CN103886339A - Oil pumping device indicator diagram dynamic identification method and device based on BP neural network - Google Patents

Abstract

The invention relates to an oil pumping device indicator diagram dynamic identification method and device based on a BP neural network. The method comprises the steps of characteristic database establishing and dynamic identification. An establishing process comprises the steps that time domain data of an oil pumping device indicator diagram sample are normalized and sampled to acquire a discrete time domain data sequence; Fourier transformation is carried out on the time domain data sequence to acquire a frequency domain data sequence; partial data in the frequency domain data sequence are used to replace partial data in the time domain data sequence, and a Fourier approximation characteristic value is acquired; the normalized oil pumping device indicator diagram sample is divided into four parts, and a characteristic vector is extracted; the Fourier approximation characteristic value and the characteristic vector are stored in a characteristic database; and a BP neural network algorithm is used to train the oil pumping device indicator diagram sample, so as to correct the Fourier approximation characteristic value in the database. According to the invention, the automation management level of an oil pumping device is enhanced; the work efficiency in an oil field is improved; and the management system of oil well operation is improved.

Description

Pumping unit load-position diagram dynamic identifying method and device based on BP neural network

Technical field

The present invention relates to device management techniques, particularly relate to a kind of pumping unit load-position diagram dynamic identifying method and device based on BP neural network.

Background technology

The pumping unit load-position diagram (being also pumping-unit workdone graphic) in oil field can reflect the operating mode of pumping unit, as the failure condition of pumping unit, pumping efficiency and oil pumper output etc.If can carry out scientific management to the monitoring analysis of pumping unit load-position diagram, can make pumping unit reach higher oil recovery rate and higher oil recovery factor.In view of this,, in current oil well operation, an important action is exactly the load-position diagram that gathers and obtain pumping unit.

Inventor finds realizing in process of the present invention: although can carry out round-the-clock information acquisition by modern measurement means at present, and can utilize corresponding software that the information collecting is depicted as to pumping unit load-position diagram, but, still need at present manually to check pumping unit load-position diagram, and pumping unit load-position diagram is carried out to manual analysis, to judge the operating mode of pumping unit.Therefore, the automatic management degree of pumping unit needs to be further improved.

Because the problem that the automatic management of existing pumping unit exists, practical experience and the professional knowledge of the inventor based on being engaged in this type of product design manufacture and enriching for many years, and the utilization of cooperation scientific principle, actively research and innovation in addition, to founding a kind of pumping unit load-position diagram dynamic identifying method and device based on BP neural network of new structure, the problem that can overcome the automatic management existence of existing pumping unit, makes it have more practicality.Through continuous research and design, and through repeatedly studying sample and improvement, finally create the present invention who has practical value.

Summary of the invention

Fundamental purpose of the present invention is, overcome the problem of the automatic management existence of existing pumping unit, and provide a kind of pumping unit load-position diagram dynamic identifying method and device based on BP neural network of new structure, technical matters to be solved is, strengthen the automatic management degree of pumping unit, and then further improve the crude production rate of oil well, and improve field operations efficiency, the management system of simultaneously improving oil well operation, is very suitable for practicality.

Object of the present invention and solve its technical matters and can adopt following technical scheme to realize.

A kind of pumping unit load-position diagram dynamic identifying method based on BP neural network proposing according to the present invention, comprise: the process of establishing of feature database and the utilization feature database that success has been set up carries out Dynamic Recognition process to the pumping unit load-position diagram gathering, the process of establishing of wherein said feature database comprises: the time domain data to pumping unit load-position diagram sample is normalized, and the time domain data after normalized is carried out to sampling processing, to obtain discrete time domain data sequence; Described discrete time domain data sequence is carried out to Fourier transform, to obtain frequency domain data sequence; Utilize the partial data in described frequency domain data sequence to substitute partial data corresponding to position in described discrete time domain data sequence, and utilize the time domain data sequence after substituting to carry out Fourier approximation computation, to obtain Fourier Approximation Characteristic value; Utilizing diagonal line partitioning scheme is four parts by the pumping unit load-position diagram sample decomposition after described normalized, and extracts proper vector from lower right-most portion; Described Fourier Approximation Characteristic value and described proper vector are stored in feature database; Utilize BP neural network algorithm to the training of pumping unit load-position diagram sample, and utilize the Fourier Approximation Characteristic value of training in the eigenvalue correction feature database obtaining.

Preferably, the aforesaid pumping unit load-position diagram dynamic identifying method based on BP neural network, the feature database that wherein said utilization is successfully set up carries out Dynamic Recognition process to the pumping unit load-position diagram gathering and comprises: the time domain data to the pumping unit load-position diagram gathering is normalized, and the time domain data after normalized is carried out to sampling processing, to obtain discrete time domain data sequence; Described discrete time domain data sequence is carried out to Fourier transform, to obtain frequency domain data sequence; Utilize the partial data in described frequency domain data sequence to substitute partial data corresponding to position in described discrete time domain data sequence, and utilize the time domain data sequence after described substituting to carry out Fourier approximation computation, to obtain Fourier Approximation Characteristic value; Utilize diagonal line partitioning scheme that the pumping unit load-position diagram after described normalized is divided into four parts, and extract proper vector from lower right-most portion; Utilize approach degree algorithm to carry out calculating based on the approach degree of feature database to Fourier Approximation Characteristic value corresponding to the pumping unit load-position diagram collecting and proper vector, to identify the pumping unit load-position diagram of described collection.

Preferably, the aforesaid pumping unit load-position diagram dynamic identifying method based on BP neural network, in the wherein said alternative described discrete time domain data sequence of partial data of utilizing in described frequency domain data sequence, partial data corresponding to position comprises: front K data in described frequency domain data sequence are substituted to front K data in described discrete time domain data sequence, wherein, K is natural number and is less than the data bulk in time domain data sequence.

The present invention proposes a kind of pumping unit load-position diagram Dynamic Recognition device based on BP neural network, and this device comprises: for set up feature database set up module and for utilizing feature database that success is set up the pumping unit load-position diagram gathering to be carried out to the identification module of Dynamic Recognition; This is set up module and comprises: normalization module, for the time domain data of pumping unit load-position diagram sample is normalized, and the time domain data after normalized is carried out to sampling processing, to obtain discrete time domain data sequence; Fourier transform module, for described discrete time domain data sequence is carried out to Fourier transform, to obtain according to substituting partial data corresponding to position in described discrete time domain data sequence, and utilize the time domain data sequence after described substituting to carry out Fourier approximation computation, to obtain Fourier Approximation Characteristic value; Extraction module, is four parts for utilizing diagonal line partitioning scheme by the pumping unit load-position diagram sample decomposition after described normalized, and extracts proper vector from lower right-most portion; Feature database, for storing described Fourier Approximation Characteristic value and described proper vector; Training module, for utilizing BP neural network algorithm to the training of pumping unit load-position diagram sample, and utilizes the Fourier Approximation Characteristic value of training in the eigenvalue correction feature database obtaining.

Preferably, aforesaid Dynamic Recognition device, wherein said identification module comprises: described normalization module, for the time domain data of the pumping unit load-position diagram gathering is normalized, and the time domain data after normalized is carried out to sampling processing, to obtain discrete time domain data sequence; Described Fourier transform module, for described discrete time domain data sequence is carried out to Fourier transform, to obtain frequency domain data sequence; Described Fourier approaches module, substitute partial data corresponding to position in described discrete time domain data sequence for the partial data that utilizes described frequency domain data sequence, and utilize the time domain data sequence after substituting to carry out Fourier approximation computation, to obtain Fourier Approximation Characteristic value; Described extraction module, for utilizing diagonal line partitioning scheme that the pumping unit load-position diagram after described normalized is divided into four parts, and extracts proper vector from lower right-most portion; Approach degree module, for utilizing approach degree algorithm to carry out calculating based on the approach degree of feature database to Fourier Approximation Characteristic value corresponding to the pumping unit load-position diagram collecting and proper vector, to identify the pumping unit load-position diagram of described collection.

Preferably, aforesaid Dynamic Recognition device, in the wherein said alternative described discrete time domain data sequence of partial data of utilizing in described frequency domain data sequence, partial data corresponding to position comprises: front K data in described frequency domain data sequence are substituted to front K data in described discrete time domain data sequence, wherein, K is natural number and is less than the data bulk in time domain data sequence.

By technique scheme, pumping unit load-position diagram dynamic identifying method based on BP neural network of the present invention and device at least have following advantages and beneficial effect: the present invention is by utilizing pumping unit load-position diagram Sample Establishing feature database, and utilize BP neural network algorithm to the training of pumping unit load-position diagram sample, with correction feature storehouse, like this, can utilize the feature database of this correction to identify fast and accurately the pumping unit load-position diagram collecting; Thereby technical scheme provided by the invention has strengthened the automatic management degree of pumping unit, and then further improve the crude production rate of oil well, and improved field operations efficiency, the simultaneously perfect management system of oil well operation.

In sum, the present invention has significant progress technically, and has significantly positive technique effect, is really a new and innovative, progressive, practical new design.

Above-mentioned explanation is only the general introduction of technical solution of the present invention, in order to better understand technological means of the present invention, and can be implemented according to the content of instructions, and for above and other object of the present invention, feature and advantage can be become apparent, below especially exemplified by preferred embodiment, and coordinate Figure of description, be described in detail as follows.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the process of establishing of the feature database based on BP neural network of the present invention;

Fig. 2 is the serious load-position diagram master sample schematic diagram lacking in liquid situation of the present invention;

Fig. 3 is the load displacement change curve schematic diagram after Fourier transform of the present invention;

Fig. 4 is of the present inventionly cut apart schematic diagram to pumping unit load-position diagram sample;

Fig. 5 is the load characteristic vector schematic diagram of the array form of storing in feature database of the present invention;

Fig. 6 is the displacement characteristic vector schematic diagram of the array form of storing in feature database of the present invention;

Fig. 7 is output layer matrix schematic diagram of the present invention;

Fig. 8 is the feature database successfully set up of utilization of the present invention carries out Dynamic Recognition process process flow diagram to the pumping unit load-position diagram gathering;

Fig. 9 is the schematic diagram of the object lesson that the pumping unit load-position diagram collecting is identified of the present invention.

Embodiment

Technological means and effect of taking for reaching predetermined goal of the invention for further setting forth the present invention, below in conjunction with accompanying drawing and preferred embodiment, to the pumping unit load-position diagram dynamic identifying method based on BP neural network proposing according to the present invention and embodiment, structure, feature and effect thereof of device, be described in detail as follows.

Embodiment mono-, pumping unit load-position diagram dynamic identifying method based on BP neural network.

The method of the present embodiment mainly comprises two parts, and a part is: the process of establishing of feature database, another part is: utilize the feature database that success has been set up to carry out Dynamic Recognition process to the pumping unit load-position diagram gathering.Below in conjunction with Fig. 1-9, this two parts content is described respectively.

One, the process of establishing of feature database, its flow process as shown in Figure 1.

In Fig. 1, S100, first the time domain data of pumping unit load-position diagram sample is normalized, afterwards, the time domain data after normalized is carried out to sampling processing, thereby can obtain discrete time domain data sequence.

Above-mentioned pumping unit load-position diagram sample also can be called master sample, and this master sample likely can exist the phenomenon that does not meet concrete regional actual conditions, therefore, need to revise the content corresponding with this sample of storing in feature database by follow-up S150.

The time domain data of pumping unit load-position diagram sample is normalized and can facilitates follow-up processing procedure, make data there is versatility.The present embodiment can adopt existing normalized mode to be normalized time domain data, normalized process is not elaborated at this.

The present embodiment can be sampled to the actual one section of continuous signal x (t) receiving (being the time domain data of pumping unit load-position diagram sample) according to predetermined sampling frequency f, thereby one section of continuous signal x (t) is converted to one section of discrete burst x[n] (being discrete time domain data series).

S110, the discrete time domain data sequence of above-mentioned acquisition is carried out to Fourier transform, to obtain frequency domain data sequence.

Discrete time domain data sequence can be converted to frequency domain data sequence by the Fourier transform in this step.The present embodiment can adopt existing Fourier transform mode to carry out the conversion of time domain data sequence to frequency domain data sequence, the specific implementation process of Fourier transform is not elaborated at this.

A concrete example, the load-position diagram master sample of well W269-10 in the situation that of serious scarce liquid is as Fig. 2, and the load displacement change curve after Fourier transform is as shown in Figure 3.

S120, utilize partial data in above-mentioned frequency domain data sequence to substitute partial data corresponding to position in above-mentioned discrete time domain data sequence, and utilize the time domain data sequence after above-mentioned substituting to carry out Fourier approximation computation, to obtain Fourier Approximation Characteristic value.

Concrete alternative Process can be: front K data in frequency domain data sequence are substituted to front K the data X (k) in discrete time domain data sequence, with reconstructed frequency domain signal, wherein, K is natural number and the quantity that is less than the data in time domain data sequence.

Fourier approaches as linear proximity, and a concrete implementation is: 2n+1 equidistant points X according to f (x) on interval [0,2 π] _ithe functional value fi=f (xi) that=2 π/(2n+1) * (i+0.5) (i=1,2...2n) locates, asks fourier series $f\left(x\right)=\frac{1}{2}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{a}_0+\underset{k=1}{\overset{\∞}{\mathrm{\Σ}}}(a_k\cos \mathrm{kx}+b_k\sin \mathrm{kx})$ Front 2n+1 coefficient a _k(k=0,1,2 ... and b n) _k(k=0,1,2 ...) approximate value.

Afterwards, calculate a _kand b _k, the example of a concrete computation process is as follows:

For (k=0,1,2 ... n) calculate according to following iterative formula:

U _2n+2=U _2n+1=0;

u _j=f _i+2coskθu _j+1-u _j+2(j=2n,2n-1...1)

Wherein, $\mathrm{\θ}=\frac{2\mathrm{\π}}{2n+1};$

Calculate cosk θ and sink θ with following recursion formula:

coskθ=cosθcos(k-1)θ-sinθsin(k-1)θ;

sinkθ=sinθcos(k-1)θ-cosθsin(k-1)θ;

Then, utilize following formula to calculate a _kand b _k;

$a_k=\frac{2}{2n+1}(f_0+u_1\cos \mathrm{k\θ}-u_2);$

$b_k=\frac{2}{2n+1}{u}_1\sin \mathrm{k\θ};$

Thereby utilize above-mentioned computation process can obtain a _kand b _ktwo feature arrays, get each feature array the first two (be a[0], a[1], b[0] and b[1]) as Fourier Approximation Characteristic value.

S130, to utilize diagonal line partitioning scheme be four parts by the pumping unit load-position diagram sample decomposition after above-mentioned normalized, and extract proper vector from lower right-most portion.

Concrete, the implementation that pumping unit load-position diagram sample is cut apart is: as shown in Figure 4, the sequence of choosing displacement smallest point and maximum point is end points, these two end points are connected to form a diagonal line; Maximum point and the smallest point of choosing load are upper and lower flex point, and upper and lower flex point is connected, and can form another diagonal line (if these two cornerwise end points are linked in sequence, can obtain a quadrilateral); These two diagonal line can be divided into four parts by the pumping unit load-position diagram after normalized.

Afterwards, in the process of extraction proper vector, can be take displacement smallest point as initial value at the selected characteristic vector of bottom right figure (being the figure in Section 4 limit) jumping characteristic.

S140, Fourier Approximation Characteristic value and the proper vector of above-mentioned acquisition are stored in feature database.

For Fig. 2, the proper vector of storing in feature database as shown in Figure 5 and Figure 6.What Fig. 5 showed is the load characteristic vector of the array form of storing in feature database.What Fig. 6 showed is the displacement characteristic vector of the array form of storing in feature database.

S150, utilize BP neural network algorithm to the training of pumping unit load-position diagram sample, and utilize the Fourier Approximation Characteristic value in the eigenvalue correction feature database that training obtains.

Concrete, this step is to meet the pumping unit load-position diagram of local conditions as the standard indicator diagram of a certain particular type for local concrete condition using one, and utilize existing BP neural network algorithm take the load-position diagram of this standard as basis, the eigenwert of the same type of having stored in feature database is proofreaied and correct.

An object lesson that utilizes BP neural network algorithm to carry out sample training and to proofread and correct is carried out to brief description below.

(1), weight coefficient W _ijput initial value:

To the weight coefficient W of each layer _ijput a less non-zero random number, but W wherein _in+ 1=-θ.

(2) a sample X of input ₀and respective desired values:

$y=e^{{-x}_0^2}$

(3) calculate the output of each layer:

Ground floor:

U _0-1=W _0-1X ₀-θ _0-1

$X_{0-1}=f\left(U_{0-1}\right)=\frac{1}{1+\exp \left({-U}_{0-1}\right)}$

The second layer:

$U_{1-i}=\underset{j=1}{\overset{2}{\mathrm{\&Sigma;}}}{X}_{1-\mathrm{ij}}{X}_{0-j}=W_{l\_\mathrm{<figure-callout\; id="0"\; label="il\; X"\; filenames="HDA00003345796300022.png,HDA00003345796300032.png"\; state="\{\{state\}\}">il</figure-callout>}}{X}_{}{}_{0\_1}-{\mathrm{\&theta;}}_{1-i}$

Wherein, i=1,2,3,4.

The 3rd layer:

$U_{2-i}=\underset{j=1}{\overset{5}{\mathrm{\&Sigma;}}}{W}_{2-\mathrm{ij}}{X}_{1-j}=\underset{j=1}{\overset{4}{\mathrm{\&Sigma;}}}{W}_{2-\mathrm{ij}}{X}_{1-j}-{\mathrm{\&theta;}}_{2-i}$

$X_{2-i}=f\left(U_{2-i}\right)=\frac{1}{1+\exp \left({-U}_{2-i}\right)}$

Wherein, i=1,2,3,4,5,6.

(4) calculate:

e=X _4-l-Y

As forwarding (8) to, e<0.003 carries out operation accordingly, otherwise, forward (5) to and carry out operation accordingly.

(5) ask the learning error of each layer:

k=3;j=1,2,3,4,5,6；

$d_{2-j}=X_{2-j}(1-X_{2-j})\underset{l}{\mathrm{\&Sigma;}}{W}_{3-1j}{d}_{3-l}$

k=2;j=1,2,3,4；

$d_{1-j}=X_{1-j}(1-X_{1-j})\underset{l}{\mathrm{\&Sigma;}}{W}_{2-1j}{d}_{2-l}$

$d_{0-j}=X_{0-j}(1-X_{0-j})\underset{l}{\mathrm{\&Sigma;}}{W}_{1-1j}{d}_{1-l}$

(6) modified weight coefficient and threshold values:

In the time of k=1:

W _0-1=W _0-1-ηd _0-1X ₀+αW _0-0

In the time of k=2:

I=1,2,3,4;j=1,2；

W _1-ij=W _1-ij-ηd _1-iX _1-j+αW _1-0

In the time of k=3:

I=1,2,3,4,5,6;j=1,2,3,4；

W _2-ij=W _2-ij-ηd _1-iX _1-j+αW _1-0

(7) return to (3) and carry out operation accordingly:

(8) export the weight coefficient of each layer.

Under VC environment, test b P algorithm is:

Printf (" please input node numbers of hidden layers: n ");

scanf("%d",&(*bp).dot)；

Printf (" please input learning rate: n ");

scanf("%lf",&(*bp).rate)；

Printf (" please input permissible error limit: n ");

scanf("%lf",&(*bp).error)；

Here through suitable this application of the following value of test input:

Node numbers of hidden layers: bp->dot=20;

Learning rate: bp->rate=0.5;

Permissible error limit: bp->error=0.0001;

The eigenwert of VC test input Fourier approximation computation is training sample matrix, and 3 values, a[0 are got in now test], a[1] and b[0];

Stochastic matrix of initialization, the output layer matrix obtaining after calculating as shown in Figure 7.By sample training, can make the coupling standard more of pumping unit load-position diagram.

Two, utilize the feature database that success has been set up to carry out Dynamic Recognition process to the pumping unit load-position diagram gathering, its flow process as shown in Figure 8.

In Fig. 8, S200, first the time domain data of the pumping unit load-position diagram collecting is normalized, and the time domain data after normalized is carried out to sampling processing, to obtain discrete time domain data sequence.

S210, above-mentioned discrete time domain data sequence is carried out to Fourier transform, to obtain frequency domain data sequence.

S220, utilize partial data in frequency domain data sequence to substitute partial data corresponding to position in above-mentioned discrete time domain data sequence, and utilize the time domain data sequence after substituting to carry out Fourier approximation computation, to obtain Fourier Approximation Characteristic value.

S230, utilize diagonal line partitioning scheme that the pumping unit load-position diagram after normalized is divided into four parts, and extract proper vector from lower right-most portion.

S240, the proper vector of utilizing Fourier Approximation Characteristic value that approach degree algorithm obtains S220 and S230 to obtain carry out calculating based on the approach degree of feature database, to identify the pumping unit load-position diagram collecting.

The specific implementation process of above-mentioned steps S200-S230 and the specific implementation process of above-mentioned steps S100-130 are basic identical, no longer describe in detail at this, below S240 are further described.

The object lesson that approach degree based on feature database calculates can be:

1, based on feature database obtain feature array (as, by the whole arrays in feature database according to the synthetic array of ID der group), obtain actual test feature array, contrast respectively;

2, if (a>b) { s1+=b; S2+=a; Else{s1+=a; S2+=b; , wherein a is feature database characteristic of correspondence array, b is actual test feature array;

3, t=s1/s2; Then, directly obtain type number according to t value.

The object lesson that the pumping unit load-position diagram collecting is identified as shown in Figure 9, on the picture showing in application software of the present invention, after double-clicking certain well, can directly obtain type number is the information warning that oil well seriously lacks liquid, thereby oil field personnel can carry out relevant operation in time to the pumping unit of this oil well.

Embodiment bis-, pumping unit load-position diagram Dynamic Recognition device based on BP neural network.

The device of the present embodiment mainly comprises: set up module and identification module.

Set up module and be mainly used in setting up feature database.The feature database of this foundation can be stored in to be set up in module, also can be stored in other memory modules.

Identification module is mainly used in after receiving the pumping unit load-position diagram of collection that external transmission comes, and utilizes above-mentionedly to set up the pumping unit load-position diagram that feature database that module successfully sets up receives it and carry out Dynamic Recognition.

The above-mentioned module of setting up mainly comprises: normalization module, Fourier transform module, Fourier approach module, extraction module, feature database and training module.Above-mentioned identification module mainly comprises: normalization module, Fourier transform module, Fourier approach module, extraction module and approach degree module.

Above-mentionedly set up module and identification module can comprise respectively that normalization module, Fourier transform module, Fourier separately approach module and extraction module; Certainly, same set of normalization module, Fourier transform module, Fourier approaches module and extraction module also can belong to simultaneously and set up module and identification module.

Normalization module is mainly used in, in the time receiving pumping unit load-position diagram sample, the time domain data of this sample being normalized, and the time domain data after normalized being carried out to sampling processing, to obtain discrete time domain data sequence; In the time receiving the pumping unit load-position diagram of collection, the time domain data of this load-position diagram is normalized, and the time domain data after normalized is carried out to sampling processing, to obtain discrete time domain data sequence.

Normalization module can adopt existing normalized mode to be normalized time domain data, the performed normalized process of normalization module is not elaborated at this.

Normalization module can be sampled to the actual one section of continuous signal x (t) receiving (being the time domain data of the time domain data of pumping unit load-position diagram sample or the pumping unit load-position diagram that collects) according to predetermined sampling frequency f, thereby normalization module can be converted to one section of continuous signal x (t) one section of discrete burst x[n] (being discrete time domain data series).

Fourier transform module is mainly used in the discrete time domain data sequence of normalization module output to carry out Fourier transform, to obtain frequency domain data sequence.

Fourier transform module can be converted to frequency domain data sequence by discrete time domain data sequence, and Fourier transform module can adopt existing Fourier transform mode to carry out the conversion of time domain data sequence to frequency domain data sequence, the specific implementation process of the performed Fourier transform of Fourier transform module is not elaborated at this.

Fourier approaches module and is mainly used in utilizing partial data in the frequency domain data sequence of Fourier transform module output to substitute partial data corresponding to position in the discrete time domain data sequence of normalization module output, and utilize the time domain data sequence after substituting to carry out Fourier approximation computation, to obtain Fourier Approximation Characteristic value.

Concrete, Fourier approaches module can substitute front K the data X (k) in discrete time domain data sequence by front K data in frequency domain data sequence, with reconstructed frequency domain signal, wherein, K is natural number and the quantity that is less than the data in time domain data sequence.

Fourier approaches a specific implementation that the performed Fourier of module approaches as the description in above-described embodiment one, no longer describes in detail at this.

Pumping unit load-position diagram sample or pumping unit load-position diagram that extraction module obtains after being mainly used in utilizing diagonal line partitioning scheme that normalization module is normalized are divided into four parts, and lower right-most portion from four parts is extracted proper vector.

The implementation that extraction module is cut apart pumping unit load-position diagram sample is: the sequence of choosing displacement smallest point and maximum point is end points, and these two end points are connected to form a diagonal line; Maximum point and the smallest point of choosing load are upper and lower flex point, and upper and lower flex point is connected, and can form another diagonal line (if these two cornerwise end points are linked in sequence, can obtain a quadrilateral); These two diagonal line can be divided into four parts by the pumping unit load-position diagram after normalized.Afterwards, extraction module, can be take displacement smallest point as initial value at the selected characteristic vector of bottom right figure (being the figure in Section 4 limit) jumping characteristic in the process of extraction proper vector.

Feature database is mainly used in, and for a pumping unit load-position diagram sample, storage Fourier approaches Fourier Approximation Characteristic value that this sample of module output is corresponding and this sample characteristic of correspondence vector of extraction module output.

Training module is mainly used in utilizing BP neural network algorithm to the training of pumping unit load-position diagram sample, and utilizes the Fourier Approximation Characteristic value of training in the eigenvalue correction feature database obtaining.

For local concrete condition, can meet the pumping unit load-position diagram of local conditions as the standard indicator diagram of a certain particular type using one, training module can utilize existing BP neural network algorithm take the load-position diagram of this standard as basis, and the eigenwert of the same type of having stored in feature database is proofreaied and correct.Training module utilize BP neural network algorithm carry out sample training and proofread and correct object lesson as the description in above-described embodiment one, no longer describe in detail at this.

Approach degree module is mainly used in eigenwert to store in feature database and proper vector and Fourier, and to approach Fourier Approximation Characteristic value and the proper vector that module and extraction module export for the pumping unit load-position diagram collecting be to calculate basis, carry out approach degree calculating, belong to the load-position diagram of a certain type to identify the pumping unit load-position diagram collecting.The object lesson that the performed approach degree of approach degree module calculates, as the description in above-described embodiment one, no longer describes in detail at this.

The above is only preferred embodiment of the present invention, not the present invention is done to any pro forma restriction, although the present invention discloses as above with preferred embodiment, but not in order to limit technology of the present invention, any those skilled in the art are not departing within the scope of technical solution of the present invention, when can utilizing the technology contents of above-mentioned announcement to make a little change or being modified to the equivalent embodiment of equivalent variations, in every case be the content that does not depart from technical solution of the present invention, any simple modification of above embodiment being done according to technical spirit of the present invention, equivalent variations and modification, all still belong in the scope of technical solution of the present invention.

Claims (6)

1. the pumping unit load-position diagram dynamic identifying method based on BP neural network, it is characterized in that, described method comprises: the process of establishing of feature database and the utilization feature database that success has been set up carries out Dynamic Recognition process to the pumping unit load-position diagram gathering, and the process of establishing of wherein said feature database comprises:

Time domain data to pumping unit load-position diagram sample is normalized, and the time domain data after normalized is carried out to sampling processing, to obtain discrete time domain data sequence;

Described discrete time domain data sequence is carried out to Fourier transform, to obtain frequency domain data sequence;

Utilize the partial data in described frequency domain data sequence to substitute partial data corresponding to position in described discrete time domain data sequence, and utilize the time domain data sequence after described substituting to carry out Fourier approximation computation, to obtain Fourier Approximation Characteristic value;

Utilizing diagonal line partitioning scheme is four parts by the pumping unit load-position diagram sample decomposition after described normalized, and extracts proper vector from lower right-most portion;

Described Fourier Approximation Characteristic value and described proper vector are stored in feature database;

Utilize BP neural network algorithm to the training of pumping unit load-position diagram sample, and utilize the Fourier Approximation Characteristic value of training in the eigenvalue correction feature database obtaining.

2. dynamic identifying method as claimed in claim 1, is characterized in that, the feature database that described utilization is successfully set up carries out Dynamic Recognition process to the pumping unit load-position diagram gathering and comprises:

Time domain data to the pumping unit load-position diagram gathering is normalized, and the time domain data after normalized is carried out to sampling processing, to obtain discrete time domain data sequence;

Described discrete time domain data sequence is carried out to Fourier transform, to obtain frequency domain data sequence;

Utilize the partial data in described frequency domain data sequence to substitute partial data corresponding to position in described discrete time domain data sequence, and utilize the time domain data sequence after described substituting to carry out Fourier approximation computation, to obtain Fourier Approximation Characteristic value;

Utilize diagonal line partitioning scheme that the pumping unit load-position diagram after described normalized is divided into four parts, and extract proper vector from lower right-most portion;

Utilize approach degree algorithm to carry out calculating based on the approach degree of feature database to Fourier Approximation Characteristic value corresponding to the pumping unit load-position diagram collecting and proper vector, to identify the pumping unit load-position diagram of described collection.

3. dynamic identifying method as claimed in claim 1 or 2, is characterized in that, in the described alternative described discrete time domain data sequence of partial data of utilizing in described frequency domain data sequence, partial data corresponding to position comprises:

Front K data in described frequency domain data sequence are substituted to front K data in described discrete time domain data sequence, and wherein, K is natural number and is less than the data bulk in time domain data sequence.

4. the pumping unit load-position diagram Dynamic Recognition device based on BP neural network, it is characterized in that, comprising: for set up feature database set up module and for utilizing feature database that success is set up the pumping unit load-position diagram gathering to be carried out to the identification module of Dynamic Recognition;

The described module of setting up comprises:

Normalization module, for the time domain data of pumping unit load-position diagram sample is normalized, and carries out sampling processing to the time domain data after normalized, to obtain discrete time domain data sequence;

Fourier transform module, for described discrete time domain data sequence is carried out to Fourier transform, to obtain frequency domain data sequence;

Fourier approaches module, substitute partial data corresponding to position in described discrete time domain data sequence for the partial data that utilizes described frequency domain data sequence, and utilize the time domain data sequence after described substituting to carry out Fourier approximation computation, to obtain Fourier Approximation Characteristic value;

Extraction module, is four parts for utilizing diagonal line partitioning scheme by the pumping unit load-position diagram sample decomposition after described normalized, and extracts proper vector from lower right-most portion;

Feature database, for storing described Fourier Approximation Characteristic value and described proper vector;

Training module, for utilizing BP neural network algorithm to the training of pumping unit load-position diagram sample, and utilizes the Fourier Approximation Characteristic value of training in the eigenvalue correction feature database obtaining.

5. Dynamic Recognition device as claimed in claim 4, is characterized in that, described identification module comprises:

Described normalization module, for the time domain data of the pumping unit load-position diagram gathering is normalized, and carries out sampling processing to the time domain data after normalized, to obtain discrete time domain data sequence;

Described Fourier transform module, for described discrete time domain data sequence is carried out to Fourier transform, to obtain frequency domain data sequence;

Described Fourier approaches module, substitute partial data corresponding to position in described discrete time domain data sequence for the partial data that utilizes described frequency domain data sequence, and utilize the time domain data sequence after described substituting to carry out Fourier approximation computation, to obtain Fourier Approximation Characteristic value;

Described extraction module, for utilizing diagonal line partitioning scheme that the pumping unit load-position diagram after described normalized is divided into four parts, and extracts proper vector from lower right-most portion;

Approach degree module, for utilizing approach degree algorithm to carry out calculating based on the approach degree of feature database to described Fourier Approximation Characteristic value corresponding to the pumping unit load-position diagram collecting and proper vector, to identify the pumping unit load-position diagram of described collection.

6. the Dynamic Recognition device as described in claim 4 or 5, is characterized in that, in the described alternative described discrete time domain data sequence of partial data of utilizing in described frequency domain data sequence, partial data corresponding to position comprises:

Front K data in described frequency domain data sequence are substituted to front K data in described discrete time domain data sequence, and wherein, K is natural number and is less than the data bulk in time domain data sequence.

Images