CN106096324A - The power transmission and transformation main equipment load data disappearance returned based on k neighbour fills up algorithm - Google Patents

Abstract

A kind of electric power power transmission and transforming equipment load data disappearance returned based on k neighbour fills up algorithm, it is characterised in that filling steps is: data set D is divided into two subsets, i.e. subset D m, subset D c.Vector x in subset D m is divided into x=[x_o；x_m].Calculate in corresponding moment value, vector x_oWith the Euclidean distance of institute's directed quantity in subset D c.According to distance size the vector in subset Dc is carried out ascending order arrangement, obtain subset D ' c.Take subset D ' front k vector (y in c₁,y₂,…,y_k).Calculate x weighting k neighbour's regressand value at the i-th moment missing values.Repetition step one is to step 6, until subset D_mIn institute's directed quantity the most processed.It provides the benefit that: need not use training set to be trained, the advantage of nearest neighbor algorithm is to be not required for knowing the distribution function of sample to be predicted in advance, therefore has directly perceived, without features such as priori statistical knowledge, unsupervised learnings.

Description

The power transmission and transformation main equipment load data disappearance returned based on k neighbour fills up algorithm

Technical field

The present invention relates to the big Data Mining of power system, particularly a kind of power transmission and transformation main equipment load data disappearance is filled out Mend algorithm.

Background technology

Along with improving constantly of electric power information degree and increasing rapidly of electric power big data quantity, research is applicable to electricity The algorithm of the big data mining of power also sets up effective Knowledge Discovery Model, and the innovation of intelligent grid business model and development are had weight Want meaning.

In power system, Various types of data collection is power system with the power transmission and transformation main equipment load data of monitoring system collection The basis of management and running, security and stability analysis, equipment state and risk assessment.But, in the actual motion of power system, one Aspect can cause observing data unusual situation occur due to data acquisition channel mistake, remote-terminal unit failure and other reasons, with Cause inconsistent with most of observations；On the other hand, due to specific event, (such as line maintenance, cutting load has a power failure, major issue is impacted Deng) cause the ANOMALOUS VARIATIONS of load, also result in observation data and go against the established rules.Additionally, data metering device or storage device event Barrier is likely to cause sub-load shortage of data.Therefore, must be to original loads before carrying out load data analysis and modeling Abnormal data in data is correspondingly filled up and revises.

The problem the most both at home and abroad treatment research of Power system load data disappearance being existed some universalitys.First, document In method be both on a small scale data set, the computational efficiency for large-scale dataset is relatively low；Second, these methods are to list The treatment effect of individual discrete bad data is preferable, and the treatment effect of bad data continuous for sheet is general.

Summary of the invention

The invention aims to solve the problems referred to above, devise a kind of electric power power transmission and transformation returned based on k neighbour and set Standby load data disappearance fills up algorithm.Specific design scheme is:

Filling steps is:

Step one, is divided into two subsets, i.e. subset D m, subset D c by data set D.

Step 2, is divided into x=[x by the vector x in subset D m_o；x_m]。

Step 3, calculates in corresponding moment value, vector x_oWith the Euclidean distance of institute's directed quantity in subset D c.

Step 4, according to distance size the vector in subset Dc is carried out ascending order arrangement, obtain subset D ' c.

Step 5, take subset D ' front k vector (y in c₁,y₂,…,y_k)。

Step 6, calculating x is at weighting k neighbour's regressand value of the i-th moment missing values:

$x^{\left(i\right)}=\underset{j=1}{\overset{k}{\Σ}}{w}_j{y}_j^{\left(i\right)}$

Wherein w_jIt is vector y_jWeight.

Step 7, repetition step one is to step 6, until subset D_mIn institute's directed quantity the most processed.

In step one, described subset D m is the load curve set comprising missing values, and subset D c is not comprise missing values Load curve set.

In step 2, described vector x_oFor intact misorientation amount, vector x_mFor disappearance vector.

In step 6, greater weight should be taken with xo apart near vector, less weight should be taken with xo apart from remote vector.

In step 6, weight function computing formula is:

$w_j=\frac{1}{k},j=1,2,...,k$

$w_j=\frac{1/dist(x_o,y_j)}{\underset{j=1}{\overset{k}{\Σ}}1/dist(x_o,y_j)},j=1,2,...,k$

$w_j=\frac{e^{-dist(x_o,y_j)}}{\underset{j=1}{\overset{k}{\Σ}}{e}^{-dist(x_o,y_j)}},j=1,2,...,k.$

In weight function computing formula, described dist (X1, Yj) represents the Euclidean distance of vector X1 Yu Yj.

The electric power power transmission and transforming equipment load data based on k neighbour recurrence obtained by the technique scheme of the present invention is lacked Algorithm is filled up in mistake, and it provides the benefit that:

K nearest neighbor algorithm is a kind of inertia learning algorithm, it is not necessary to using training set to be trained, its time complexity is O (n), the sample number during wherein n is training set.The advantage of k nearest neighbor algorithm is to be not required for knowing the distribution of sample to be predicted in advance Function, therefore has directly perceived, without features such as priori statistical knowledge, unsupervised learnings.

Accompanying drawing explanation

Fig. 1 be the present invention under the conditions of single missing values (t=23), by average fill obtain fill up result；

Fig. 2 be the present invention under the conditions of single missing values (t=23), by kNN return fill w3 obtain fill up result；

Under the conditions of Fig. 3 is consecutive miss value (t=21-25) of the present invention, fill up result by what average filling obtained；

Under the conditions of Fig. 4 is consecutive miss value (t=21-25) of the present invention, fill up knot by what kNN recurrence filling w3 obtained Really；

Detailed description of the invention

Below in conjunction with the accompanying drawings the present invention is specifically described.

Filling steps is:

Step one, is divided into two subsets, i.e. subset D m, subset D c by data set D.

Step 2, is divided into x=[x by the vector x in subset D m_o；x_m]。

Step 3, calculates in corresponding moment value, vector x_oWith the Euclidean distance of institute's directed quantity in subset D c.

Step 4, according to distance size the vector in subset Dc is carried out ascending order arrangement, obtain subset D ' c.

Step 5, take subset D ' front k vector (y in c₁,y₂,…,y_k)。

Step 6, calculating x is at weighting k neighbour's regressand value of the i-th moment missing values:

$x^{\left(i\right)}=\underset{j=1}{\overset{k}{\Σ}}{w}_j{y}_j^{\left(i\right)}$

Wherein w_jIt is vector y_jWeight

Step 7, repetition step one is to step 6, until subset D_mIn institute's directed quantity the most processed.

In step one, described subset D m is the load curve set comprising missing values, and subset D c is not comprise missing values Load curve set.

In step 2, described vector x_oFor intact misorientation amount, vector x_mFor disappearance vector.

In step 6, greater weight should be taken with xo apart near vector, less weight should be taken with xo apart from remote vector.

In step 6, weight function computing formula is:

$w_j=\frac{1}{k},j=1,2,...,k$

$w_j=\frac{1/dist(x_o,y_j)}{\underset{j=1}{\overset{k}{\Σ}}1/dist(x_o,y_j)},j=1,2,...,k$

$w_j=\frac{e^{-dist(x_o,y_j)}}{\underset{j=1}{\overset{k}{\Σ}}{e}^{-dist(x_o,y_j)}},j=1,2,...,k$

In weight function computing formula, described dist (X1, Yj) represents the Euclidean distance of vector X1 Yu Yj.

Embodiment 1

Take certain provincial power network 185 main apparatus load data of a year 365 days, totally 67525 load curves.Every 48 points of load curve record whole day, totally 3241200 data points.Wherein 2 curves are artificially manufactured part missing data, bent Line 1 is single missing values, lacks at t=23；Curve 2 is consecutive miss value, and data lack at t=21-25

Data set D is all 67525 and meets collection of curves, and being classified as two load curve set Dm and Dc, Dm is Having two collection of curves of missing values, Dc is that remaining is worth collection of curves without disappearance.

Consider the curve X1 of single missing values, be classified as intact become estranged disappearance two parts X1=[x1 ..x22, x24,…x48；x23].

Calculating X1 and the Euclidean distance of all curves in subset D c, the load curve in Dc uses owning in addition to x23 Value.

According to distance size, the vector in Dc is carried out ascending order arrangement, obtain D ' c.

According to k value take front k vector (Y1, Y2 ..., Yk).

Calculating X1 is at weighting k neighbour's regressand value of the i-th moment missing values:

$x^{\left(i\right)}=\underset{j=1}{\overset{k}{\Σ}}{w}_j{y}_j^{\left(i\right)}$

Three kinds of weight w of consideration in the present embodiment:

$w_{1j}=\frac{1}{k},j=1,2,...,k$

$w_{2j}=\frac{1/dist(X_1,Y_j)}{\underset{j=1}{\overset{k}{\Σ}}1/dist(X_1,Y_j)},j=1,2,...,k$

$w_{3j}=\frac{e^{-dist(X_1,Y_j)}}{\underset{j=1}{\overset{k}{\Σ}}{e}^{-dist(X_1,Y_j)}},j=1,2,...,k$

Wherein dist (X1, Yj) represents the Euclidean distance of vector X1 Yu Yj.

After completing the above-mentioned steps of the present embodiment, by existing average fill method, complete described in the present embodiment Certain saves shortage of data and calculates.

Fig. 1 be the present invention under the conditions of single missing values (t=23), by average fill obtain fill up result；Fig. 2 is The present invention, under the conditions of single missing values (t=23), fills up result by what kNN recurrence filling w3 obtained, institute as shown in Figure 1, Figure 2 Showing, for single discrete missing values, the filling effect that k neighbour returns is substantially better than average filling mode.

Under the conditions of Fig. 3 is consecutive miss value (t=21-25) of the present invention, fill up result by what average filling obtained；Fig. 4 Under the conditions of being consecutive miss value (t=21-25) of the present invention, fill up result for connecting in flakes by what kNN recurrence filling w3 obtained Continuous missing values, the filling effect that k neighbour returns is substantially better than average filling mode.

Embodiment 2

Take certain provincial power network 185 main apparatus load data of a year 365 days, totally 67525 load curves.Every 48 points of load curve record whole day, totally 3241200 data points.Wherein 2 curves are artificially manufactured part missing data, bent Line 1 is single missing values, lacks at t=23；Curve 2 is consecutive miss value, and data lack at t=21-25

Data set D is all 67525 and meets collection of curves, and being classified as two load curve set Dm and Dc, Dm is Having two collection of curves of missing values, Dc is that remaining is worth collection of curves without disappearance.

Consider the curve X1 of single missing values, be classified as intact become estranged disappearance two parts X1=[x1 ..x22, x24,…x48；x23].

Calculating X1 and the Euclidean distance of all curves in subset D c, the load curve in Dc uses owning in addition to x23 Value.

According to distance size, the vector in Dc is carried out ascending order arrangement, obtain D ' c.

According to k value take front k vector (Y1, Y2 ..., Yk).

Calculating X1 is at weighting k neighbour's regressand value of the i-th moment missing values:

$x^{\left(i\right)}=\underset{j=1}{\overset{k}{\Σ}}{w}_j{y}_j^{\left(i\right)}$

Three kinds of weight w of consideration in the present embodiment:

$w_{1j}=\frac{1}{k},j=1,2,...,k$

$w_{2j}=\frac{1/dist(X_1,Y_j)}{\underset{j=1}{\overset{k}{\Σ}}1/dist(X_1,Y_j)},j=1,2,...,k$

$w_{3j}=\frac{e^{-dist(X_1,Y_j)}}{\underset{j=1}{\overset{k}{\Σ}}{e}^{-dist(X_1,Y_j)}},j=1,2,...,k$

Wherein dist (X1, Yj) represents the Euclidean distance of vector X1 Yu Yj.

After completing the above-mentioned steps of the present embodiment, by existing average fill method, complete described in the present embodiment Certain saves shortage of data and calculates；

After completing the above-mentioned steps of the present embodiment, by existing linear straight cutting method, complete described in the present embodiment Certain saves shortage of data and calculates；

After completing the above-mentioned steps of the present embodiment, by existing batten straight cutting method, complete described in the present embodiment Certain saves shortage of data and calculates；

Return (taking three kinds of weight functions) lack with average filling, linear interpolation, cubic spline interpolation, k neighbour respectively Value is filled up, with using mean absolute percentage error (mean absolute percentage error, MAPE) index evaluation X1 load curve fill up effect, described MAPE index computing formula is

$MAPE=\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}\left|\frac{x_i-{\hat{x}}_i}{x_i}\right|$

Wherein x_iIt is actual value,It it is predictive value.The accuracy of visible MAPE the least explanation prediction is the highest.

Table 1 is the MAPE index (t=23) of the various complementing method of single missing values

The MAPE (t=21-25) that table 2 is the various complementing method of consecutive miss value

As shown in table 1, table 2,

1., for single discrete missing values, the filling effect that k neighbour returns is substantially better than other modes.

2., for consecutive miss value in blocks, the filling effect that k neighbour returns is better than additive method.

3. use the forecast error of exponential weight function w3 less than other two kinds of weight functions.

Technique scheme only embodies the optimal technical scheme of technical solution of the present invention, those skilled in the art Some variations may made some of which part all embody the principle of the present invention, belong to protection scope of the present invention it In.

Claims (6)

1. the electric power power transmission and transforming equipment load data disappearance returned based on k neighbour fills up algorithm, it is characterised in that fill up step Suddenly it is:

Step one, is divided into two subsets, i.e. subset D m, subset D c by data set D.

Step 2, is divided into x=[x by the vector x in subset D m_o；x_m]。

Step 3, calculates in corresponding moment value, vector x_oWith the Euclidean distance of institute's directed quantity in subset D c.

Step 4, according to distance size the vector in subset Dc is carried out ascending order arrangement, obtain subset D ' c.

Step 5, take subset D ' front k vector (y in c₁,y₂,…,y_k)。

Step 6, calculating x is at weighting k neighbour's regressand value of the i-th moment missing values:

$x^{\left(i\right)}=\underset{j=1}{\overset{k}{\Σ}}{w}_j{y}_j^{\left(i\right)}$

Wherein w_jIt is vector y_jWeight

Step 7, repetition step one is to step 6, until subset D_mIn institute's directed quantity the most processed.

2. fill up algorithm according to the electric power power transmission and transforming equipment load data disappearance based on k neighbour recurrence described in claim 1, It is characterized in that, in step one, described subset D m is the load curve set comprising missing values, and subset D c is not comprise missing values Load curve set.

3. fill up algorithm according to the electric power power transmission and transforming equipment load data disappearance based on k neighbour recurrence described in claim 1, It is characterized in that, in step 2, described vector x_oFor intact misorientation amount, vector x_mFor disappearance vector.

4. fill up algorithm according to the electric power power transmission and transforming equipment load data disappearance based on k neighbour recurrence described in claim 1, It is characterized in that, in step 6, greater weight should be taken with xo apart near vector, relatively little Quan should be taken with xo apart from remote vector Weight.

5. fill up algorithm according to the electric power power transmission and transforming equipment load data disappearance based on k neighbour recurrence described in claim 1, It is characterized in that, in step 6, weight function computing formula is:

$w_j=\frac{1}{k},j=1,2,...,k$

$w_j=\frac{1/dist(x_o,y_j)}{\underset{j=1}{\overset{k}{\Σ}}1/dist(x_o,y_j)},j=1,2,...,k$

$w_j=\frac{e^{-dist(x_o,y_j)}}{\underset{j=1}{\overset{k}{\Σ}}{e}^{-dist(x_o,y_j)}},j=1,2,...,k.$

6. fill up algorithm according to the electric power power transmission and transforming equipment load data disappearance based on k neighbour recurrence described in claim 5, It is characterized in that, in weight function computing formula, described dist (X₁,Y_j) represent vector X₁With Y_jEuclidean distance.