CN105426991A - Transformer defect prediction method and transformer defect prediction system - Google Patents

Abstract

The invention provides a transformer defect prediction method. The method comprises the following steps of: determining a to-be-predicted month, and in transformer historical data, starting from previous month of the to-be-predicted month to select actual measurement defect rates respectively corresponding to a plurality of sequential months according to a time inverted sequence; according to a preset method, executing seasonal decomposition for an original defect rate sequence, obtaining corresponding time sequence of each mode component, analyzing the corresponding time sequence of each mode component, screening out a model meeting a certain condition from each mode component, and obtaining a value of each mode component corresponding to the screened model thereof; accumulating the value of each mode component corresponding to the screened model thereof, and taking the accumulated sum as a predicted defect rate of the to-be-predicted month. By implementation of the method and the system, different feature information can be separated, modeling difficulty can be reduced, and prediction accuracy of transformer defect rate can be improved.

Description

A kind of method and system of transformer ratio of defects prediction

Technical field

The present invention relates to transformer detection technique field, particularly relate to the method and system of a kind of transformer ratio of defects prediction.

Background technology

The reliability service of transformer is related to the safety of whole electrical network, because transformer device structure is complicated, after long-time load carrying, inevitably having the generation of various defect, is fault once above-mentioned development of defects, huge economic loss and social influence will be caused, therefore need to carry out degree of depth excavation to above-mentioned defect, scientificlly and effectively predict defect probability of happening, formulate measure targetedly in time, reasonable arrangement O&M strength, to the safe important in inhibiting of support equipment.

Due to transformer defect generation and equipment quality, to supervise level of making, mounting process, O&M measure, environmental factor, climate effect etc. relevant, make defective data have stronger randomness, contain abundant characteristic information simultaneously.

In prior art, transformer ratio of defects Forecasting Methodology is mainly divided three classes: traditional mathematics modelling (as regression analysis), nonlinear model method (as artificial neural network, fuzzy neural network) and the modelling (as wavelet analysis method) based on structure.

Inventor finds, the equal Shortcomings part of above-mentioned three class Forecasting Methodology: in first kind method, is applicable to the prediction of rule change ordered series of numbers, unpredictable to the data of unconventional change ordered series of numbers; In Equations of The Second Kind method, although be applicable to the prediction of unconventional change ordered series of numbers, modeling difficulty is higher, and algorithm is complicated; In the 3rd class methods, simply predict, but in model, the degree of correlation of each characteristic information is more weak although be applicable to multiple dimensioned ordered series of numbers relation, precision of prediction needs to improve further.

Therefore, need the method for a kind of transformer ratio of defects prediction badly, different characteristic information separated can be come, the precision of prediction of modeling difficulty and raising transformer ratio of defects can be reduced.

Summary of the invention

Embodiment of the present invention technical matters to be solved is, provides the method and system that a kind of transformer ratio of defects is predicted, different characteristic information separated can be come, can reduce modeling difficulty, and improve the precision of prediction of transformer ratio of defects.

In order to solve the problems of the technologies described above, embodiments provide the method for a kind of transformer ratio of defects prediction, described method comprises:

A, determine month to be measured, and in transformer historical data, chose continuous multiple month for starting point according to the mode of time inverted sequence with the upper January in described month to be measured and distinguish corresponding actual measurement ratio of defects, and the described multiple actual measurement ratio of defects got all are formed genetic defects rate sequence as sample;

The method that b, basis are preset, carry out seasonality to described genetic defects rate sequence to decompose, obtain the time series that each mode component is corresponding respectively, and the time series that described each mode component obtained is corresponding is respectively analyzed, in each mode component, all filter out the model meeting certain condition, and obtain the value of each mode component its screening model corresponding;

C, the value of its screening model corresponding for described each mode component obtained to be added up, and using the ratio of defects of described cumulative sum as described month prediction to be measured.

Wherein, described step b specifically comprises:

According to the seasonal adjustment method preset, seasonality is carried out to described genetic defects rate sequence and decomposes, obtain the time series that multiple mode component is corresponding respectively;

The time series corresponding respectively to each mode component is analyzed, and obtains difference auto regressive moving average ARIMA (p, d, the q) model of each mode component; Wherein, p is autoregression item; Q is moving average item number, the difference number of times of d for doing when time series becomes steady;

At the ARIMA (p of described each mode component obtained, d, q), in model, all adopt correlogram and partial correlation figure process, realize carrying out parameter estimation to p, d, q, determine the ARIMA (p of each mode component described, d, q) meet the value corresponding respectively of p, d, q under certain condition in model, and obtain the ARIMA (p meeting certain condition in each mode component described, d, q) value that model is corresponding.

Wherein, described multiple mode component has three, comprises Seasonal component, trend and cyclical component and random fluctuation component.

The embodiment of the present invention additionally provides the system of a kind of transformer ratio of defects prediction, and described system comprises:

Genetic defects rate retrieval unit, for determining month to be measured, and in transformer historical data, chose continuous multiple month for starting point according to the mode of time inverted sequence with the upper January in described month to be measured and distinguish corresponding actual measurement ratio of defects, and the described multiple actual measurement ratio of defects got all are formed genetic defects rate sequence as sample;

Series Decomposition and analytic unit, for the method that basis is preset, carry out seasonality to described genetic defects rate sequence to decompose, obtain the time series that each mode component is corresponding respectively, and the time series that described each mode component obtained is corresponding is respectively analyzed, in each mode component, all filter out the model meeting certain condition, and obtain the value of each mode component its screening model corresponding;

Ratio of defects predicted value acquiring unit, for adding up the value of described each mode component obtained its screening model corresponding, and using the ratio of defects of described cumulative sum as described month prediction to be measured.

Wherein, described Series Decomposition and analytic unit comprise:

Series Decomposition module, for according to the seasonal adjustment method preset, carries out seasonality to described genetic defects rate sequence and decomposes, obtain the time series that multiple mode component is corresponding respectively;

Series analysis model builds module, analyzes, obtain difference auto regressive moving average ARIMA (p, d, the q) model of each mode component for the time series corresponding respectively to each mode component; Wherein, p is autoregression item; Q is moving average item number, the difference number of times of d for doing when time series becomes steady;

Predicted value acquisition module, for in ARIMA (p, d, the q) model of described each mode component obtained, all adopt correlogram and partial correlation figure process, realize carrying out parameter estimation to p, d, q, determine ARIMA (p, the d of each mode component described, q) under meeting certain condition in model, p, d, q distinguish corresponding value, and the value that ARIMA (p, d, the q) model obtaining meeting in each mode component described certain condition is corresponding.

Wherein, described multiple mode component has three, comprises Seasonal component, trend and cyclical component and random fluctuation component.

Implement the embodiment of the present invention, there is following beneficial effect:

In embodiments of the present invention, because the genetic defects rate sequence formed by transformer historical data carries out the mode that seasonality is decomposed and Time Series AR IMA model prediction combines, the ratio of defects treating moon sight part is predicted, both simplify model and again reduce interference in model between different component and coupling, thus different characteristic information separated can be come, reduce modeling difficulty, the comparatively conventional time series method that predicts the outcome is more accurate, reaches the object improving precision of prediction.

Accompanying drawing explanation

In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, be briefly described to the accompanying drawing used required in embodiment or description of the prior art below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the prerequisite not paying creative work, the accompanying drawing obtaining other according to these accompanying drawings still belongs to category of the present invention.

The process flow diagram of the method for a kind of transformer ratio of defects prediction that Fig. 1 provides for the embodiment of the present invention;

The oscillogram that in the method application scenarios of a kind of transformer ratio of defects prediction that Fig. 2 provides for the embodiment of the present invention, original ratio of defects sequence is formed;

The oscillogram that in the method application scenarios of a kind of transformer ratio of defects prediction that Fig. 3 provides for the embodiment of the present invention, Seasonal sequence is formed;

The oscillogram that in the method application scenarios of a kind of transformer ratio of defects prediction that Fig. 4 provides for the embodiment of the present invention, trend and cyclic sequence are formed;

The oscillogram that in the method application scenarios of a kind of transformer ratio of defects prediction that Fig. 5 provides for the embodiment of the present invention, random fluctuation vector sequence is formed;

The comparison of wave shape figure that ratio of defects is formed with actual measurement ratio of defects is predicted in the method application scenarios of a kind of transformer ratio of defects prediction that Fig. 6 provides for the embodiment of the present invention;

The structural representation of the system of a kind of transformer ratio of defects prediction that Fig. 7 provides for the embodiment of the present invention.

Embodiment

For making the object, technical solutions and advantages of the present invention clearly, below in conjunction with accompanying drawing, the present invention is described in further detail.

As shown in Figure 1, in the embodiment of the present invention, the method for a kind of transformer ratio of defects prediction provided, described method comprises:

Step S1, determine month to be measured, and in transformer historical data, chose continuous multiple month for starting point according to the mode of time inverted sequence with the upper January in described month to be measured and distinguish corresponding actual measurement ratio of defects, and the described multiple actual measurement ratio of defects got all are formed genetic defects rate sequence as sample;

Detailed process is, in actual motion, because the working environment of transformer is more severe, be easier to occur the problems such as instantaneous electromagnetic interference (EMI) and communication failure, cause the data of transformer monitoring easily to be polluted, thus cause follow-up insignificant off-line analysis, increase non-productive work amount and the various resource of waste, therefore choose the multiple continuous month data analysis before month to be measured, be conducive to improving the degree of accuracy predicted the outcome.

In embodiments of the present invention, with the upper January in month to be measured for starting point, continuous multiple month is chosen according to the mode of time inverted sequence, in transformer historical data, the actual measurement ratio of defects that selected by extracting, month is corresponding is respectively analyzed as sample, makes the data in genetic defects rate sequence and actual measurement ratio of defects one_to_one corresponding.

The method that step S2, basis are preset, carry out seasonality to described genetic defects rate sequence to decompose, obtain the time series that each mode component is corresponding respectively, and the time series that described each mode component obtained is corresponding is respectively analyzed, in each mode component, all filter out the model meeting certain condition, and obtain the value of each mode component its screening model corresponding;

Detailed process is, according to the seasonal adjustment method preset, carries out seasonality and decomposes, obtain the time series that multiple mode component is corresponding respectively to genetic defects rate sequence; In one embodiment, mode component has three, comprises Seasonal component, trend and cyclical component and random fluctuation component;

The time series corresponding respectively to each mode component is analyzed, and obtains difference auto regressive moving average ARIMA (p, d, the q) model of each mode component; Wherein, p is autoregression item; Q is moving average item number, the difference number of times of d for doing when time series becomes steady;

At the ARIMA (p of each mode component obtained, d, q), in model, all adopt correlogram and partial correlation figure process, realize carrying out parameter estimation to p, d, q, determine the ARIMA (p of each mode component, d, q) meet the value corresponding respectively of p, d, q under certain condition in model, and obtain the ARIMA (p meeting certain condition in each mode component, d, q) value that model is corresponding.

In one embodiment, the method preset of employing is seasonal adjustment method, and this adjusting method is the X-11 seasonal adjustment method of the Census Bureau of US Department of Commerce research and development, and based on the addition model of the method for moving average, performing step is specific as follows:

By the n subcenter method of moving average, calculate the Trendline item of estimation item irregular with season again by " 3*3 " and " 3*5 " Henderson method of moving average, calculate the item in season of estimation and after seasonal adjustment then, after iteration twice, seasonal each component decomposed can be obtained, as Seasonal component, trend and cyclical component and random fluctuation component.

Wherein, seasonal because of component, shown in (1):

$\begin{array}{c}{S}_t^{\left(2\right)}={\hat{S}}_t^{\left(2\right)}-({\hat{S}}_{t-T/2}^{\left(2\right)}+2{\hat{S}}_{t-T/2-1}^{\left(2\right)}+\\ \mathrm{...}+{\widehat{2S}}_{t+T/2-1}^{\left(2\right)}+{\hat{S}}_{t+T/2}^{\left(2\right)}/2T\end{array}---\left(1\right)$

Trend and cyclical component, shown in (2):

${\mathrm{TC}}_t^{\left(3\right)}=\underset{j=-H}{\overset{H}{\Σ}}{h}_j^{2H+1}{\mathrm{TCI}}_{t+j}^{\left(2\right)}---\left(2\right)$

In formula (2), for the flexible strategy of 2H+1 item Henderson ,-H≤J≤H, T is the Seasonal time series cycle.

Random fluctuation component, shown in (3):

$I_t^{\left(3\right)}={\mathrm{TCI}}_t^{\left(2\right)}-{\mathrm{TC}}_t^{\left(3\right)}---\left(3\right)$

Due to the difference of the product quality of transformer, operating condition and maintenance levels, the health status between each transformer, degree of aging is made all to there is certain difference, and the defect of transformer and product defects, weather, load variations etc. are closely related, have obvious seasonal characteristics, therefore transformer generation defect cause roughly can be divided into following three classes according to above-mentioned mode component:

A (), transformer their location have Rules of Seasonal Changes; In addition, the load that transformer is born each season has certain rule, as being summer that peak of power consumption, transformer load are obviously higher than (that is: the Seasonal components of transformer defect cause) such as other seasons;

(b), along with the development of urban construction and the increase of need for electricity, number transformer also increases thereupon, and along with the lengthening of working time, transformer is under the combined action such as hot, electric, mechanical, and the quantity that defect occurs also has the trend of increase (that is: the trend of transformer defect cause and cyclical component);

C (), transformer because of the defect that the reasons such as misoperation, repair and maintenance are improper, outside destroy cause, have uncertainty (that is: the random fluctuation component of transformer defect cause) in operational process.

As can be seen here, first, the genetic defects rate sequence got can be resolved into multiple mode component C by seasonality _itime series corresponding respectively, as X (t), t=1,2,3 ..., N} carries out seasonality and resolves into three mode component time series corresponding respectively, and N is the sum of transformer number of defects sample, i=3;

Secondly, the time series of decomposing each component of obtaining corresponding is respectively analyzed, set up ARIMA (p, d, q) model, same way is all adopted to carry out analysis modeling as selected Box-Jenkins method to the time series of each component, tranquilization process according to current time sequence determines d value, again according to autocorrelation function and the partial autocorrelation function figure of tranquilization current time sequence, and bayesian information criterion (BayesianInformationCriterions, BIC) carry out determining rank, select most suitable p, q value;

Then, ARIMA (p, d, the q) model determined by each component above-mentioned is as the filtered out model satisfied condition and calculate, and as the value calculated a kth component, following formula (4) can be adopted to obtain:

$C_k(t+\mathrm{\τ})=\underset{i=1}{\overset{p}{\Σ}}{\mathrm{\φ}}_{ki}{C}_k\left(t-i\right)+\underset{j=0}{\overset{q}{\Σ}}{\mathrm{\θ}}_{kj}{\mathrm{\α}}_k(t-j)---\left(4\right)$

Wherein, τ-prediction step, for auto-regressive parameter, θ _kjfor running mean parameter.

Step S3, the value of its screening model corresponding for described each mode component obtained to be added up, and using the ratio of defects of described cumulative sum as described month prediction to be measured.

Detailed process is, the value of its screening model corresponding for each mode component is added up, that obtain and for predicting in month to be measured ratio of defects.

As shown in Figures 2 to 5, the application scenarios of the method for transformer ratio of defects prediction in the embodiment of the present invention is described further:

In transformer historical data, get the sequence of actual measurement ratio of defects of continuous 60 months in January, 2009 to 2013 years 12, i.e. genetic defects rate sequence (as shown in Figure 2), totally 60 samples;

X-11 seasonal adjustment method is adopted first to carry out Seasonal analysis to genetic defects rate sequence, as shown in table 1:

Table 1

As known from Table 1, be in March, August, September on the occasion of, and apparently higher than other months, its meaning be these three months larger by seasonal effect factor; On the contrary, the negative values such as June, October, November are little by seasonal effect factor, prove that the generation of defect is certain and have close correlativity season.

Using in January, 2014 (sample point 61) as month to be measured (i.e. sample to be tested), utilize 60 samples of genetic defects rate sequence to carry out the prediction of ratio of defects successively, detailed process is as follows:

Front 60 samples are carried out seasonality to decompose, obtain the time series that 3 mode component are corresponding respectively, as Seasonal sequence C ₁, trend and cyclic sequence C ₂with random fluctuation sequence C ₃, as shown in Figures 3 to 5.Known, Seasonal sequence C ₁with trend and cyclic sequence C ₂all there is significant change rule, random fluctuation sequence C ₃change there is certain randomness.

ARIMA forecast model is set up to the time series of these components.Such as, trend and cyclic sequence C ₂can stationary sequence be obtained after carrying out twice difference, therefore d=2.Corresponding model order, BIC, stably R ²value is listed in the table below shown in 2:

Table 2

Obviously, the BIC value of ARIMA (2,2,4) model is minimum, R ²be worth maximum, therefore in above-mentioned 9 kinds of models, the fitting effect of ARIMA (2,2,4) model is best, by that analogy, carries out ARIMA modeling to other component, obtains the value of fitting effect the best.

Finally, superpose the value that each component corresponding A RIMA obtains, and using the ratio of defects that superposition sum was predicted as in January, 2014, this predicated error is 4%.

As shown in Figure 6, be 60 and predict the outcome and the comparison of wave shape figure of 60 actual measurement samples, degree of fitting inspection is carried out to actual defects rate in Fig. 6 and prediction ratio of defects curve, obtains difference average 0.45%, R side 90.6%, can see that the degree of fitting of two curves is higher.

Decompose if do not carry out seasonality to genetic defects rate sequence, and directly operate time sequence is predicted, in January, 2014 with the predicated error of pure Time Series Method for 18%.

As shown in Figure 7, in the embodiment of the present invention, the system of a kind of transformer ratio of defects prediction provided,

Genetic defects rate retrieval unit 710, for determining month to be measured, and in transformer historical data, chose continuous multiple month for starting point according to the mode of time inverted sequence with the upper January in described month to be measured and distinguish corresponding actual measurement ratio of defects, and the described multiple actual measurement ratio of defects got all are formed genetic defects rate sequence as sample;

Series Decomposition and analytic unit 720, for the method that basis is preset, carry out seasonality to described genetic defects rate sequence to decompose, obtain the time series that each mode component is corresponding respectively, and the time series that described each mode component obtained is corresponding is respectively analyzed, in each mode component, all filter out the model meeting certain condition, and obtain the value of each mode component its screening model corresponding;

Ratio of defects predicted value acquiring unit 730, for adding up the value of described each mode component obtained its screening model corresponding, and using the ratio of defects of described cumulative sum as described month prediction to be measured.

Wherein, described Series Decomposition and analytic unit 720 comprise:

Series Decomposition module 7201, for according to the seasonal adjustment method preset, carries out seasonality to described genetic defects rate sequence and decomposes, obtain the time series that multiple mode component is corresponding respectively;

Series analysis model builds module 7202, analyzes, obtain difference auto regressive moving average ARIMA (p, d, the q) model of each mode component for the time series corresponding respectively to each mode component; Wherein, p is autoregression item; Q is moving average item number, the difference number of times of d for doing when time series becomes steady;

Predicted value acquisition module 7203, for in ARIMA (p, d, the q) model of described each mode component obtained, all adopt correlogram and partial correlation figure process, realize carrying out parameter estimation to p, d, q, determine ARIMA (p, the d of each mode component described, q) under meeting certain condition in model, p, d, q distinguish corresponding value, and the value that ARIMA (p, d, the q) model obtaining meeting in each mode component described certain condition is corresponding.

Wherein, described multiple mode component has three, comprises Seasonal component, trend and cyclical component and random fluctuation component.

Implement the embodiment of the present invention, there is following beneficial effect:

In embodiments of the present invention, because the genetic defects rate sequence formed by transformer historical data carries out the mode that seasonality is decomposed and Time Series AR IMA model prediction combines, the ratio of defects treating moon sight part is predicted, both simplify model and again reduce interference in model between different component and coupling, thus different characteristic information separated can be come, reduce modeling difficulty, the comparatively conventional time series method that predicts the outcome is more accurate, reaches the object improving precision of prediction.

It should be noted that in said system embodiment, each included system unit is carry out dividing according to function logic, but is not limited to above-mentioned division, as long as can realize corresponding function; In addition, the concrete title of each functional unit, also just for the ease of mutual differentiation, is not limited to protection scope of the present invention.

One of ordinary skill in the art will appreciate that all or part of step realized in above-described embodiment method is that the hardware that can carry out instruction relevant by program has come, described program can be stored in a computer read/write memory medium, described storage medium, as ROM/RAM, disk, CD etc.

Above disclosedly be only present pre-ferred embodiments, certainly can not limit the interest field of the present invention with this, therefore according to the equivalent variations that the claims in the present invention are done, still belong to the scope that the present invention is contained.

Claims (6)

1. a method for transformer ratio of defects prediction, it is characterized in that, described method comprises:

A, determine month to be measured, and in transformer historical data, chose continuous multiple month for starting point according to the mode of time inverted sequence with the upper January in described month to be measured and distinguish corresponding actual measurement ratio of defects, and the described multiple actual measurement ratio of defects got all are formed genetic defects rate sequence as sample;

The method that b, basis are preset, carry out seasonality to described genetic defects rate sequence to decompose, obtain the time series that each mode component is corresponding respectively, and the time series that described each mode component obtained is corresponding is respectively analyzed, in each mode component, all filter out the model meeting certain condition, and obtain the value of each mode component its screening model corresponding;

C, the value of its screening model corresponding for described each mode component obtained to be added up, and using the ratio of defects of described cumulative sum as described month prediction to be measured.

2. the method for claim 1, is characterized in that, described step b specifically comprises:

According to the seasonal adjustment method preset, seasonality is carried out to described genetic defects rate sequence and decomposes, obtain the time series that multiple mode component is corresponding respectively;

The time series corresponding respectively to each mode component is analyzed, and obtains difference auto regressive moving average ARIMA (p, d, the q) model of each mode component; Wherein, p is autoregression item; Q is moving average item number, the difference number of times of d for doing when time series becomes steady;

At the ARIMA (p of described each mode component obtained, d, q), in model, all adopt correlogram and partial correlation figure process, realize carrying out parameter estimation to p, d, q, determine the ARIMA (p of each mode component described, d, q) meet the value corresponding respectively of p, d, q under certain condition in model, and obtain the ARIMA (p meeting certain condition in each mode component described, d, q) value that model is corresponding.

3. method as claimed in claim 2, it is characterized in that, described multiple mode component has three, comprises Seasonal component, trend and cyclical component and random fluctuation component.

4. a system for transformer ratio of defects prediction, it is characterized in that, described system comprises:

Genetic defects rate retrieval unit, for determining month to be measured, and in transformer historical data, chose continuous multiple month for starting point according to the mode of time inverted sequence with the upper January in described month to be measured and distinguish corresponding actual measurement ratio of defects, and the described multiple actual measurement ratio of defects got all are formed genetic defects rate sequence as sample;

Series Decomposition and analytic unit, for the method that basis is preset, carry out seasonality to described genetic defects rate sequence to decompose, obtain the time series that each mode component is corresponding respectively, and the time series that described each mode component obtained is corresponding is respectively analyzed, in each mode component, all filter out the model meeting certain condition, and obtain the value of each mode component its screening model corresponding;

Ratio of defects predicted value acquiring unit, for adding up the value of described each mode component obtained its screening model corresponding, and using the ratio of defects of described cumulative sum as described month prediction to be measured.

5. system as claimed in claim 4, it is characterized in that, described Series Decomposition and analytic unit comprise:

Series Decomposition module, for according to the seasonal adjustment method preset, carries out seasonality to described genetic defects rate sequence and decomposes, obtain the time series that multiple mode component is corresponding respectively;

Series analysis model builds module, analyzes, obtain difference auto regressive moving average ARIMA (p, d, the q) model of each mode component for the time series corresponding respectively to each mode component; Wherein, p is autoregression item; Q is moving average item number, the difference number of times of d for doing when time series becomes steady;

Predicted value acquisition module, for in ARIMA (p, d, the q) model of described each mode component obtained, all adopt correlogram and partial correlation figure process, realize carrying out parameter estimation to p, d, q, determine ARIMA (p, the d of each mode component described, q) under meeting certain condition in model, p, d, q distinguish corresponding value, and the value that ARIMA (p, d, the q) model obtaining meeting in each mode component described certain condition is corresponding.

6. system as claimed in claim 5, it is characterized in that, described multiple mode component has three, comprises Seasonal component, trend and cyclical component and random fluctuation component.