CN103777091A - High-speed rail electric energy quality monitoring data classification method based on K mean value - Google Patents

Abstract

The invention brings forward a high-speed rail electric energy quality monitoring data classification method based on a K mean value. The method comprises, first of all, separating electric energy quality data of only one high-speed rail on a power supply arm, then by using operation time of the high-speed rail and the statistical value of an electric energy quality indicator as sample data, performing normalization, carrying out K mean value clustering, and finally obtaining classification of electric energy quality monitoring data of different models. According to electric energy quality monitoring data of a high-speed rail traction station, the invention provides a method for easily and rapidly separating the electric energy quality data of only one high-speed rail on a certain power supply arm; by use of a K mean value cluster algorithm, electric energy quality data classification of high-speed rail models is finished; and not only can a basis be provided for research on the electric energy quality problem brought by high-speed rails of a certain type to a power network, the electric energy quality monitoring data classification can also be carried out for various conditions, and a basis is provided for further refining the electric energy quality characteristic analysis of a high-speed rail load.

Description

A kind of high ferro electric energy quality monitoring data classification method based on K average

Technical field

The invention belongs to the power quality data analysis field of high ferro, provide and relate to a kind of high ferro electric energy quality monitoring data classification method based on K average.

Background technology

High-speed railway with energy-saving and environmental protection, the feature such as efficient, safe, comfortable, quick, punctual in China's fast development, the trip that is changing people and life that it is just making rapid progress.But high-speed railway locomotive power is large, road speed is high, bring certain impact to the quality of power supply of the electrical network along the line of undertaking its power supply task.

China Express Railway adopts the operation mode of AC-DC-AC high-power locomotive, high-speed (more than highest running speed 350km/h), high density (minimum tracking interval 3min), large marshalling (maximum 16 marshallings); The high-speed railway particularly Transportation Model of Beijing-Shanghai High-Speed Railway is taked bullet train and the mixed pattern of running of overline train, singularity and the complicacy of this decision maker's high-speed railway traction load.Electrified high-speed railway locomotive belongs to non-linear and impact load, and it mainly brings the impact of the quality of power supply such as negative phase-sequence and harmonic wave to electrical network; Cause the quality of power supply characteristic difference, particularly harmonic characteristic of electrical network when the locomotive operation of different automobile types simultaneously.These not only cause adverse effect to miscellaneous equipment in electrical network, and stability and reliability to self-operating constitutes a threat to.The electric locomotive of electric railway is the high-power single-phase rectification load that undulatory property is very large, and due to the various states of the acceleration of train in operational process, coasting, braking, and the variation of train quantity on the factor such as line slope, turning radius, meteorological condition, driver operation and supply arm, traction load random fluctuation.Therefore,, in the time carrying out the analysis of electric energy metrical, must fully take into account the part throttle characteristics of electric locomotive.Adopt different friendships-straight electric locomotive, the harmonic content difference of its generation.In order further to study the quality of power supply characteristic after certain vehicle puts into operation, assess the impact that it brings to electrical network, the present invention proposes a kind of method of for vehicle, electric energy quality monitoring data having been carried out sort research.

Summary of the invention

In order to overcome the defect of prior art, the object of the invention is to propose a kind of high ferro electric energy quality monitoring data classification method based on K average, first the method is isolated the power quality data that only has a high ferro operation on supply arm, then working time, power quality index and the statistical value thereof that utilizes high ferro as sample data normalization after, carry out K mean cluster, finally obtain the classification of the electric energy quality monitoring data of different automobile types.This invention is the quality of power supply characteristic after further studying certain vehicle and putting into operation, and assesses the impact that it brings to electrical network, and then takes positive counter-measure for electrical network and provide the foundation.

The present invention is achieved through the following technical solutions:

A high ferro electric energy quality monitoring data classification method based on K average, is characterized in that, the method comprises the steps:

(1) obtain the electric energy quality monitoring data of high ferro Traction Station by online electric energy quality monitor;

(2) electric energy quality monitoring data when foundation has car operation and high ferro, in the working time of supply arm, filter out the power quality data that only has a car operation on supply arm;

(3) for the power quality data that only has a car operation on supply arm, calculate the working time of a high ferro and the statistical value of power quality index;

(4) using the statistical value of the working time in step (3) and power quality index as sample data, and be normalized;

(5) sample data after being normalized is carried out to K mean cluster, to obtain the electric energy quality monitoring Data classification of different automobile types.

Further, in step (1), the electric energy quality monitoring data of described high ferro Traction Station comprise voltage deviation, electric current, negative-sequence current, frequency departure, active power, reactive power, electric weight, harmonic wave, a harmonic wave, phase angle, voltage fluctuation and flicker, tri-phase unbalance factor, voltage swell, voltage dip and short time voltage interruption etc.

Further, in step (2), described in have the electric energy quality monitoring data in car when operation to comprise active power, negative-sequence current and phase angle etc.

Further, in step (2), described in the power quality data that filters out while only having a car operation on supply arm comprise the steps:

(a), when the electric energy quality monitoring data in the time having car operation are greater than pre-set threshold value simultaneously, being considered as supply arm has the moment of car, has the moment of car to sort from front to back supply arm and obtains the time series of car period;

(b) according to there being car time series, calculate each time span that has continuously the car period;

(c) judge each time span that has continuously a car period and the size of short transit time, to there being continuously the car period to divide;

(d) judge each time span that has continuously a car period and the size of long transit time, by have continuously the car period be divided into the operation of many cars have the car period and only have a car operation have a car period, and then draw the power quality data while only having a car to move on supply arm.

Further, in step (3), described power quality index comprises content and the number of times etc. of active power, reactive power, electric current, negative-sequence current and harmonic wave; The statistical value of described power quality index comprises maximal value, minimum value, average and the variance etc. of power quality index.

Further, in step (3), the working time of a described high ferro is according to described corresponding of power quality data while only having the operation of car, only the absolute difference in moment obtains;

Suppose that power quality index p is P by the vector of Time alignment, P=[p (1), p (2), ..., p (i) ..., p (n)], wherein p (i) and p (n) are respectively the power quality index of i and n sampling instant, and the maximal value of described power quality index index p is p _max=max (P), minimum value are p _min=min (P), mean value are

variance is

Further, in step (4), the statistical value of the working time of a high ferro and power quality index is mapped to [0 ,+1] interval, sets X ' _i=(x ' _{i, l}, x ' _{i, 2}..., x ' _{i, j}... x ' _{i, M}) be i sample data, tool is M eigenwert altogether, by following formula, sample data is normalized to calculating:

$x_{i,j}=\frac{x_{i,j}^{\′}-\underset{j=1,...,m}{\min }\left(x_{i,j}^{\′}\right)}{\underset{j=1,...,m}{\max }\left(x_{i,j}^{\′}\right)-\underset{j=1,...,m}{\min }\left(x_{i,j}^{\′}\right)}$

In formula, x ' _{i, j}j eigenwert of i sample data, x _{i, j}it is the normalized value of j feature of i sample data.

Further, in step (5), K-mean cluster process is as follows:

(a) from sample data, choose 4 samples, as the initial cluster center of 4 bunches;

(b) sample data is distributed to the most contiguous cluster centre according to minimal distance principle;

(c), according to cluster result, recalculate sample average in each cluster as new cluster centre;

(d) repeating step b and c until cluster centre no longer change.

(e) cluster finishes, and obtains 4 bunches.

Compared with prior art, beneficial effect of the present invention is:

1) the present invention is according to high ferro Traction Station electric energy quality monitoring data, provides a kind of simple and efficient by the method that only has the power quality data of a high ferro operation to separate on certain supply arm; Utilize K means clustering algorithm, completed the power quality data classification for high ferro vehicle.

2) the present invention is that the power quality problem bringing to electrical network in certain type high ferro operational process of research provides the foundation.In addition, institute of the present invention extracting method also can carry out electric energy quality monitoring Data classification for operating mode (comprising load braking, startup, stable operation etc.), for the quality of power supply specificity analysis of the further high iron load of refinement provides basis.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the classification of the electric energy quality monitoring data for high ferro vehicle of the present invention;

Fig. 2 is the power quality data classification process figure while only having a car operation on supply arm of the present invention.

Embodiment

Below in conjunction with accompanying drawing, the specific embodiment of the present invention is described in further detail.

In this example, set forth a kind of high ferro electric energy quality monitoring data classification method based on K average, its process flow diagram as shown in Figure 1, comprises the steps:

(1) obtain the electric energy quality monitoring data of high ferro Traction Station by online electric energy quality monitor.

High ferro traction power supply station electric energy quality monitoring packet is containing active power (P), reactive power (Q), electric current (I), harmonic wave (I _h, I ₁represent fundamental current), a harmonic wave (I _ih), voltage (merit, flickering (P _st), negative phase-sequence (μ ₂) etc. index.

(2) electric energy quality monitoring data when foundation has car operation and high ferro, at the transit time of supply arm, filter out the power quality data that only has a car operation on supply arm, and its specific implementation process as shown in Figure 2.

First I the present invention has the car period and carries out preliminary classification without the power quality data of car period supply arm for indexs such as active power, negative-sequence current and the phase angles of the quality of power supply.

According to there being high ferro to enter after supply arm, using some index of Monitoring Data such as power, negative-sequence current and phase angle are greater than predetermined threshold value as criterion simultaneously, obtain supply arm moment t1 (the i) (i=1 that takes up an official post, 2, n, n represents sampled point number) have a car situation: have car y=1 and without car y=0.The time series t (j) of car period will there be is moment corresponding to car y=1 situation to be arranged in order to be formed as.

Then II according to there being car time series, calculates each time span that has continuously the car period.

After utilizing, be a bit whether t (j+1)-t (j) is greater than 10s as criterion with the more front mistiming, find and on supply arm, have continuously the initial time t_b of car period and stop moment t_e.Detailed process is as follows: if t (j+1)-t (j) is <10s, j=j+1, enters next point and judge; If t (j+1)-t (j) >=10s, think that t (j) is n and has continuously the termination moment t_e (n) of car period, t (j+1) is n+1 the initial time t_b (n+1) of car period continuously.

Therefore, there is continuously the difference between the initial time t_b (n+1) of car period can try to achieve each time span that has continuously the car period according to a certain termination moment t_e (n) that has continuously a car period and this.

III has the division of car period continuously.

High ferro the normal transit time of supply arm be probably 4min between 6min, the shortest transit time is 4min, the longest run time is 6min.In order to obtain the more accurate complete car period that has continuously, the present invention utilizes the shortest transit time 4min of high ferro on supply arm further to judge.Concrete operations are as follows: calculating n has the time span t_interal=t_e-t_b of car period, if t_interal >=4min, representing this n, to have the car period be complete; If t_interal<4min, connect thereafter have a car period, form the new car period that has, until this new time span that has the car period is greater than the shortest transit time 4min, what this was new like this have, and the car period is also complete has a car period.

IV finally by have continuously the car period be divided into the operation of many cars have the car period and only have that a car moves have a car period.

In order to obtain only having the power quality data of a high ferro operation, the present invention utilizes the maximum duration of single high ferro operation as criterion, if there is the time span > 6min of car period, that thinks many cars operations has a car period; If there is time span≤6min of car period, that thinks the operation of car has a car period, and the power quality data that this period is corresponding, is exactly the power quality data that only has a high ferro operation that the present invention will obtain.

(3) power quality data when only having the operation of high ferro on supply arm, calculates working time, power quality index and the statistical value thereof of a high ferro.

Wherein can subtract each other acquisition according to the obtained corresponding start-stop of the power quality data moment that only has a high ferro operation working time of a car.Power quality index mainly refers to the indexs such as power, electric current, harmonic content, overtone order, negative-sequence current.The statistical value of power quality index comprises maximal value, minimum value, average, variance of power quality index etc.What suppose power index p is P by the vector of Time alignment, P=[p (1), p (2), ..., p (i) ..., p (n)], wherein p (i) and p (n) are respectively the power quality index value of i and n sampling instant, and the maximal value of this index p is p _max=max (P), minimum value are p _min=min (P), mean value are $p_{\mathrm{mean}}=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}p\left(i\right)/n,$ Variance is $p_{\mathrm{var}}=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{(p\left(i\right)-p_{\mathrm{mean}})}^2}{n-1}.$

(4) using working time, power quality index and statistical value thereof as sample data, be normalized.

For improving the generalization ability of model, the time of minimizing procedural training, sample data is carried out to normalized.In this example, the working time of all high ferros, power quality index and statistical value thereof are mapped to [0 ,+1] interval, suppose X ' _i=(x ' _{i, 1}x ' _{i, 2}' ..., x ' _{i, m}) be i sample data, there is x ' _{i, 1}, x ' _{i, 2}..., x ' _{i, m}m dimension variable, i.e. m eigenwert altogether.

Normalization formula is: $x_{i,j}=\frac{x_{i,j}^{\&prime;}-\underset{j=1,...,m}{\min }\left(x_{i,j}^{\&prime;}\right)}{\underset{j=1,...,m}{\max }\left(x_{i,j}^{\&prime;}\right)-\underset{j=1,...,m}{\min }\left(x_{i,j}^{\&prime;}\right)};$

In above formula, x ' _i,jj eigenwert of i sample data, x _{i, j}be the normalized value of j feature of i sample data, i sample data after normalization is X _i=x _{i, 1}, x _{i, 2}..., x _{i, m}).

(5) carry out K-mean cluster

(5a), according to vehicle: bullet train 8 is organized into groups, bullet train 16 is organized into groups, medium trains 8 is organized into groups and medium trains 16 is organized into groups, the power quality data during by car operation is only divided into four classes;

(5b) choose 4 initial cluster center V _i(1), i=1,2 ..., the sequence number of 4, i representation class, parenthetic 1 represents primary iteration operation times;

For example: 4 initial cluster centers can be chosen to obtain comparatively according to actual conditions and disperse, such as 16 marshaling power are generally 2 times of 8 marshaling power, so just can go out to select initial cluster center at 1 times power and 2 times powers respectively.

(5c) one by one by data sample (X to be clustered ₁, X ₂..., X _j..., X _n) distribute to the some V in 4 cluster centres by minimum distance criterion _i(t).Calculate j data sample X _jwith 4 cluster centre V _i(t), i=1,2 ..., 4 distance d (X _j, V _i(t)), i=1,2,3,4, if d is (X _j, V _i(t)) minimum, X _jbe i class, be designated as X ' _j;

Wherein, X _jand X _nrepresent a j and N data sample; T represents interative computation number of times, is initially 1; The sequence number of i representation class, i=1,2 ..., 4;

(5d) calculate 4 new cluster centres: .C _irepresent the number of samples of i class.

If (5e) V _i(t+1)=V _i(t), i=1,2 ..., 4, algorithm finishes, and the most at last sample data to gather be four classes, obtain four class sample datas; Otherwise t=t+1, returns to step (5c).

(6) the four class cluster sample datas that obtain in step (5) are exactly the electric energy quality monitoring data of four kinds of different automobile types.

Finally should be noted that: above embodiment is only in order to illustrate that technical scheme of the present invention is not intended to limit, although the present invention is had been described in detail with reference to above-described embodiment, those of ordinary skill in the field are to be understood that: still can modify or be equal to replacement the specific embodiment of the present invention, and do not depart from any modification of spirit and scope of the invention or be equal to replacement, it all should be encompassed in the middle of claim scope of the present invention.

Claims (11)

1. the high ferro electric energy quality monitoring data classification method based on K average, is characterized in that, the method comprises the steps:

(1) obtain the electric energy quality monitoring data of high ferro Traction Station by online electric energy quality monitor;

(2) electric energy quality monitoring data when foundation has car operation and high ferro, in the working time of supply arm, filter out the power quality data that only has a car operation on supply arm;

(3) for the power quality data that only has a car operation on supply arm, calculate the working time of a high ferro and the statistical value of power quality index;

(4) using the statistical value of the working time in step (3) and power quality index as sample data, and be normalized;

(5) sample data after being normalized is carried out to K mean cluster, to obtain the electric energy quality monitoring Data classification of different automobile types.

2. the high ferro electric energy quality monitoring data classification method based on K average according to claim 1, it is characterized in that, in step (1), the electric energy quality monitoring data of described high ferro Traction Station comprise voltage deviation, electric current, negative-sequence current, frequency departure, active power, reactive power, electric weight, harmonic wave, a harmonic wave, phase angle, voltage fluctuation and flicker, tri-phase unbalance factor, voltage swell, voltage dip and short time voltage interruption.

3. the high ferro electric energy quality monitoring data classification method based on K average according to claim 1, is characterized in that, in step (2), described in have the electric energy quality monitoring data in car when operation to comprise active power, negative-sequence current and phase angle.

4. the high ferro electric energy quality monitoring data classification method based on K average according to claim 1, is characterized in that, in step (2), described in the power quality data that filters out while only having a car operation on supply arm comprise the steps:

(a), when the electric energy quality monitoring data in the time having car operation are greater than pre-set threshold value simultaneously, being considered as supply arm has the moment of car, has the moment of car to sort from front to back supply arm and obtains the time series of car period;

(b) according to there being car time series, calculate each time span that has continuously the car period;

(c) judge each time span that has continuously a car period and the size of short transit time, to there being continuously the car period to divide;

(d) judge each time span that has continuously a car period and the size of long transit time, by have continuously the car period be divided into the operation of many cars have the car period and only have a car operation have a car period, and then draw the power quality data while only having a car to move on supply arm.

5. the high ferro electric energy quality monitoring data classification method based on K average according to claim 4, is characterized in that, in step (b), calculates each time span that has continuously the car period and comprises the steps:

After judgement has in car time series in adjacent two moment, whether the mistiming of a moment and previous moment is greater than 10s;

As <10s, enter with next moment in a rear moment and judge;

As >=10s, previous moment being considered as to this has the termination moment of car period continuously, and a rear moment is considered as to next the initial time of car period continuously, determines each time span that has continuously the car period with this.

6. the high ferro electric energy quality monitoring data classification method based on K average according to claim 4, is characterized in that, in step (c), comprises the steps: there being continuously the car period to divide

Judge each time span that has continuously a car period and the size of short transit time;

As >=the shortest transit time, represent that this has the car period is continuously complete;

Transit time as the shortest in <, has this car period and rear one to have continuously the car period to form the new car period that has continuously, until this new time span that has continuously the car period is more than or equal to the shortest transit time continuously.

7. the high ferro electric energy quality monitoring data classification method based on K average according to claim 4, it is characterized in that, in step (d), there is continuously the car period to be divided into there is the car period and only has the car period that has of a car operation to comprise the steps: of many car operations by described

Judge each time span that has continuously a car period and the size of long transit time;

Transit time as the longest in >, that thinks the operation of many cars has a car period;

As≤the longest transit time, that thinks only to have a car operation has a car period, and power quality data corresponding to this period is the power quality data while only having a car operation.

8. the high ferro electric energy quality monitoring data classification method based on K average according to claim 1, it is characterized in that, in step (3), described power quality index comprises content and the number of times of active power, reactive power, electric current, negative-sequence current and harmonic wave; The statistical value of described power quality index comprises maximal value, minimum value, average and the variance of power quality index.

9. the high ferro electric energy quality monitoring data classification method based on K average according to claim 8, it is characterized in that, in step (3), the working time of a described high ferro is according to described corresponding of power quality data while only having the operation of car, only the absolute difference in moment obtains;

Suppose that power quality index p is P by the vector of Time alignment, P=[p (1), p (2), ..., p (i) ..., p (n)], wherein p (i) and P (n) are respectively the power quality index of i and n sampling instant, and the maximal value of described power quality index index p is p _max=max (P), minimum value are p _min=min (P), mean value are variance is

10. the high ferro electric energy quality monitoring data classification method based on K average according to claim 1, is characterized in that, in step (4), the statistical value of the working time of a high ferro and power quality index is mapped to [0 ,+1] interval, sets X ' _j=(x ' _{i, 1}x ' _{i, 2}..., x ' _{i, j}..., x ' _{i, M}) be i sample data, tool is M eigenwert altogether, by following formula, sample data is normalized to calculating:

$x_{i,j}=\frac{x_{i,j}^{\&prime;}-\underset{j=1,...,m}{\min }\left(x_{i,j}^{\&prime;}\right)}{\underset{j=1,...,m}{\max }\left(x_{i,j}^{\&prime;}\right)-\underset{j=1,...,m}{\min }\left(x_{i,j}^{\&prime;}\right)}$

In formula, x ' _{i, j}j eigenwert of i sample data, x _{i, j}it is the normalized value of j feature of i sample data.

The 11. high ferro electric energy quality monitoring data classification methods based on K average according to claim 1, is characterized in that, in step (5), the method for the sample data after being normalized being carried out to K mean cluster comprises the steps:

(a) according to vehicle, the power quality data when only having the operation of car is divided into four classes, and described vehicle comprises that bullet train 8 is organized into groups, bullet train 16 is organized into groups, medium trains 8 is organized into groups and medium trains 16 is organized into groups;

(b) choose at random initialization cluster centre V _i(1);

Wherein, i=1,2 ..., the sequence number of 4, i representation class; 1 represents primary iteration operation times;

(c) one by one by sample data (X to be clustered ₁, X ₂..., X _j..., X _n) give the some V in 4 cluster centres according to following minor increment regular allocation _i(t), described minor increment rule is:

Calculate j sample data X _jwith 4 cluster centre V _i(t), i=1,2 ..., 4 distance d (X _j, V _i(t)), i=1,2,3,4, if d is (X _j, V _i(t)) minimum, X _jbe i class, be designated as X ' _j;

Wherein, X _jand X _nrepresent a j and N sample data; T represents interative computation number of times, is initially 1; The sequence number of i representation class, i=1,2 ..., 4;

(d) calculate 4 new cluster centres by following formula:

$V_i(t+1)=\frac{1}{\left|C_i\right|}\underset{k=1}{\overset{\left|C_i\right|}{\mathrm{\&Sigma;}}}{X}_k^i,i=\mathrm{1,2},...,4$

Wherein, C _irepresent the number of i class sample data;

(e) if V _i(t+1)=V _i(t), i=1,2 ..., 4, this K means clustering algorithm finishes, and sample data is gathered is four classes; Otherwise, make interative computation number of times t=t+1, jump to step (c).

Images