CN105866725A - Method for fault classification of smart electric meter based on cluster analysis and cloud model - Google Patents

Abstract

The invention relates to a method for fault classification of a smart electric meter based on cluster analysis and a cloud model. The method comprises a step 1) of obtaining historical smart electric meter fault data sample points, and adopting a K-means algorithm to divide historical smart electric meter fault data sample points into K large fault classes and central values corresponding to each large class; a step 2) of taking the central values corresponding to each large class as sample means, taking smart electric meter fault data sample points contained in each large fault class as data points, and generating a corresponding K-class electric meter fault cloud model; a step 3) of adopting a reverse normal cloud generator to calculate the electric meter fault cloud model, and obtaining qualitative cloud characteristics of the electric meter fault cloud model; and a step 4) of subdividing the K large fault classes into a plurality of small fault classes according to the qualitative cloud characteristics. Compared with the prior art, the method has the advantages of qualitative analysis and fine classification.

Description

A kind of intelligent electric meter Fault Classification based on cluster analysis and cloud model

Technical field

The present invention relates to intelligent electric meter Fault Classification, especially relate to a kind of based on cluster analysis with the intelligence of cloud model Can electric meter fault sorting technique.

Background technology

The fast development of computer and network technologies so that acquisition of information and analysis means are increasingly advanced, data mining is own Becoming enterprise, the important tool of mechanism's data rule, data mining platform is broadly divided into storage calculating platform and data are dug Pick algorithm, for existing intelligent electric energy meter data characteristics, present intelligent electric energy meter fault data quantity is numerous and jumbled various, does not also have It can be separated by a kind of method of qualitative analysis by fault type, and existing one of being badly in need of is by quantitatively classifying to accident analysis qualitatively Method.

Summary of the invention

Defect that the purpose of the present invention is contemplated to overcome above-mentioned prior art to exist and a kind of qualitative analysis, classification are provided Careful intelligent electric meter Fault Classification based on cluster analysis and cloud model.

The purpose of the present invention can be achieved through the following technical solutions:

A kind of intelligent electric meter Fault Classification based on cluster analysis and cloud model, comprises the following steps:

1) obtain history intelligent electric meter fault data sample point, and use K-means algorithm by history intelligent electric meter fault Data sample point is divided into K the big class of fault and central value corresponding to each big class；

2) using central value corresponding for each big class as sample average, the intelligent electric meter number of faults that the big class of each fault comprises According to sample point as data point, generate corresponding K level electrical energy meter fault cloud model；

3) use reverse Normal Cloud generator that electrical energy meter fault cloud model is calculated, obtain electrical energy meter fault cloud model Qualitative cloud feature；

4) multiple fault group will be divided under K the big class of fault according to qualitative cloud feature.

Described step 1) specifically include following steps:

11) select K history intelligent electric meter fault data sample point as the cluster centre of the big class of primary fault；

12) to remaining history intelligent electric meter fault data sample point, according to itself and the distance of each cluster centre, by it Conclude into the closest big class of fault；

13) calculate the sample average of the big class of each fault, and calculate canonical measure function, it may be assumed that

$E=\underset{i=1}{\overset{K}{\Σ}}\underset{\mathrm{\τ}\∈Ci}{\Σ}|\mathrm{\τ}-m_i{|}^2$

$m_i=\frac{1}{N_i}\underset{\mathrm{\τ}\∈Ci}{\Σ}\mathrm{\τ}$

In formula, m_iFor the electrical energy meter fault data sample average of i-th bunch, C_iElectrical energy meter fault data sample for i-th bunch This point is gathered, N_iFor the electrical energy meter fault data sample sum of i-th bunch, E is canonical measure function, and τ is electrical energy meter fault number According to sample point；

14) replace initial cluster center with the sample average of the big class of each fault, repeat step 12)～14), until standard Measure function is restrained.

Described step 3) in, by the sample average of electrical energy meter fault cloud modelSingle order sample Absolute Central Moment T and sample This variance S²Respectively as reverse Normal Cloud generator input value, i.e. expected value Ex, entropy En and super entropy He.

Described sample averageSingle order sample Absolute Central Moment T and sample variance S²Calculating formula be:

$\stackrel{\‾}{X}=\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{x}_i$

$T=\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}|x_i-\stackrel{\‾}{X}|$

$S^2=\frac{1}{n-1}\underset{i=1}{\overset{n}{\Σ}}{(x_i-\stackrel{\‾}{X})}^2$

Wherein, n is the quantity of the data point that the big class of fault comprises, x_iValue for i-th data point.

Described expected value Ex, entropy En and the calculating formula of super entropy He be:

$Ex=\stackrel{\‾}{X}$

$En=\sqrt{\frac{\mathrm{\π}}{2}}\×\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}|x_i-Ex|$

$He=\sqrt{S^2-{\mathrm{En}}^2}.$

A kind of intelligent electric meter Fault Classification based on cluster analysis and cloud model the most according to claim 1, It is characterized in that, described step 4) in, K the big class of fault includes error connection, multiplexing electric abnormality, stealing, battery failure and equipment Fault.

A kind of intelligent electric meter Fault Classification based on cluster analysis and cloud model the most according to claim 6, It is characterized in that, in described step 4) fault group under K the big class of fault is categorized as:

Compared with prior art, the invention have the advantages that

One, qualitative analysis: the present invention initially with K-means clustering algorithm by numerous and jumbled fault data rough segmentation be 5 big Class, is carrying out quantitatively changing to qualitatively according to reverse Normal Cloud generator class big to each fault respectively, number of faults the most at last According to point multiple fault groups, classification is accurately.

Two, classify careful: fault data is divided into 5 big classes, 96 groups by the present invention, and classification is fine, it is simple to in the future Electrical energy meter fault judge and fault divide.

Accompanying drawing explanation

Fig. 1 is One-Dimensional Normal Cloud model profile figure.

Fig. 2 is Normal Cloud Generator principle schematic.

Fig. 3 is backward cloud generator principle schematic.

Fig. 4 is intelligent electric energy meter evaluation index system.

Fig. 5 is index chart of frequency distribution.

Detailed description of the invention

The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.

Embodiment

K-means algorithm is a kind of most widely used clustering algorithm.This algorithm is with the minimum classification of canonical measure function Principle, is divided into K bunch N number of electrical energy meter fault data sample point.Its cluster result makes the electrical energy meter fault data sample of same bunch This point has higher similarity, and the electrical energy meter fault data sample point similarity degree between each bunch is relatively low.K-means algorithm Concrete classifying step is as follows:

(1) arbitrarily select K electrical energy meter fault data sample point as initial cluster center；

(2) to remaining each electrical energy meter fault data sample point, according to itself and the distance at each bunch of center, it is given Nearest bunch；

(3) calculate the sample average of each bunch, and calculate canonical measure function, it may be assumed that

$m_i=\frac{1}{N_i}\underset{\mathrm{\τ}\∈Ci}{\Σ}\mathrm{\τ},E=\underset{i=1}{\overset{K}{\Σ}}\underset{\mathrm{\τ}\∈Ci}{\Σ}|\mathrm{\τ}-m_i{|}^2$

In formula, m_iElectrical energy meter fault data sample average for i-th bunch；Ci is the electrical energy meter fault data of i-th bunch Sample point set；Ni is the electrical energy meter fault data sample sum of i-th bunch；E is canonical measure function；τ is electrical energy meter fault Data sample point.

(4) replace initial cluster center with the sample average of each bunch, repeat step (2)～step (4), until standard is surveyed Degree function convergence.Concrete application herein is: electrical energy meter fault data classification indicators and classification number K are input to K-means and calculate In method, the K obtaining electrical energy meter fault data classification indicators hives off, each classification indicators correspondence one electrical energy meter fault, thus Obtain the classification results of electrical energy meter fault data.

Traditional electrical energy meter fault analyzes method, often with catastrophe failure as evaluation criterion, it is impossible to the actual electricity of reflection Can random included in table fault data, fuzzy message and between relatedness.Additionally, part electrical energy meter fault differentiate be based on Empirical data obtains, if utilizing inaccurate empirical data to provide an explicit value as electrical energy meter fault result of calculation, aobvious It is so irrational.Accordingly, it would be desirable to introduce rational data analysing method, in conjunction with randomness, ambiguity etc. indetermination theory, Electrical energy meter fault data are carried out qualitative and quantitative analysis.

This project uses the Clouds theory produced on the basis of theory of probability and fuzzy mathematics theory to come electric energy meter event Unascertained information in barrier data carries out mining analysis.Clouds theory by the firm academician of China Li De set up by uncertain The data analysing method that qualitativing concept and quantitative data are mutually changed.Clouds theory reflects concept and probabilistic two of knowledge Aspect, i.e. ambiguity and randomness, and can combine both, set up qualitative and quantitatively between correspondence mappings relation.

If U is quantitative domain space, whereinT is the qualitativing concept in domain space, if elementRight Degree of membership C of T concept_TX () has a numerical value tended towards stability on [0,1], then in the mapping of [0,1] from qualitativing concept to domain Distribution becomes cloud.I.e.

C_T(x):U→[0,1]

$\∀x\∈U:x\→C_T\left(x\right)$

Based on these mapping relations, cloud model can be obtained in conjunction with Normal Distribution Theory.Normal distribution is a kind of the most common Distribution, it is widely used in natural phenomena, social phenomenon.According to central limit theorem, if produced by each determiner Affect the most small, do not have any factor can play inundatory effect, then this stochastic variable just can be close to Step response.Actually in engineer applied, a lot of parameter models are processed as normal distribution characteristic the most simply and study, because it There is universality and the suitability.As it is shown in figure 1, what cloud model produced on the basis of being namely based on this distribution, it is most basic Cloud model is distributed.

Expect Ex: be concept central value in domain, the value of this qualitativing concept can be represented.

Entropy En: weighing ambiguity and the index amount of randomness, ambiguity reflects is subordinate on longitudinal coordinate at qualitativing concept Genus degree uncertain, randomness reflects the most received range size of qualitativing concept.Entropy is the biggest, qualitativing concept The most macroscopical, probabilistic measuring just is more difficult to.

Super entropy He: being the entropy of entropy, it reflects the dispersion degree of water dust, compared with normal distribution, the foundation of this parameter It it is the degree size in order to characterize deviation normal distribution.The value of super entropy is the biggest, represents that it is the biggest with the difference of normal distribution, water dust The most discrete, the randomness of degree of membership is the biggest.

Cloud generator:

Cloud generator is by the algorithm of mutually conversion between qualitativing concept and quantitative value.Mechanism of production and side according to cloud To, there are Normal Cloud Generator and backward cloud generator.Forward cloud is most basic cloud algorithm, the numerical characteristic of cloud produce cloud Drip, i.e. realize the transformation process of the scope from the qualitative information of language performance to quantitative data and the regularity of distribution.As in figure 2 it is shown, Forward cloud is under given YUNSHEN number, produces a point i.e. water dust Drop (x obeying Normal Cloud distribution_i,μ_i)。

The specific algorithm of forward Normal Cloud generator is as follows:

Input quantity: three YUNSHEN number Ex of qualitativing concept A, En, He and water dust sum M；

Output: M water dust and each water dust degree of certainty to concept A.

Step1: generate the normal distribution random number x that obedience (Ex, En) is variance；

Step2: generate the random number En' of the normal distribution that obedience (Ex, He) is variance；

Step3: if the most concrete quantized value that x is concept A；

Step4: making y is the x degree of certainty to concept, calculates y=exp{-(x-Ex)²/2(En')}；

Step5: setting up an office, (x y) is a water dust；

Step6: repeat above step until producing M water dust of number required.

If the given water dust sample (xi, yi) meeting a certain Normal Cloud regularity of distribution, be converted to appropriate qualitative The process of Linguistic Value (Ex, En, He), i.e. conversion from quantitative values to qualitativing concept, this and the input of above-mentioned Normal Cloud Generator With output contrast, referred to as backward cloud generator, as shown in Figure 3.The reverse Normal Cloud generator that this project uses is in nothing Determining foundation on Information base, its specific algorithm is as follows:

Input: the quantitative values xi of M water dust；

Output: the expected value Ex of qualitativing concept, entropy En, super entropy He.

Step1: calculated the sample average of data set by xSingle order sample Absolute Central Moment Sample variance

Step2: can be expected by upper step

Step3: entropy can be obtained by sample average

Step4: can be obtained by the sample variance in step1 and the entropy in step3

The event that acquisition terminal and electric energy meter are generated, by following Rules Filtering:

(1) repeating to report with 1 event, event content includes that the time is the most identical, only carries out main website intelligence by the 1st article Diagnosis, its complementary event is not involved in main website intelligent diagnostics.

(2) repeating to report with 1 class event, event title is identical but content different, period analysis be judged as wrong report, k Main website intelligent diagnostics (k recommended value is 15, it is proposed that arrange scope 5～60) it is no longer participate in it.

(3) reject content and do not meet the event that communications protocol format requires, including data mess code and answer number completion according to for empty Situation.

(4) reject the event that content is the most wrong, be later than early than equipment set-up time and event time including event time Situation (k recommended value is 5, it is proposed that arrange scope 2～30) after current time k days.

(5) for the event that should occur in pairs, if event is not paired, by curve data and terminal heart beating, message is logged in Carry out auxiliary judgment etc. data, such as, can have a power failure interval, according to curve data according to terminal heart beating and login message auxiliary judgment The out-of-limit interval of auxiliary judgment.

For acquisition terminal and the data of electric energy meter, by following Rules Filtering:

(1) positive/negative take advantage of the numerical value of multiplying power to meritorious general power (k recommended value is 50, builds more than k times of user's contract capacity View arranges scope 2～50), belong to abnormal data.

(2) freeze day positive/negative to the calculated electricity of electric energy indicating value, more than user's day maximum power consumption (contract capacity × 24h) k times (k recommended value is 50, it is proposed that arrange scope 2～50), belong to abnormal data.

(3) freeze by the moon positive/negative to the calculated electricity of electric energy indicating value, more than user's moon maximum power consumption (contract capacity × 24h × 30 day) k times (k recommended value is 50, it is proposed that arrange scope 2～50), belong to abnormal data.

(4) day month is freezed maximum demand and is taken advantage of the numerical value of multiplying power (k recommended value is 50, builds more than k times of user's contract capacity View arranges scope 2～50), belong to abnormal data.

(5) group electric flux curve calculated period electricity is always added, more than User window maximum power consumption (contract capacity × Period Length) k times (k recommended value is 50, it is proposed that arrange scope 2～50), belong to abnormal data.

(6) secondary side magnitude of voltage more than rated secondary voltage value k times (k recommended value is 2, it is proposed that arrange scope 2～ 10), abnormal data is belonged to.

(7) secondary side current value take advantage of multiplying power be more than current transformer primary side load current value k times (k recommended value is 2, Suggestion arranges scope 2～10), belong to abnormal data.

According to above-mentioned screening rule, electric energy meter abnormal data can be divided into: electricity abnormity diagnosis, voltage x current are abnormal examines Disconnected, exception electrodiagnosis, load abnormity diagnosis, clock abnormity diagnosis, wiring abnormity diagnosis, expense are controlled abnormity diagnosis and are amounted to 7 classes 29 Individual intelligent diagnostics analyzes model (concrete diagnostic cast is shown in annex one).

Electricity abnormity diagnosis (R1): electric energy meter indicating value uneven (R11), electric energy meter fly away (R12), electric energy meter fall away (R13), Electric energy meter stops walking (R14), electric energy meter rate arranges exception (R15).

Voltage x current abnormity diagnosis (R2): electricity sampling open-phase (R21), voltage out-of-limit (R22), Voltage unbalance (R23), high confession High meter B phase abnormal (R24), electric current defluidization (R25), current imbalance (R26).

Abnormal electrodiagnosis (R3): electric energy meter cover opening (R31), measuring gate opening and closing (R32), stationary magnetic field interference (R33), electricity Measure differential exception (R34), the differential exception of power (R35), power-off event extremely (R36).

Load abnormity diagnosis (R4): the super appearance (R41) of requirement, load super appearance (R42), overcurrent (R43), based model for load duration surpass Lower limit (R44), power factor are abnormal (R45).

Clock abnormity diagnosis (R5): terminal clock abnormal (R51), clock of power meter are abnormal (R52).

Wiring abnormity diagnosis (R6): reversely electricity abnormal (R61), phase sequence abnormal (R62), trend are reversely (R63).

Expense control abnormity diagnosis (R7): expense control issues exception (R71), remaining sum extremely (R72).

The fault data occurred due to scene, it may be possible to caused by the fault of electric energy meter own, or due to external cause (surreptitiously Electricity, electrical network are abnormal) cause, therefore need fault data is carried out mining analysis, thus help staff to solve targetedly Certainly problem.

The electric energy meter anomalous event that this project collects to December main website with District of Shanghai in October, 2015 is recorded as data Source (takes the anomalous event of 500 electric energy meters), is analyzed it.Taking K value is 7, and each event data value is 1 or 0 (1 table Showing that event occurs, 0 expression event does not occurs), set up intelligent electric energy meter evaluation index system, as shown in Figure 4.

For preferably embodying each index significance level in evaluation system and internal relation, index weights need to be introduced.Often Weight Determination have expert graded, feature vector method, analytic hierarchy process (AHP) etc..This project is true by analytic hierarchy process (AHP) Determining the weights of each index in evaluation system, each weights are arranged by adding up data over the years, as shown in table 1.

Table 1 evaluation index weighted value

Based on the 500 groups of data collected, calculate desired value R (being accurate to 0.05), according to electric energy meter in different numerical value districts The quantity in territory, draws index chart of frequency distribution, as shown in Figure 5.

Curve of frequency distribution based on each event of Cloud transform algorithm process, utilizes Cloud transform algorithm to extract the cloud of sample data Concept is distributed, and its principle is expressed as

$\left\{\begin{array}{c}f\left(x\right)\&RightArrow;\underset{i=1}{\overset{K}{\&Sigma;}}\left(a_i{C}_i\right(E_{xi},E_{ni},H_{ei}\left)\right)\\ 0<|f\left(x\right)-\underset{i=1}{\overset{K}{\&Sigma;}}\left(a_i{C}_i\right(E_{xi},E_{ni},H_{ei}\left)\right)|<\mathrm{\&epsiv;}\end{array}\right.$

In formula, f (x) represents frequency distribution function, a_iRepresent range coefficient, C_i(E_xi,E_ni,H_ei) represent certain after Cloud transform Cloud concept, K is the quantity (i.e. Cluster space number) of cloud concept after conversion, and ε represents the error threshold values of conversion.Cloud transform calculation procedure As follows:

(1) all of peak point in the frequency distribution to i event in the fault data of space (as electric energy meter flies away) is found, respectively Peak point abscissa value xj is as the expectation E of cloud concept_xj(j=1,2 ... n).

(2) each cloud concept of digital simulation primary frequency distribution function f (x) (is desired for E respectively_xj) entropy E_nj, and will ask The probability density expectation function of all cloud concepts Cj solved

$fi\left(x\right)=e^{\frac{{(x-E_{xi})}^2}{2{\left(E_{ni}\right)}^2}}$

Based on the reverse cloud algorithm without degree of certainty, the super entropy of each cloud concept is calculated.After Cloud transform, 2 adjacent clouds At a distance of the nearest between concept, even can there is certain cloud concept and be included by another cloud concept, consequent concept repeats And unnecessary meeting divides to the said concepts of subsequent samples data and causes difficult and confusing, therefore need the cloud that above-mentioned steps is obtained general Thought merges and rises to, and makes each cloud concept more independent and clear.

According to the cognition of electric energy meter and the technical standard of IEC60599:2007, and in view of the range coefficient between cloud Impact, the close cloud concept obtained by Cloud transform merges.

Each cloud concept numerical characteristic (Ex, En, He) is as follows: C1 (0.0860,0.0781,0.0039), C2 (0.1741,0.0611,0.0045), C3 (0.3532,0.0475,0.0024), C4 (0.5867,0.0632,0.0051), C5 (0.7911,0.0685,0.0063).Each cloud has a concept qualitatively, according to electric energy meter this project of practical operation situation Each cloud is respectively defined as: error connection, multiplexing electric abnormality, doubtful stealing, battery failure and equipment fault.For visual representation, incite somebody to action Electrical energy meter fault data diagnosis model is summarized as following 5 classes 96 diagnostic casts, is specifically shown in Table 2.

Table 2 electrical energy meter fault data clusters model

Claims (7)

1. an intelligent electric meter Fault Classification based on cluster analysis and cloud model, it is characterised in that comprise the following steps:

1) obtain history intelligent electric meter fault data sample point, and use K-means algorithm by history intelligent electric meter fault data Sample point is divided into K the big class of fault and central value corresponding to each big class；

2) using central value corresponding for each big class as sample average, the intelligent electric meter fault data sample that the big class of each fault comprises This point, as data point, generates corresponding K level electrical energy meter fault cloud model；

3) use reverse Normal Cloud generator that electrical energy meter fault cloud model is calculated, obtain determining of electrical energy meter fault cloud model Property cloud feature；

4) multiple fault group will be divided under K the big class of fault according to qualitative cloud feature.

A kind of intelligent electric meter Fault Classification based on cluster analysis and cloud model the most according to claim 1, it is special Levy and be, described step 1) specifically include following steps:

11) select K history intelligent electric meter fault data sample point as the cluster centre of the big class of primary fault；

12) to remaining history intelligent electric meter fault data sample point, according to itself and the distance of each cluster centre, concluded Enter the big class of closest fault；

13) calculate the sample average of the big class of each fault, and calculate canonical measure function, it may be assumed that

$E=\underset{i=1}{\overset{K}{\&Sigma;}}\underset{\mathrm{\&tau;}\&Element;Ci}{\&Sigma;}|\mathrm{\&tau;}-m_i{|}^2$

$m_i=\frac{1}{N_i}\underset{\mathrm{\&tau;}\&Element;Ci}{\&Sigma;}\mathrm{\&tau;}$

In formula, m_iFor the electrical energy meter fault data sample average of i-th bunch, C_iElectrical energy meter fault data sample point for i-th bunch Set, N_iFor the electrical energy meter fault data sample sum of i-th bunch, E is canonical measure function, and τ is electrical energy meter fault data sample This point；

14) replace initial cluster center with the sample average of the big class of each fault, repeat step 12)～14), until canonical measure Function convergence.

A kind of intelligent electric meter Fault Classification based on cluster analysis and cloud model the most according to claim 1, it is special Levy and be, described step 3) in, by the sample average of electrical energy meter fault cloud modelSingle order sample Absolute Central Moment T and sample This variance S²Respectively as reverse Normal Cloud generator input value, i.e. expected value Ex, entropy En and super entropy He.

A kind of intelligent electric meter Fault Classification based on cluster analysis and cloud model the most according to claim 3, it is special Levy and be, described sample averageSingle order sample Absolute Central Moment T and sample variance S²Calculating formula be:

$\stackrel{\&OverBar;}{X}=\frac{1}{n}\underset{i=1}{\overset{n}{\&Sigma;}}{x}_i$

$T=\frac{1}{n}\underset{i=1}{\overset{n}{\&Sigma;}}|x_i-\stackrel{\&OverBar;}{X}|$

$S^2=\frac{1}{n-1}\underset{i=1}{\overset{n}{\&Sigma;}}{(x_i-\stackrel{\&OverBar;}{X})}^2$

Wherein, n is the quantity of the data point that the big class of fault comprises, x_iValue for i-th data point.

A kind of intelligent electric meter Fault Classification based on cluster analysis and cloud model the most according to claim 4, it is special Levying and be, described expected value Ex, entropy En and the calculating formula of super entropy He be:

$\begin{array}{c}Ex=\stackrel{\&OverBar;}{X}\\ En=\sqrt{\frac{\mathrm{\&pi;}}{2}}\&times;\frac{1}{n}\underset{i=1}{\overset{n}{\&Sigma;}}|x_i-Ex|\end{array}$

$He=\sqrt{S^2-{\mathrm{En}}^2}.$

A kind of intelligent electric meter Fault Classification based on cluster analysis and cloud model the most according to claim 1, it is special Levy and be, described step 4) in, K the big class of fault includes error connection, multiplexing electric abnormality, stealing, battery failure and equipment fault.

A kind of intelligent electric meter Fault Classification based on cluster analysis and cloud model the most according to claim 6, it is special Levy and be, in described step 4) fault group under K the big class of fault is categorized as: