CN102043900A - Failure prediction method of rod pumping system based on indicator diagram - Google Patents

Abstract

The invention discloses a failure prediction method of a rod pumping system based on an indicator diagram. The failure prediction method is as follows: taking normal (or steady state) sample ground indicator diagram as an entry point; on the basis of failure foretoken analysis, adopting an autoregressive moving average (ARMA) time sequence model to carry out prediction on the characteristic quantity of a predicated sample; and finally, adopting 'a failure hierarchical method of the rod pumping system based on the indicator diagram' to carry out failure diagnosis and failure recognition. By utilizing the failure prediction method, a mechanical model of the rod pumping system does not need to be established and solved, and the problem of a training set does not exist; and simultaneously, by adopting the failure foretoken analysis, a large amount of time is saved, the efficiency of failure predication is improved, and the predication accuracy is effectively improved due to adoption of the failure foretoken analysis and the ARMA time sequence model together.

Description

Rod pumping system failure prediction method based on load-position diagram

Technical field

The present invention relates to a kind of rod pumping system failure prediction method, belong to the technical field of rod pumping system failure prediction method based on load-position diagram.

Background technology

The condition based maintenance of system is subject to people's attention gradually, and the failure prediction technology, all is significant for improving production safety, reduce production costs and prolonging service life of equipment as the core technology of condition based maintenance.At present, in the maintenance system of sucker rod machinery and equipment, each oil field is still based on maintenance, periodic maintenance after the fault, but the two major defect is to owe to safeguard and safeguard excessively and deposit now, its result will cause equipment to go to work braving one's illness, or the anosis diagnosis and treatment of equipment, obviously this can not satisfy the requirement of enterprise's lean mode of production.At present, both at home and abroad also seldom to the research of rod pumping system failure prediction.

It is being carried out in the system of intelligent diagnostics and predictive maintenance,, grasping the fault rule of plant equipment, and the trend of its running status predicted just seeming extremely important for monitoring and the maintenance that makes equipment obtains the optimal economic technique effect.Various plant equipment in use, its performance or state progressively descend along with the passing of service time, before taking place, a lot of faults have some omens, so-called incipient fault that Here it is, it shows that a kind of functional fault be about to take place, and functional fault shows that then system or equipment lost the performance standard of regulation.From the time interval that incipient fault changes functional fault into, can be by several seconds to several years, the interval of mutation failure is just very short.

The rod pumping system fault has two kinds of mutation failure and gradual failures, and mutation failure comprises: plunger is deviate from working barrel, rod parting, pump seizure, piston, and to meet card, valve malfunctioning etc.Gradual failure all has a process from quantitative change to qualitative change, initial bud, hidden failure state and the malfunction of experience fault.In the development and evolution process of fault, the running status of rod pumping system is producing corresponding the variation.

Rod pumping system is a complicated nonlinear systems, and it has at many levels, not only there are differences on 26S Proteasome Structure and Function between each straton system, and exists very complicated coupled relation between the subsystem.The strong nonlinearity characteristic of rod pumping system makes the generation of the system failure be caused by many factors often.System is on the course of work, and its parts promptly cause the change of inherent characteristic, thereby make system can not produce normal output because wearing and tearing, factor such as tired, aging can cause the deterioration or the inefficacy of equipment; Also can change, install factors such as improper, mutual alignment change between subsystem or the parts owing to parameter and cause deterioration or inefficacy, thereby cause system to depart from normal state; The unusual input of system also can make the state of system change, if these variations exceed certain scope, will produce unusual output.Meanwhile, these outputs can cause that again other equipment states change, and these influences are propagated step by step, and the state of total system is changed, and produce output unusually.Therefore, the fault of rod pumping system usually shows: level, timeliness, correlativity, ambiguity, randomness, not knowing property and relativity.Rod pumping system shows the strong nonlinear characteristics as a complicated mechanical system.Therefore, rod pumping system sequence working time can be carried out long-term or short-term forecasting theoretically.

Reflection oil pumping system polished rod load is called surface dynamometer card or polished rod load-position diagram with the figure of its change in displacement rule, is called for short load-position diagram, and it is the closed curve that indicator measures in suction period of oil pumper.Surface dynamometer card is the firsthand information of well on the beam oil recovery collection in worksite, the variation that the corresponding apparatus state is taking place along with the fault progression process, embodiment that all can be certain on load-position diagram.The failure prediction that realizes rod pumping system based on surface dynamometer card is to removing the oil well fault, guaranteeing that oil well ordinary production or raising oil well output etc. just seem extremely important and have realistic meaning.

Summary of the invention

Technical matters: the present invention seeks at the rod pumping system failure prediction research at present present situation all few with application, with normal (or plateau) sample surface dynamometer card as point of penetration, a kind of rod pumping system failure prediction method based on load-position diagram is provided, on the basis that the fault omen is analyzed, adopt the ARMA time series models that the characteristic quantity of forecast sample is predicted, adopt " the rod pumping system fault based on load-position diagram is passed the rank diagnostic method " to carry out fault at last and differentiate and Fault Identification.It has " the rod pumping system fault based on load-position diagram is passed the rank diagnostic method " same advantage: do not need to set up and find the solution the rod pumping system mechanical model, also do not have the training set problem.Simultaneously, a large amount of time has been saved in the analysis of fault omen, improves the efficient of failure prediction, and has effectively improved precision of prediction with the ARMA time series models.

The present invention adopts following technical scheme for achieving the above object:

The rod pumping system failure prediction method that the present invention is based on load-position diagram comprises the steps:

1) safety zone and the hazardous location of 15 characteristic quantities of calculating:, calculate the safety zone [d, c] and hazardous location (∞ of each characteristic quantity according to the characteristic quantity statistical law, d) ∪ (c ,+∞), the hazardous location is divided into left explosive area (∞, d) and right explosive area (c ,+∞); For load mean change amount, significant hazardous location be right explosive area (c ,+∞);

2) the fault omen is analyzed: whether the sample that first judgement will be predicted (establishing catalogue number(Cat.No.) is N+1) exists the fault omen, as if no omen, does not then need to predict, judges that directly this forecast sample is normal sample, finishes; If omen is arranged, then calculate the data source of N+1 sample of prediction, jump procedure 2);

3) characteristic quantity of forecast sample prediction:, predict 15 characteristic quantities of N+1 sample and whether have the knotting point according to data source;

4) to predicting the outcome, adopt " the rod pumping system fault based on load-position diagram is passed the rank diagnostic method " to carry out fault and differentiate, judge whether N+1 sample be normal, if fault sample further carries out the identification of fault type.

Preferably, step 1) described characteristic quantity safety zone and hazardous location computing method are as follows:

11) each characteristic quantity is chosen unified probable value α ₁, α ₁For sample falls into interval, safety zone [d ₀, c ₀] probability, α ₁=99%, and sample falls into interval, hazardous location (∞, d ₀) and interval, hazardous location (c ₀The probability of ,+∞) is identical, then

P{-∞＜X＜d ₀}+P{d ₀≤X≤c ₀}+P{c ₀＜X＜+∞}＝1 (1)

P{-∞＜X＜d ₀}＝P{c ₀＜X＜+∞} (2)

$P\{-\&infin;<X<d_0\}=P\{c_0<X<+\&infin;\}=\frac{1-P\{d_0\&le;X\&le;c_0\}}{2}=\frac{1-{\mathrm{\&alpha;}}_1}{2}---\left(3\right)$

So $P\{-\&infin;<X<c_0\}=1-{\mathrm{\&alpha;}}_1+\frac{1-{\mathrm{\&alpha;}}_1}{2}=\frac{3+{\mathrm{\&alpha;}}_1}{2}---\left(4\right)$

12) according to the distribution pattern and the parameter of formula (3), (4) and this characteristic quantity, calculate c ₀, d ₀, can obtain interval, safety zone [d ₀, c ₀] and interval, hazardous location (∞, d ₀) ∪ (c ₀,+∞); For load mean change amount, significant hazardous location is (c ₀,+∞);

13) if the regularity of distribution assay of certain characteristic quantity is to meet a plurality of distributions, then respectively by between a plurality of regularity of distribution computationally secure area region, and the union of getting them is this interval, characteristic quantity safety zone.

14), get then that the safety zone is interval to be [μ-3 σ, μ+3 σ], wherein if its regularity of distribution is can not determine in the test of hypothesis of certain characteristic quantity

Be sample average,

Be the sample standard deviation variance.

15) make d ₁=d ₂-λ (c ₂-d ₂), c ₁=c ₂+ λ (c ₂-d ₂), λ is a preset threshold, λ=0.1, wherein d ₂, c ₂Be respectively the minimum value and the maximal value of training sample.Interval [d ₁, c ₁] with step 11) to 14) union in the interval, safety zone that calculates is the final interval, safety zone [d, c] of this characteristic quantity, then the final interval, hazardous location of this characteristic quantity (∞, d) ∪ (c ,+∞).

Preferably, step 2) condition of described fault omen is:

(a) certain characteristic quantity of N sample enters a left side or right explosive area, and sample N-4, N-3, and N-2, N-1, this characteristic quantity of N reduces continuously or increases;

(b) N-9, N-8, L, certain characteristic quantity of N sample increases continuously or reduces continuously;

(c) N-4, N-3, N-2, N-1, N sample standard deviation enters the hazardous location;

(d) N-4, N-3, N-2, N-1, there is the knotting point in N sample standard deviation

If satisfy one of above four conditions, judge that then there is the fault omen in the N+1 sample that will predict.

Preferably, step 2) data source of N+1 sample of described prediction is meant if meet the condition of fault omen, continuation is the data source of N+1 sample of prediction to returning a characteristic quantity and a knotting information that whether exists of seeking all samples that meet identical rule continuously.

Preferably, the ARMA time series models are adopted in the prediction of the characteristic quantity of the described forecast sample of step 3), and its modeling and prediction flow process are as follows:

21) obtain the value of the sample coefficient of autocorrelation and the sample PARCOR coefficients of this sequence;

22) come preliminary judgement model parameter p and q according to the truncation or the hangover property of sample coefficient of autocorrelation and sample PARCOR coefficients, carry out model fitting;

23) unknown parameter in the estimation model;

24) model optimization is set up a plurality of models, chooses the more excellent model of match, and predicts.

In the inventive method, the α of step 11) ₁, step 15) λ be preset threshold, mainly determine according to accuracy requirement and calculating experience, the invention discloses its preferred numerical value.

Beneficial effect: the rod pumping system failure prediction method that the present invention is based on load-position diagram is keeping having improved accuracy of predicting under calculating and the simple prerequisite of model; It does not need to set up and find the solution the rod pumping system mechanical model, does not have the training set problem yet; Simultaneously, a large amount of time has been saved in the analysis of fault omen, has improved the efficient of failure prediction, and has effectively improved precision of prediction with the ARMA time series models.The present invention utilizes the load-position diagram characteristic quantity sequence that characterizes the rod pumping system state to realize the prediction of sucker rod state future development trend, and example shows that forecasting efficiency and precision are higher, especially for gradual change or the concentrated fault type that occurs.

Description of drawings

Fig. 1: based on the process flow diagram of the rod pumping system failure prediction method of load-position diagram

Embodiment

The present invention is directed to the research at present of rod pumping system failure prediction and use all few present situation, with normal (or plateau) sample surface dynamometer card as point of penetration, a kind of rod pumping system failure prediction method based on load-position diagram is provided, on the basis that the fault omen is analyzed, adopt the ARMA time series models that the characteristic quantity of forecast sample is predicted, adopt " the rod pumping system fault based on load-position diagram is passed the rank diagnostic method " to carry out fault at last and differentiate and Fault Identification.The present invention utilizes the load-position diagram characteristic quantity sequence that characterizes the rod pumping system state to realize the prediction of sucker rod state future development trend, and forecasting efficiency and precision height are especially for gradual change or the concentrated fault type that occurs.

As Fig. 1, the rod pumping system failure prediction method that the present invention is based on load-position diagram comprises the steps:

1) safety zone and the hazardous location of 15 characteristic quantities of calculating:, calculate the safety zone [d, c] and hazardous location (∞ of each characteristic quantity according to the characteristic quantity statistical law, d) ∪ (c ,+∞), the hazardous location is divided into left explosive area (∞, d) and right explosive area (c ,+∞); For load mean change amount, significant hazardous location be right explosive area (c ,+∞);

2) the fault omen is analyzed: whether the sample that first judgement will be predicted (establishing catalogue number(Cat.No.) is N+1) exists the fault omen, as if no omen, does not then need to predict, judges that directly this forecast sample is normal sample, finishes; If omen is arranged, then calculate the data source of N+1 sample of prediction, jump procedure 2);

3) characteristic quantity of forecast sample prediction:, predict 15 characteristic quantities of N+1 sample and whether have the knotting point according to data source;

4) to predicting the outcome, adopt " the rod pumping system fault based on load-position diagram is passed the rank diagnostic method " to carry out fault and differentiate, judge whether N+1 sample be normal, if fault sample further carries out the identification of fault type.

It is as follows that described rod pumping system fault based on load-position diagram is passed the rank diagnostic method:

Passing the rank diagnostic method based on the rod pumping system fault of load-position diagram comprises the steps:

1a) fault is differentiated the stage, and the fault stage of differentiating is divided into two stages again:

11a) the training stage: the training stage by the geometric feature on 15 load-position diagram of normal or steady sample extraction, obtain ASSOCIATE STATISTICS information according to the characteristic quantity of these samples:

111a) manually choose n normal or steady sample, carry out 15 features extraction, wherein n 〉=50;

112a) characteristic quantity that extracts is carried out test of outlier,, then reject this sample, and carry out test of outlier again, until there not being exceptional value if unusual;

113a) sample behind the test of outlier is carried out regularity of distribution check;

114a) calculate normal region and fault zone according to regularity of distribution assay;

12a) the resolution stage: whether exist knotting point and test sample book whether to have characteristic quantity to fall into the fault zone according to test sample book itself and judge whether fault:

121a) judging whether test sample book itself exists the knotting point, if exist, then is fault sample, jump procedure 2a); Otherwise jump procedure 122a);

122a) 15 characteristic quantities of extraction test sample book;

123a) whether test sample book has characteristic quantity to fall into the fault zone, if having, then is fault sample, jump procedure 2a); Otherwise jump procedure 124a);

124a) judge that test sample book is normal sample, do not need to carry out Fault Identification, finish;

2a) the Fault Identification stage: adopt search tree method, fault sample is carried out the identification of fault type based on the rod pumping system Fault Identification of load-position diagram.

Preferably, described 15 characteristic quantities are that bottom dead centre (E point) load, upper dead center (F point) displacement and load, standing valve are opened the displacement of point (B point) and displacement that load, travelling valve are opened point (D point) and load, area, maximum load, minimum load, the total mean change amount of load, the mean change amount of EB, BF, FD, DE section load.

Preferably, step 112a) method of described rejecting abnormalities sample adopts t test criterion rejecting abnormalities data method:

21a) in n observed reading, find out with mean value mutually the feature value of the normal sample of ratio error maximum as dubious value x _k, wherein observed reading is a characteristic quantity of normal sample;

22a) to not comprising dubious value x _kIn an interior n-1 observed reading, calculating mean value:

$\stackrel{\&OverBar;}{x}=\left(\underset{\underset{i=1}{i\&NotEqual;k}}{\overset{n}{\mathrm{\&Sigma;}}}{x}_i\right)/(n-1)---\left(1a\right)$

And standard deviation estimated value:

$\stackrel{\&OverBar;}{s}=\sqrt{\left[\underset{\underset{i=1}{i\&NotEqual;k}}{\overset{n}{\mathrm{\&Sigma;}}}{(x_i-\stackrel{\&OverBar;}{x})}^2\right]/(n-2)}---\left(2a\right)$

23a) determine relative risk α _T, α _T=0.001, find A from the t distribution table _T(α _TN-2), calculate again:

$K({\mathrm{\&alpha;}}_T,n)=A_T({\mathrm{\&alpha;}}_T;n-2)\&CenterDot;\sqrt{\frac{n}{n-1}}---\left(3a\right)$

24a) check x _kIf have

$|x_k-\stackrel{\&OverBar;}{x}|>K({\mathrm{\&alpha;}}_T,n)\&CenterDot;\stackrel{\&OverBar;}{s}---\left(4a\right)$

Set up, then x _kBe exceptional value, should reject jump procedure 25a); Otherwise x _kNot exceptional value, can not reject, and stop checking;

25a) if x _kBe exceptional value, after it is rejected, to n-1 observed reading repetition above-mentioned steps of remainder, up to no longer including exceptional value.

Preferably, step 113a) describedly sample is carried out the regularity of distribution verifies as sample distribution is being supposed on the basis, adopt χ ²Whether test of fitness of fot method test-hypothesis distribution is consistent with actual distribution:

31a) hypothesis H is proposed ₀: X obeys certain alternative distribution pattern; Choose 6 continuity stochastic distribution commonly used (be respectively: exponential distribution, evenly distribution, Weibull distribution, normal distribution, rayleigh distributed and gamma distribute) and be alternative distribution pattern, carry out test of hypothesis respectively, contain r unknown number in the distribution function of described distribution pattern, r is a natural number;

32a) real number axis is divided into k disjoint interval (a ₀, a ₁], (a ₁, a ₂] ..., (a _K-1, a _k], wherein: when n≤200, k gets 5,7 respectively ... 13; When n＞200, k gets 11,13 respectively ... 21; a ₀, a _kPress formula (5a) cycle calculations:

a ₀＝x _min-(x _max-x _min)×j×1％，j＝1，2，L?20 (5a)

a _k＝x _max+(x _max-x _min)×j×1％，j＝1，2，L?20

33a) computational data falls into each interval frequency n _i, i=1,2,3...k;

34a) at H ₀Under the condition of setting up, calculate X and fall into each interval Probability p _i:

p _i＝P(a _i-1＜X≤a _i) (6a)

And then obtain theoretical frequency np _i(i=1,2, A, k);

35a) with n _i, np _iThe substitution formula

Obtain χ ²Value;

36a) determine level of significance α _χ, α _χ=0.01, look into χ ²Distribution table gets

37a) if

Then refuse H ₀, otherwise, can accept H ₀

Preferably, step 114a) described characteristic quantity normal region and fault zone computing method are as follows:

41a) each characteristic quantity is chosen unified probable value α, α is that sample falls into interval, normal region [b ₀, a ₀] probability, α=99.99%, and sample falls into interval, fault zone (∞, b ₀) and interval, fault zone (a ₀The probability of ,+∞) is identical, then

P{-∞＜X＜b ₀}+P{b ₀≤X≤a ₀}+P{a ₀＜X＜+∞}＝1 (7a)

P{-∞＜X＜b ₀}＝P{a ₀＜X＜+∞} (8a)

$P\{-\&infin;<X<b_0\}=P\{a_0<X<+\&infin;\}=\frac{1-P\{b_0\&le;X\&le;a_0\}}{2}=\frac{1-\mathrm{\&alpha;}}{2}---\left(9a\right)$

So

$P\{-\&infin;<X<a_0\}=1-\mathrm{\&alpha;}+\frac{1-\mathrm{\&alpha;}}{2}=\frac{3+\mathrm{\&alpha;}}{2}---\left(10a\right)$

42a), calculate a according to the distribution pattern and the parameter of formula (9a), (10a) and this characteristic quantity ₀, b ₀, can obtain interval, normal region [b ₀, a ₀] and interval, fault zone (∞, b ₀) ∪ (a ₀,+∞); For load mean change amount, significant fault zone is (a ₀,+∞);

If 43a) regularity of distribution assay of certain characteristic quantity is to meet a plurality of distributions, then calculate the interval, normal region by a plurality of regularities of distribution respectively, and the union of getting them is this interval, characteristic quantity normal region.

If 44a) test of hypothesis of certain characteristic quantity be can not determine its regularity of distribution, get then that the normal region is interval to be [μ-5 σ, μ+5 σ], wherein Be sample average,

Be the sample standard deviation variance.

45a) make b ₁=b ₂λ (a ₂-b ₂), a ₁=a ₂λ (a ₂-b ₂), λ is a preset threshold, λ=0.2, wherein b ₂, a ₂Be respectively the minimum value and the maximal value of sample.Interval [b ₁, a ₁] with step 41a) to 44a) union in the interval, normal region that calculates is the final interval, normal region [b, a] of this characteristic quantity, then the final interval, fault zone of this characteristic quantity (∞, b) ∪ (a ,+∞).

Preferably, step 121a) the described method of judging whether test sample book itself exists knotting to put is:

51a) the minimum and maximum value of calculating test sample book displacement;

52a) displacement on the test sample book load-position diagram is divided into J part, J=10000, j=1,2, L J-1;

53a) calculate j bar straight line x _j=x _Min+ (x _Max-x _MinThe intersection point set U{H of) * j/J and load-position diagram ₁, H ₂;

54a) obtain H with method of interpolation ₁Corresponding load y _J1

55a) obtain H with method of interpolation ₂Corresponding load y _J2

56a) if | y _J1-y _J2|/y _J1＜ε, ε are preset threshold, and then there is the knotting point in ε=0.001, finishes;

If 57a) j＜J then upgrades j=j+1, jump procedure 53a), otherwise judge that there is not the knotting point in test sample book, finish.

Preferably, step 2a) described search tree is divided into four big classes: area change class, the class of tiing a knot, vibrate class, bump the extension class, wherein the area change class is divided three classes again: area increases class, the minimum class of area, area is less than normal or normal class.The fault type that area class fault search tree is used to discern comprises: standing valve leakage, sucker rod middle and upper part be disconnected takes off, travelling valve is malfunctioning, the leakage of oil pipe bottom, connect take out that band spray, sucker rod bottom are disconnectedly taken off, gas lock, pump seizure, serious feed flow deficiency, oil pipe leakage, travelling valve leakage, that travelling valve cuts out is slow, plunger is deviate from working barrel, feed flow deficiency, gases affect, inertial load are big; The fault type that knotting class fault search tree is used to discern comprises: bump pump, secondary vibration, plunger down and deviate from working barrel; The fault type that vibration class fault search tree is used to discern comprises: shake out, vibrate excessive; Bump and hang the fault type that class fault search tree is used to discern and comprise: on bump extension.

Preferably, step 1) described characteristic quantity safety zone and hazardous location computing method are as follows:

11) each characteristic quantity is chosen unified probable value α ₁, α ₁For sample falls into interval, safety zone [d ₀, c ₀] probability, α ₁=99%, and sample falls into interval, hazardous location (∞, d ₀) and interval, hazardous location (c ₀The probability of ,+∞) is identical, then

P{-∞＜X＜d ₀}+P{d ₀≤X≤c ₀}+P{c ₀＜X＜+∞}＝1 (1)

P{-∞＜X＜d ₀}＝P{c ₀＜X＜+∞} (2)

$P\{-\&infin;<X<d_0\}=P\{c_0<X<+\&infin;\}=\frac{1-P\{d_0\&le;X\&le;c_0\}}{2}=\frac{1-{\mathrm{\&alpha;}}_1}{2}---\left(3\right)$

So $P\{-\&infin;<X<c_0\}=1-{\mathrm{\&alpha;}}_1+\frac{1-{\mathrm{\&alpha;}}_1}{2}=\frac{3+{\mathrm{\&alpha;}}_1}{2}$

(4)

12) according to the distribution pattern and the parameter of formula (3), (4) and this characteristic quantity, calculate c ₀, d ₀, can obtain interval, safety zone [d ₀, c ₀] and interval, hazardous location (∞, d ₀) ∪ (c ₀,+∞); For load mean change amount, significant hazardous location is (c ₀,+∞);

13) if the regularity of distribution assay of certain characteristic quantity is to meet a plurality of distributions, then respectively by between a plurality of regularity of distribution computationally secure area region, and the union of getting them is this interval, characteristic quantity safety zone.

14), get then that the safety zone is interval to be [μ-3 σ, μ+3 σ], wherein if its regularity of distribution is can not determine in the test of hypothesis of certain characteristic quantity

Be sample average,

Be the sample standard deviation variance.

15) make d ₁=d ₂-λ (c ₂-d ₂), c ₁=c ₂+ λ (c ₂-d ₂), λ is a preset threshold, λ=0.1, wherein d ₂, c ₂Be respectively the minimum value and the maximal value of training sample.Interval [d ₁, c ₁] with step 11) to 14) union in the interval, safety zone that calculates is the final interval, safety zone [d, c] of this characteristic quantity, then the final interval, hazardous location of this characteristic quantity (∞, d) ∪ (c ,+∞).

Preferably, step 2) condition of described fault omen is:

(a) certain characteristic quantity of N sample enters a left side or right explosive area, and sample N-4, N-3, and N-2, N-1, this characteristic quantity of N reduces continuously or increases;

(b) N-9, N-8, L, certain characteristic quantity of N sample increases continuously or reduces continuously;

(c) N-4, N-3, N-2, N-1, N sample standard deviation enters the hazardous location;

(d) N-4, N-3, N-2, N-1, there is the knotting point in N sample standard deviation

If satisfy one of above four conditions, judge that then there is the fault omen in the N+1 sample that will predict.

Preferably, step 2) data source of N+1 sample of described prediction is meant if meet the condition of fault omen, continuation is the data source of N+1 sample of prediction to returning a characteristic quantity and a knotting information that whether exists of seeking all samples that meet identical rule continuously.

Preferably, the ARMA time series models are adopted in the prediction of the characteristic quantity of the described forecast sample of step 3), and its modeling and prediction flow process are as follows:

21) obtain the value of the sample coefficient of autocorrelation and the sample PARCOR coefficients of this sequence;

22) come preliminary judgement model parameter p and q according to the truncation or the hangover property of sample coefficient of autocorrelation and sample PARCOR coefficients, carry out model fitting;

23) unknown parameter in the estimation model;

24) model optimization is set up a plurality of models, chooses the more excellent model of match, and predicts.

The thought of ARMA time series forecasting method is: the future of a phenomenon of prediction is when changing, future is predicted in past behavior with this phenomenon, promptly disclose the time dependent rule of phenomenon by the seasonal effect in time series historical data, and this rule extended to future, thereby to making prediction the future of this phenomenon.Arma modeling is to mainly contain three kinds of citation forms: autoregressive model (AR model), moving average model(MA model) (MA model) and ARMA model (arma modeling).

The prediction mode of AR model is the linear combination prediction by the observed reading in past and present interference value:

(5)

In the formula: the exponent number of p-autoregressive model

(i=1,2, Lp)-undetermined coefficient of model

e _i-error time sequence

The prediction mode of MA model is the linear combination prediction by the interference value in past and present interference value:

X _t＝e _t-θ ₁e _t-1-θ ₂e _t-2-L-θ _qe _t-q

(6)

In the formula: the exponent number of q-model

θ _j(j=1,2, L, q)-undetermined coefficient of model

e _i-error

X _t-moving average sequence

Arma modeling (ARMA model) is that basis " mixing " constitutes by AR model and MA model:

It is carried out statistical treatment, can obtain coefficient of autocorrelation:

Coefficient of autocorrelation p ₁Be

And θ ₁Function, autocorrelation function is from p ₁Beginning is exponential damping.If The autocorrelation function exponential damping is level and smooth; If

The autocorrelation function exponential damping is an alternate, vibrates between positive negative value.

Claims (5)

1. the rod pumping system failure prediction method based on load-position diagram is characterized in that comprising the steps:

1) safety zone and the hazardous location of 15 characteristic quantities of calculating:, calculate the safety zone [d, c] and hazardous location (∞ of each characteristic quantity according to the characteristic quantity statistical law, d) ∪ (c ,+∞), the hazardous location is divided into left explosive area (∞, d) and right explosive area (c ,+∞); For load mean change amount, significant hazardous location be right explosive area (c ,+∞);

2) the fault omen is analyzed: whether the sample that first judgement will be predicted (establishing catalogue number(Cat.No.) is N+1) exists the fault omen, as if no omen, does not then need to predict, judges that directly this forecast sample is normal sample, finishes; If omen is arranged, then calculate the data source of N+1 sample of prediction, jump procedure 2);

3) characteristic quantity of forecast sample prediction:, predict 15 characteristic quantities of N+1 sample and whether have the knotting point according to data source;

4) to predicting the outcome, adopt " the rod pumping system fault based on load-position diagram is passed the rank diagnostic method " to carry out fault and differentiate, judge whether N+1 sample be normal, if fault sample further carries out the identification of fault type.

2. according to claim 1 described rod pumping system failure prediction method, it is characterized in that step 1) described characteristic quantity safety zone and hazardous location computing method are as follows based on load-position diagram:

11) each characteristic quantity is chosen unified probable value α ₁, α ₁For sample falls into interval, safety zone [d ₀, c ₀] probability, α ₁=99%, and sample falls into interval, hazardous location (∞, d ₀) and interval, hazardous location (c ₀The probability of ,+∞) is identical, then

P{-∞＜X＜d ₀}+P{d ₀≤X≤c ₀}+P{c ₀＜X＜+∞}＝1 (1)

P{-∞＜X＜d ₀}＝P{c ₀＜X＜+∞} (2)

$P\{-\&infin;<X<d_0\}=P\{c_0<X<+\&infin;\}=\frac{1-P\{d_0\&le;X\&le;c_0\}}{2}=\frac{1-{\mathrm{\&alpha;}}_1}{2}---\left(3\right)$

So $P\{-\&infin;<X<c_0\}=1-{\mathrm{\&alpha;}}_1+\frac{1-{\mathrm{\&alpha;}}_1}{2}=\frac{3+{\mathrm{\&alpha;}}_1}{2}$

(4)

12) according to the distribution pattern and the parameter of formula (3), (4) and this characteristic quantity, calculate c ₀, d ₀, can obtain interval, safety zone [d ₀, c ₀] and interval, hazardous location (∞, d ₀) ∪ (c ₀,+∞); For load mean change amount, significant hazardous location is (c ₀,+∞);

13) if the regularity of distribution assay of certain characteristic quantity is to meet a plurality of distributions, then respectively by between a plurality of regularity of distribution computationally secure area region, and the union of getting them is this interval, characteristic quantity safety zone;

14), get then that the safety zone is interval to be [μ-3 σ, μ+3 σ], wherein if its regularity of distribution is can not determine in the test of hypothesis of certain characteristic quantity

Be sample average,

Be the sample standard deviation variance;

15) make d ₁=d ₂-λ (c ₂-d ₂), c ₁=c ₂+ λ (c ₂-d ₂), λ is a preset threshold, λ=0.1, wherein d ₂, c ₂Be respectively the minimum value and the maximal value of training sample.Interval [d ₁, c ₁] with step 11) to 14) union in the interval, safety zone that calculates is the final interval, safety zone [d, c] of this characteristic quantity, then the final interval, hazardous location of this characteristic quantity (∞, d) ∪ (c ,+∞).

3. according to claim 1 described rod pumping system failure prediction method, it is characterized in that step 2 based on load-position diagram) condition of described fault omen is:

(a) certain characteristic quantity of N sample enters a left side or right explosive area, and sample N-4, N-3, and N-2, N-1, this characteristic quantity of N reduces continuously or increases;

(b) N-9, N-8, L, certain characteristic quantity of N sample increases continuously or reduces continuously;

(c) N-4, N-3, N-2, N-1, N sample standard deviation enters the hazardous location;

(d) N-4, N-3, N-2, N-1, there is the knotting point in N sample standard deviation;

If satisfy one of above four conditions, judge that then there is the fault omen in the N+1 sample that will predict.

4. according to claim 1 described rod pumping system failure prediction method based on load-position diagram, it is characterized in that step 2) data source of N+1 sample of described prediction is meant if meet the condition of fault omen, continuation is the data source of N+1 sample of prediction to returning a characteristic quantity and a knotting information that whether exists of seeking all samples that meet identical rule continuously.

5. according to claim 1 described rod pumping system failure prediction method based on load-position diagram, it is characterized in that the ARMA time series models are adopted in the characteristic quantity prediction of the described forecast sample of step 3), its modeling and prediction flow process are as follows:

21) obtain the value of the sample coefficient of autocorrelation and the sample PARCOR coefficients of this sequence;

22) come preliminary judgement model parameter p and q according to the truncation or the hangover property of sample coefficient of autocorrelation and sample PARCOR coefficients, carry out model fitting;

23) unknown parameter in the estimation model;

24) model optimization is set up a plurality of models, chooses the more excellent model of match, and predicts.