor sequence, obtain the synthetic relationship model of electricity sales amount and economic factors;

(3-2) utilize electricity sales amount in fit equation to ask local derviation to each economic factor, obtain the susceptibility S that the lower electricity sales amount of different economic factor impact changes it respectively _i; What this susceptibility represented is monthly mean level, by t _awith t ' _abetween the difference of each economic factor, the economic factors calculated is to electricity sales amount difference modified value, that is: ${\mathrm{\ΔL}}_1=\underset{i=1}{\overset{n}{\Σ}}{S}_i\·\mathrm{\Δ}t;$

In formula: S _ibe the susceptibility that i-th economic factor is corresponding, Δ t is t _awith t' _athe difference of two interval average economic levels; t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month, t' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date;

(3-3) by period to be calculated and not corresponding period average economic factors mathematic interpolation electricity difference, revise, the monthly power demand to be calculated of economic factors can be considered

$L_{T^{\′}1}=L_{T_a}+(L_{t_a}-L_{t_a^{\′}})+{\mathrm{\ΔL}}_1;$

In formula: for electricity sales amount of checking meter by this moon; for interval t _awith t' _aelectricity sales amount difference; t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month, t' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date; represent t _atotal electricity sales amount of segment; represent t' _atotal electricity sales amount of segment; Δ L ₁represent by t _awith t' _athe difference of two interval average economic levels calculates the economic factors that obtains to electricity sales amount difference modified value; for the revised electricity sales amount difference of economic factors.

Further, the detailed process of described step (4) is as follows:

(4-1) historical data cluster

Collect historical data, by clustering method, after a class is divided on ground similar for the factor such as economy, weather, then for each class data mining, eliminate other aspect factors to the interference of electric quantity change;

(4-2) according to the impact of different random sexual factor on electricity sales amount, electricity sales amount amplitude of variation absolute value is classified; Enchancement factor arranged, in conjunction with actual production, living quantizes simultaneously;

(4-3) correlation rule of random factor and electricity sales amount amplitude of variation is excavated;

(4-4) level of economic development membership function is built;

(4-5) integrated economics development level membership function, quantizes the electricity sales amount amplitude of variation under the impact of each enchancement factor.

Further, the detailed process of described step (5) is as follows:

The monthly electricity sales amount of final reality is $L_{T^{\′}}=L_{T_a}+(L_{t_a}+L_1^{\′})(1+{\mathrm{\γ\ηt}}_{j1}^{\′})-(L_{t_a}+L_1^{\′\′})(1+{\mathrm{\γ\ηt}}_{j2}^{\′}),$

In formula: for this month actual electricity sales amount of checking meter; γ is of that month level of economic development degree of membership; T' _j1for a certain enchancement factor is at total hourage of this month not corresponding interval generation; T' _j2for a certain enchancement factor total hourage that not corresponding interval occurs in last month; η be in the unit time a certain enchancement factor on the impact of electricity sales amount amplitude of variation; L' ₁, L " ₁be respectively the electricity sales amount affected by average economic level of this month and last month; t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month, t' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date; T' _arepresent that each type load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date; represent t _atotal electricity sales amount of segment; represent t' _atotal electricity sales amount of segment;

After in like manner calculating the total actual electricity sales amount of all Overload Class, according to statistical line losses rate computing formula, calculate actual count line loss per unit.

Further, the detailed process of described step (6) is as follows:

(6-1) nondimensionalization process is carried out to long data when economy, meteorology, enchancement factor;

(6-2) model choosing is carried out to sample;

(6-3) mark post distance calculates, the rationality of verification computation result.

Compared with prior art, tool of the present invention has the following advantages:

1, calculate electricity sales amount based on LSSVM algorithm, Consideration is comprehensive, and accuracy is high;

2, solve the asynchronous problem of supply and distribution in statistical line losses from background, make statistical line losses more accurately rationally, further for management line loss analyzing lays the foundation;

3, be combined with membership function by association rule mining, quantize random occurrence to electricity sales amount variable effect, make result more precisely rationally, practicality is high.

Accompanying drawing explanation

Fig. 1 is that each type load of the inventive method is checked meter cycle schematic diagram.

Fig. 2 is the economic membership function schematic diagram of the inventive method.

Fig. 3 is enchancement factor classification situation schematic diagram in the inventive method.

Fig. 4 is the calculated population process flow diagram of the inventive method.

Embodiment

Below in conjunction with accompanying drawing, the present invention will be further described:

The inventive method does not propose in situation at consideration statistical line losses supply and distribution the same period.

Composition graphs 1 and Fig. 4, the concrete steps of computing method of the present invention are as follows:

(1) carry out processing to the meter reading method of low-voltage platform area mesolow user and simplify, determining meter reading method and the cycle of power load;

(1-1) determine that equivalence is checked meter example day

Because low-voltage customer quantity is many, distribution wide, example of checking meter almost is difficult to reach the unified stipulated time day.Therefore, in the methods of the invention, equivalence interval introducing of checking meter only carrys out quantification treatment for low-voltage customer.Equivalence check meter example refer to day all similar client electricity all equivalence check meter in this sky.The equivalence that first need calculate resident is checked meter example day, and equivalence example of checking meter is determined by following formula day:

Wherein e _ibe electricity of checking meter on the i-thth, n is number of days of monthly checking meter, E _jfor every monthly power demand, be the proportion that electricity of checking meter on the i-thth accounts for when monthly power demand;

(1-2) determine that equivalence is checked meter interval

It is common method in Fusion that optimal weighting data fusion is sent out.It is the weighted mean value asking each sensor output data, thus multiple data are equivalent to one group of data.

In conjunction with the data fusion method of optimal weighting, the meter reading data equivalence of every day is considered as a sensor, check meter example day result of calculation for reference value with equivalence, obtaining the weight of meter reading data every day, select relative variance reckling, is optimal estimation;

Weight asks for formula: $\mathrm{\ω}=\frac{1}{{\mathrm{\σ}}_i^2}/\underset{i=1}{\overset{n}{\Σ}}\frac{1}{{\mathrm{\σ}}_i^2}$

Wherein, σ _ibe that meter reading data on the i-thth and equivalence are checked meter the difference of example day data, n for monthly checking meter number of days,

And then obtain optimal estimation:

$\hat{X}=\underset{i=1}{\overset{n}{\Σ}}\[\frac{1}{{\mathrm{\σ}}_i^2}/\underset{i=1}{\overset{n}{\Σ}}\frac{1}{{\mathrm{\σ}}_i^2}\]X_i,$

Wherein, X _ibe meter reading data on the i-thth, for optimal estimation.

By what obtain carry out variance with each day electricity to calculate and contrast, be on duty for the dayly decided to be equivalence by minimum for variance and check meter interval from date, be namely equivalent to all low-voltage customers and be unified in this date and check meter, equivalence electricity of checking meter is check meter electricity sum in all low-voltage platform areas.

(2) calculate adjacent month supply and distribution not the same period interval electricity sales amount;

(2-1) sample data is determined

The daily load choosing somewhere continuous a period of time and corresponding day weather data, take weather data as sample training, weather data comprises the lowest temperature, the highest temperature, medial humidity, quantity of precipitation etc.;

(2-2) electricity sales amount is calculated based on LSSVM

Overload Class is different, its cycle of checking meter accordingly is also different, this patent is divided into residential electricity consumption, commercial power, commercial power, farming power four class, utilizes LSSVM algorithm to calculate each type load respectively and to check meter cycle and total electricity sales amount of calendar month not corresponding period, do difference to it; For residential electricity consumption load, be analyzed as follows, other classification loads as commercial power, commercial power, farming power computation process in like manner.Then corresponding actual monthly power demand is: $L_{T^{\′}}=L_{T_a}+(L_{t_a}-L_{t_a^{\′}}),$

In formula: for electricity sales amount of checking meter by this moon, for interval t _awith t' _aelectricity sales amount difference; t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month, t' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date; represent t _atotal electricity sales amount of segment, represent t' _atotal electricity sales amount of segment.

To difference carry out revising further and can obtain more accurate monthly power demand.

(3) local economic development level correction electricity sales amount result of calculation is utilized;

(3-1) choose monthly electricity sales amount and carry out Multiple Non Linear Regression with corresponding each monthly economic factor sequence (comprising total output value, the primary industry, secondary industry, the tertiary industry etc.), obtain the synthetic relationship model of electricity sales amount and economic factors;

(3-2) utilize electricity sales amount in fit equation to ask local derviation to each economic factor, obtain the susceptibility S that the lower electricity sales amount of different economic factor impact changes it respectively _i; What this susceptibility represented is monthly mean level, by t _awith t ' _abetween the difference of each economic factor, calculate the difference of economic factors to electricity sales amount modified value, that is: ${\mathrm{\ΔL}}_1=\underset{i=1}{\overset{n}{\Σ}}{S}_i\·\mathrm{\Δ}t;$

In formula: S _ibe the susceptibility that i-th economic factor is corresponding, Δ t is t _awith t' _athe difference of two interval average economic levels; t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month, t' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date;

(3-3) by period to be calculated and not corresponding period average economic factors mathematic interpolation electricity difference, revise, the monthly power demand to be calculated of economic factors can be considered

$L_{T^{\′}1}=L_{T_a^{\′}}+(L_{t_a}-L_{t_a^{\′}})+{\mathrm{\ΔL}}_1;$

In formula: for electricity sales amount of checking meter by this moon; for interval t _awith t' _aelectricity sales amount difference; Wherein, t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month, t' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date, represent t _atotal electricity sales amount of segment, represent t' _atotal electricity sales amount of segment; Δ L ₁represent by t _awith t' _athe difference of two interval average economic levels calculates the economic factors that obtains to electricity sales amount difference modified value; for the revised electricity sales amount difference of economic factors.

(4) consideration that random occurrence is occurred is added, corrected Calculation result again;

(4-1) historical data cluster

Collect historical data, by clustering method, after a class is divided on ground similar for the factor such as economy, weather, then for each class data mining, eliminate other aspect factors to the interference of electric quantity change; The strict mathematical of cluster is described below: studied sample set is the nonvoid subset that E, class C are defined as E, namely and cluster is exactly the class c meeting following two conditions ₁, c ₂..., c _kset: (l) C ₁∪ C ₂∪ C ₃∪ ... ∪ C _k=E (2) (any i, j), from first condition, each sample in sample set E must belong to some classes; From second condition, each sample in sample set E at most only belongs to a class.K-Means method is a kind of conventional clustering technique, and the method basic thought is: in space, carry out cluster centered by k point, sorts out the object near them.By the method for iteration, successively upgrade the value of each cluster centre, until obtain best cluster result.

In the present invention, choosing somewhere, to comprise all 6 months every mean daily temperatures, levels of economic development enumerating the enchancement factor time be training data sample, and composition more than 180 organizes sequence, and each group sequence is a sample and clustering object X _i, therefore whole sample set is X={X ₁, X ₂... X _n.

(4-2) according to the impact of different random sexual factor on electricity sales amount, electricity sales amount amplitude of variation absolute value is classified; Enchancement factor arranged, in conjunction with actual production, living quantizes, as shown in Figure 3 simultaneously;

(4-3) correlation rule of random factor and electricity sales amount amplitude of variation is excavated;

Correlation rule represents in database not have between same area to meet the rule that certain specifies the incidence relation required.In correlation rule, remember that transaction database is the number of all subset affairs T in D, D, be designated as | D|.The frequency of occurrences of certain subset affairs T middle term collection A is the number of transactions comprising item collection, i.e. the frequency of item collection A, is designated as f (A).If item collection and then shape as formula be referred to as correlation rule.Correlation rule support be the support of item collection A ∪ B, namely comprise the ratio of A ∪ B in transaction database D, be designated as

$sup(A\⇒B)=P(A\∪B)=\frac{f(A\∪B)}{\left|D\right|}\×100\%$ Formula (1)

Note for the minimum support threshold value of this correlation rule, generally getting 70%, when being greater than minimum support threshold value, claiming item to integrate A as frequent item set.

Correlation rule degree of confidence be comprise A in transaction database D while comprise again the ratio of B, namely conditional probability P (B|A), is designated as

$conf(A\⇒B)=P\left(B\right|A)=\frac{f(A\∪B)}{f\left(A\right)}\×100\%$ Formula (2)

Support and degree of confidence reflect validity and the determinacy of this correlation rule respectively, and wherein support characterizes the probability of occurrence of correlation rule in transaction database or significance level, and namely support is higher, and its correlation degree is higher; Degree of confidence characterizes the credibility of this correlation rule, and namely degree of confidence is higher, and its confidence level is higher.

Choose complete historical data sample, analyze its relevance, note correlation rule is: R _n→ L _m.By electricity sales amount amplitude of variation respective items collection L _mbe designated as transaction database D _m, the corresponding interval time span occurred of this amplitude of variation is designated as | D _m|.In total sample, each random factor R _nthe time span occurred is f (R _n); | D _m| in, each random factor R _nthe time span occurred is f (R _n∪ L _m), then can compute associations rule R by formula (1) _n→ L _msupport.As correlation rule R _n→ L _msupport when being greater than 70%, just think that this correlation rule is with practical value, claim item collection R _nfor frequent item set, i.e. this random factor R _nwith electricity sales amount amplitude of variation L _mbetween there is tight association relation; For non-Frequent Set, then abandon it.Finally set up the incidence relation of each random factor and electricity sales amount amplitude of variation, such as m the interval L of electricity sales amount amplitude of variation _mcorresponding random factor collection R _m,nrepresent, note: wherein N _mfor causing L _mthe number of all random factors.

Degree of confidence is tried to achieve by formula (2), calculate corresponding normal weight coefficient respectively, because only considering certain the random factor frequent item set corresponding to all kinds of electricity sales amount amplitude of variation when calculating, its normal weight coefficient can be obtained, therefore can using the electricity sales amount forecast value revision coefficient of this coefficient as its corresponding random factor:

wherein, ω _m,nfor the interval L of electricity sales amount amplitude of variation _mmiddle random factor R _m,nnormal weight coefficient; C _m,nfor corresponding degree of confidence.

(4-4) level of economic development membership function is built

Consider time, difference spatially, and the electricity sales amount change that under different economy level of development, same enchancement factor causes may be different, therefore, build level of economic development membership function as accompanying drawing 2, revise correlation rule result in conjunction with degree of membership γ, solve the problem that when correlation rule is set up, the meteorological background of the economy of two computation intervals is inconsistent.

Utilize the sample moon and the moon to be predicted level of economic development difference come x _ncarry out quantification treatment:

$x_n=1-\frac{{\mathrm{GDP}}_{yangben}}{{\mathrm{GDP}}_{yuce}}$

(4-5) integrated economics development level membership function, quantizes the electricity sales amount amplitude of variation under the impact of each enchancement factor

Based on above-mentioned association rule mining result, integrated economics development level degree of membership, result is done following process:

The all enchancement factors comprised under each electricity sales amount amplitude of variation, are multiplied by the maximal value in this amplitude of variation interval by coefficient, divided by the T.T. that this random factor occurs after suing for peace, be the impact on electricity sales amount change in this enchancement factor unit interval.Such as, enchancement factor R _j, in its unit interval, (h) can be quantified as the impact of electricity sales amount amplitude of variation:

${\mathrm{\η}}_j=\frac{\underset{i=1}{\overset{m}{\Σ}}{\mathrm{\ω}}_{i,j}\·L_i}{t_j}$

Therefore, the correction difference of three stage electricity sales amounts can just be obtained that is:

${\mathrm{\ΔL}}_2=(L_{t_a}+L_1^{\′})(1+{\mathrm{\ηt}}_{j1}^{\′})-(L_{t_a}+L_1^{\′\′})(1+{\mathrm{\ηt}}_{j2}^{\′})=\underset{j=1}{\overset{n}{\Σ}}{R}_j\·t_{j1}^{\′}-\underset{j=1}{\overset{n}{\Σ}}{R}_j\·t_{j2}^{\′}$

Wherein, t' _j1for factor R _jat total hourage of this month not corresponding interval generation, t' _j2for factor R _jtotal hourage that not corresponding interval occurs in last month.L' ₁, L " ₁be respectively the electricity sales amount affected by average economic level of this month and last month.

(5) calculate electricity sales amount, utilize formulae discovery this month statistical line losses;

The monthly electricity sales amount of final reality is $L_{T^{\′}}=L_{T_a}+(L_{t_a}+L_1^{\′})(1+{\mathrm{\γ\ηt}}_{j1}^{\′})-(L_{t_a}+L_1^{\′\′})(1+{\mathrm{\γ\ηt}}_{j2}^{\′}),$

In formula: for this month actual electricity sales amount of checking meter; γ is of that month level of economic development degree of membership; T' _j1for a certain enchancement factor is at total hourage of this month not corresponding interval generation; T' _j2for a certain enchancement factor total hourage that not corresponding interval occurs in last month; η be in the unit time a certain enchancement factor on the impact of electricity sales amount amplitude of variation; L' ₁, L " ₁be respectively the electricity sales amount affected by average economic level of this month and last month; t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month, t' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date; represent t _atotal electricity sales amount of segment; represent t' _atotal electricity sales amount of segment;

After in like manner calculating the total actual electricity sales amount of all categories load, according to statistical line losses rate computing formula, calculate actual count line loss per unit.

(6) electricity sales amount obtained step (3) and step (5) and statistical line losses electricity carry out mark soundness verification.

(6-1) because data bulk level, the unit etc. such as economy, meteorology, enchancement factor duration are inconsistent, need data to carry out following nondimensionalization process;

$x_{ij}^{*}=\frac{x_{ij}-\min x_i}{\max x_i-\min x_i}$

Wherein, for result after the process of jth monthly factor i nondimensionalization, x _ijfor the actual value of jth monthly factor i, minx _ifor factor i minimum value, maxx _ifor factor i maximal value.

(6-2) model choosing is carried out to sample;

For electricity sales amount and loss electricity result two aspect, carry out soundness verification respectively,

Electricity sales amount is verified: 1. the comprehensive random factor curve of meteorological variation tendency 3. of the corresponding month economic trend 2. corresponding moon.

Loss electricity checking: the 1. corresponding monthly load factor of corresponding month supply load 2. corresponding monthly average power factor 3..

(6-3) mark post distance calculates, the rationality of verification computation result.

In the present invention, using above-mentioned mark post sample as reference sample.The electricity sales amount of calculating, loss electricity and the electricity of checking meter, loss electricity and the distance of mark post sample respectively, adopts following form calculus:

$D_i=\sqrt{\underset{j=1}{\overset{p}{\Σ}}{(x_{ij}-y_{ij})}^2},i=1,2,...,n$

Wherein, y _ijfor the mark post sample value of jth monthly factor i, x _ijfor the actual value of jth monthly factor i.

If distance is less, then i-th sample point is more near from mark post sample point, and the rationality of expression i-th sample is higher.The rationality of the revised monthly electricity sales amount of the present invention and statistical line losses electricity can be verified by the method, obtain more real statistical line losses rate.

Above-described embodiment is only be described the preferred embodiment of the present invention; not scope of the present invention is limited; under not departing from the present invention and designing the prerequisite of spirit; the various distortion that those of ordinary skill in the art make technical scheme of the present invention and improvement, all should fall in protection domain that claims of the present invention determines.

Claims (7)

1., based on statistical line losses computing method for LSSVM algorithm and association rule mining, it is characterized in that, the step of described computing method is as follows:

(1) carry out processing to the meter reading method of low-voltage platform area mesolow user and simplify, determining meter reading method and the cycle of power load;

(2) calculate adjacent month supply and distribution not the same period interval electricity sales amount;

(3) local economic development horizontal corrected Calculation electricity sales amount result is utilized;

(4) consideration that random occurrence is occurred is added, corrected Calculation result again;

(5) calculate electricity sales amount, utilize formulae discovery this month statistical line losses;

(6) electricity sales amount obtained step (3) and step (5) and statistical line losses electricity carry out mark soundness verification.

2. a kind of statistical line losses computing method based on LSSVM algorithm and association rule mining according to claim 1, it is characterized in that, the detailed process of described step (1) is as follows:

(1-1) determine that equivalence is checked meter example day

Equivalence check meter client electricity that example refers to all residents day all equivalence check meter in this sky, equivalence example of checking meter is determined by following formula day,

Wherein e _ibe electricity of checking meter on the i-thth, n is number of days of monthly checking meter, E _jfor every monthly power demand, be the proportion that electricity of checking meter on the i-thth accounts for when monthly power demand;

(1-2) determine that equivalence is checked meter interval

In conjunction with the data fusion method of optimal weighting, the meter reading data equivalence of every day is considered as a sensor, check meter example day result of calculation for reference value with equivalence, obtaining the weight of meter reading data every day, select relative variance reckling, is optimal estimation;

Weight asks for formula: $\mathrm{\ω}=\frac{1}{{\mathrm{\σ}}_i^2}/\underset{i=1}{\overset{n}{\Σ}}\frac{1}{{\mathrm{\σ}}_i^2}$

Wherein, σ _ibe that meter reading data on the i-thth and equivalence are checked meter the difference of example day data, n for monthly checking meter number of days, and then optimal estimation:

$\hat{X}=\underset{i=1}{\overset{n}{\Σ}}\[\frac{1}{{\mathrm{\σ}}_i^2}/\underset{i=1}{\overset{n}{\Σ}}\frac{1}{{\mathrm{\σ}}_i^2}\]X_i$

Wherein, X _ibe meter reading data on the i-thth, for optimal estimation;

By what obtain carry out variance with each day electricity to calculate and contrast, be on duty for the dayly decided to be equivalence by minimum for variance and check meter interval from date, be namely equivalent to all low-voltage customers and be unified in this date and check meter, equivalence electricity of checking meter is check meter electricity sum in all low-voltage platform areas.

3. a kind of statistical line losses computing method based on LSSVM algorithm and association rule mining according to claim 1, it is characterized in that, the detailed process of described step (2) is as follows:

(2-1) sample data is determined

The daily load choosing somewhere continuous a period of time and corresponding day weather data, take weather data as sample training, weather data comprises the lowest temperature, the highest temperature, medial humidity, quantity of precipitation;

(2-2) electricity sales amount is calculated based on LSSVM

Overload Class is different, its cycle of checking meter accordingly is also different, Overload Class is divided into residential electricity consumption, commercial power, commercial power, farming power four class, utilizes LSSVM algorithm to calculate each type load respectively and to check meter cycle and total electricity sales amount of calendar month not corresponding period, difference is done to it; For residential electricity consumption load, be analyzed as follows, other classification loads as commercial power, commercial power, farming power computation process in like manner; Then corresponding actual monthly power demand is:

In formula: for electricity sales amount of checking meter by this moon, for interval t _awith t' _aelectricity sales amount difference; t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month, t' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date, represent t _atotal electricity sales amount of segment, represent t' _atotal electricity sales amount of segment;

To difference carry out revising further and can obtain more accurate monthly power demand.

4. a kind of statistical line losses computing method based on LSSVM algorithm and association rule mining according to claim 1, it is characterized in that, the detailed process of described step (3) is as follows:

(3-1) choose monthly electricity sales amount and carry out Multiple Non Linear Regression with corresponding each monthly economic factor sequence, obtain the synthetic relationship model of electricity sales amount and economic factors;

(3-2) utilize electricity sales amount in fit equation to ask local derviation to each economic factor, obtain the susceptibility S that the lower electricity sales amount of different economic factor impact changes it respectively _i; What this susceptibility represented is monthly mean level, by t _awith t ' _abetween the difference of each economic factor, the economic factors calculated is to electricity sales amount difference modified value, that is: ${\mathrm{\ΔL}}_1=\underset{i=1}{\overset{n}{\Σ}}{S}_i\·\mathrm{\Δ}t;$

In formula: S _ibe the susceptibility that i-th economic factor is corresponding, Δ t is t _awith t' _athe difference of two interval average economic levels; t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month, t' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date;

(3-3) by period to be calculated and not corresponding period average economic factors mathematic interpolation electricity difference, revise, the monthly power demand to be calculated of economic factors can be considered

$L_{T^{\′}1}=L_{T_a}+(L_{t_a}-L_{t_a^{\′}})+{\mathrm{\ΔL}}_1;$

In formula: for electricity sales amount of checking meter by this moon; for interval t _awith t' _aelectricity sales amount difference; t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month; T' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date; represent t _atotal electricity sales amount of segment; represent t' _atotal electricity sales amount of segment; Δ L ₁represent by t _awith t' _athe difference of two interval average economic levels calculates the economic factors that obtains to electricity sales amount difference modified value; for the revised electricity sales amount difference of economic factors.

5. a kind of statistical line losses computing method based on LSSVM algorithm and association rule mining according to claim 1, it is characterized in that, the detailed process of described step (4) is as follows:

(4-1) historical data cluster

Collect historical data, by clustering method, after a class is divided on ground similar for the factor such as economy, weather, then for each class data mining, eliminate other aspect factors to the interference of electric quantity change;

(4-2) according to the impact of different random sexual factor on electricity sales amount, electricity sales amount amplitude of variation absolute value is classified; Enchancement factor arranged, in conjunction with actual production, living quantizes simultaneously;

(4-3) correlation rule of random factor and electricity sales amount amplitude of variation is excavated;

(4-4) level of economic development membership function is built;

(4-5) integrated economics development level membership function, quantizes the electricity sales amount amplitude of variation under the impact of each enchancement factor.

6. a kind of statistical line losses computing method based on LSSVM algorithm and association rule mining according to claim 1, it is characterized in that, the detailed process of described step (5) is as follows:

The monthly electricity sales amount of final reality is $L_{T^{\′}}=L_{T_a}+(L_{t_a}+L_1^{\′})(1+{\mathrm{\γ\ηt}}_{j1}^{\′})-(L_{t_a}+L_1^{\′\′})(1+{\mathrm{\γ\ηt}}_{j2}^{\′}),$

In formula: for this month actual electricity sales amount of checking meter; γ is of that month level of economic development degree of membership; T' _j1for a certain enchancement factor is at total hourage of this month not corresponding interval generation; T' _j2for a certain enchancement factor total hourage that not corresponding interval occurs in last month; η be in the unit time a certain enchancement factor on the impact of electricity sales amount amplitude of variation; L' ₁, L " ₁be respectively the electricity sales amount affected by average economic level of this month and last month; t _arepresent the load of residential electricity consumption classification check meter end cycle time apart from the segment at the end of month; T' _arepresent that residential electricity consumption classification load sale of electricity period ratio of checking meter is powered the segment that the cycle of checking meter shifts to an earlier date; represent t _atotal electricity sales amount of segment; represent t' _atotal electricity sales amount of segment;

After in like manner calculating the total actual electricity sales amount of all Overload Class, according to statistical line losses rate computing formula, calculate actual count line loss per unit.

7. a kind of statistical line losses computing method based on LSSVM algorithm and association rule mining according to claim 1, it is characterized in that, the detailed process of described step (6) is as follows:

(6-1) nondimensionalization process is carried out to long data when economy, meteorology, enchancement factor;

(6-2) model choosing is carried out to sample;

(6-3) mark post distance calculates, the rationality of verification computation result.