CN110348137A - A kind of water-drive pool seepage field evaluation method based on Vector Autoression Models - Google Patents
A kind of water-drive pool seepage field evaluation method based on Vector Autoression Models Download PDFInfo
- Publication number
- CN110348137A CN110348137A CN201910636131.1A CN201910636131A CN110348137A CN 110348137 A CN110348137 A CN 110348137A CN 201910636131 A CN201910636131 A CN 201910636131A CN 110348137 A CN110348137 A CN 110348137A
- Authority
- CN
- China
- Prior art keywords
- data
- well
- model
- water
- injection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The water-drive pool seepage field evaluation method based on Vector Autoression Models that the invention discloses a kind of belongs to water flooding adjustment field, collects the historical production data and/or history water filling data of target reservoir；Historical production data and/or history water filling data are pre-processed；Data model of fit is filled the water according to pretreated historical production data and/or history, and verifies model；Analysis of uncertainty, while the oil recovery contribution amount according to the model parameter evaluation injection well after fitting are carried out according to the produced quantity in the model prediction extraction well future after fitting, and to the prediction result of model.The present invention, which solves existing seepage field evaluation method and exists, to be calculated higher cost and there is a problem of that convergence is poor under accuracy is lower and complex geological condition.
Description
Technical field
The invention belongs to water floodings to adjust field, be related to a kind of water-drive pool infiltration based on Vector Autoression Models
Flow field evaluation method.
Background technique
Waterflooding development improves the means of recovery ratio as main oil reservoir, has and is extremely widely applied.However it is big at present
After part oil reservoir is developed after long-period water drive, remaining oil distribution is mixed and disorderly and disperses, it is difficult to and effectively understanding water-drive pool employs rule,
Cause seepage field adjustment difficulty big, influences waterflooding development efficiency.
Domestic scholars adjust decision for seepage field and provide branch by determining that seepage field parameter evaluates seepage field
Support, however its method needs to be evaluated by expertise, it is subjective, cause evaluation result accuracy lower.It is external
Scholar's multi-pass crosses the methods of streamline simulation and predicts and optimize water-drive pool water filling system, but its method is under complex geological condition,
Method haves the defects that convergence is poor.
For seepage field, since the prior art can not directly observe underground fluid mobility status, researcher is logical
The mode for crossing conservation of matter equation of the solution based on darcy or non-Darcy's law simulates underground fluid flowing, and then characterizes underground and seep
Flow field.And in reservoir engineer by completing reservoir model-building to geologic feature, while history matching is carried out by numerical simulation, into
After making to one step reservoir model more closing to reality geological condition, it is still necessary to which the model completed based on fitting is determined to seepage field
Waterflooding adjustment method, to further increase sweep efficiency and recovery ratio.For Microreservoir, empirical method can be passed through
Adjustment well operations mode adjusts flow field in turn, however in view of most of oil reservoir has complicated geological conditions and mining method,
The determination of oil reservoir prioritization scheme will become extremely challenge.
Therefore, in view of the above-mentioned problems, the invention proposes a kind of water-drive pool seepage field based on Vector Autoression Models
Evaluation method.
Summary of the invention
It is an object of the invention to: provide a kind of water-drive pool seepage field evaluation side based on Vector Autoression Models
Method, solves existing seepage field evaluation method and exists and calculate higher cost and deposit under accuracy is lower and complex geological condition
In the problem that convergence is poor.
The technical solution adopted by the invention is as follows:
A kind of water-drive pool seepage field evaluation method based on Vector Autoression Models, comprising the following steps:
The historical production data and/or history for collecting target reservoir fill the water data；
Historical production data and/or history water filling data are pre-processed；
Data model of fit is filled the water according to pretreated historical production data and/or history, and verifies model；
According to the produced quantity in the model prediction extraction well future after fitting, and the prediction result of model is carried out uncertain
Analysis, while the oil recovery contribution amount according to the model parameter evaluation injection well after fitting.
Further, the historical production data for collecting target reservoir and/or history fill the water data, specially collect and adopt
The history of the historical production data of well and/or injection well fills the water data out, arranges as the text of " * .xlsx " or " * .csv " format
Part, every row includes at least days, pound sign, daily oil production data and/or daily water-injection rate data in table.
Further, described that historical production data and/or history water filling data are pre-processed, comprising: to daily output
Oil mass data and/or daily water-injection rate data carry out sliding window smoothing techniques and normalized.
Further, described that data model of fit is filled the water according to pretreated historical production data and/or history, and test
Model of a syndrome the following steps are included:
By pretreated daily oil production data and/or daily water-injection rate data configuration at time series format, i.e., every columns
According to the different wells of expression, each row of data indicates the data of different moments；
Data are filled the water according to historical production data and/or history by vector auto regression method, are fitted historical production data,
And reasonable lag order is chosen for model by the method that lag order is chosen；
Choose historical data end some months data as verifying collection, using it before data as training set train mould
Type；
Verifying is collected by the model after training and carries out prediction effect, then the effect of prediction is evaluated, if verifying collection
Prediction effect it is better, show to future time instance produce well production prediction effect it is better and pre- to injection well oil recovery contribution amount
It is higher to survey precision.
Further, described that data are filled the water according to historical production data and/or history by vector auto regression method, intend
Historical production data is closed, and is that model chooses reasonable lag order by the method that lag order is chosen, comprising:
Construct vector are as follows:
Wherein, yi,tFor i-th mouthful of extraction well produced quantity of t moment, unit is m3/ the moon；ei,tFor i-th mouthful of injection well note of t moment
Enter amount, unit is m3/ the moon；kPFor extraction well sum；kIFor injection well sum；YtWell produced quantity vector is produced for t moment,EtFor t moment injection well injection rate vector,
It is as follows that extraction well yield prediction model is established by the correlation of flow between well:
Wherein, AiWell parameter matrix is produced for the i-th rank, for describing the discharge relation between extraction well, BiFor the i-th rank injection well parameter matrix, for describing interactive relation between injection-production well,P is lag order, and characterizing at most previous p months extraction well yields can be to present extraction
Well yield has an impact, pIFor injection well lag order, at most previous p is characterizedIA month injection well yield can be to present extraction
Well yield has an impact, and c is deviation factors,utFor the residual values of t moment,
Model is using of that month and previous p month extraction well daily oil production as Yt+…+Yt-p+1, and with 1 month and its preceding p- following
The diurnal injection in January is as inputPredict lower monthly oil production estimated valueAnd iteration carries out, it may be assumed that
Wherein, the symbol for taking triangle such as estimates parameter matrixIndicate that the symbol is estimated value,
Estimate that parameter matrix is solved by simultaneous linear equations, if n represents Observable data volume, solution procedure are as follows:
Y=DZ+U,
U=[u1, u2..., un],
Wherein, Y is total dependent variable matrix, and Z is total independent variable matrix, and U is total residual matrix, and D is total parameter matrix；
It is solved again by least square method:
Because when well quantity is excessive or lag order is excessive, there may be over-fittings for obtained result, therefore need to be in side
Journey group adds regularization term；
The lag order of target data set is chosen according to lag order selection method again.
Further, the model by after training, which collects verifying, carries out prediction effect, then to the effect of prediction into
Row evaluation, comprising:
Injection well is evaluated by parameter, Appreciation gist be any injection well yield is increased into fixed numbers, if
Evaluation extraction well is then the extraction well yield for increasing fixed numbers, and observing it influences total system, further can be to injection
Well or extraction well are evaluated, injection well judgement schematics are as follows:
Wherein,For diagonal unit matrix,ziIt is injection well in the following i moment oil recovery contribution amount, nothing
Dimension.
Further, the prediction result to model carries out analysis of uncertainty, while according to the model ginseng after fitting
The oil recovery contribution amount of number evaluation injection well, comprising:
Based on the time series process of script is thought of as random process, succinct matrix is expressed as follows:
Wherein,For average forecasting error matrix, n is Observable data volume；
Consider the long lasting effect of parameter:
Wherein,For diagonal unit matrix,ΦiFor the following i-th hours cumulative affecting parameters matrix,∑yIt (h) is future h hours cumulative influence matrix,σjIt (h) will be jth mouth well in future
The h moment is estimated to predict error, wherein σj(h) corresponding extraction well j walks estimation when predicting in h and predicts error；
For VAR model, wherein important hyper parameter is maximum lagged value p, which represent assume most Yt-pAnd
Et-p+1The size of value will affect Yt, the selection of the value completed by information criterion, firstly, computation model is in lagged value p
Likelihood estimator L, and the size by considering likelihood function value and model parameter, provide information criterion, formula is as follows:
AIC=-2ln (L)+2k,
BIC=-2ln (L)+ln (n) k,
HQ=-2ln (L)+ln (ln (n)) k,
Wherein, L is likelihood estimator, and AIC is AIC information criterion evaluation of estimate, and BIC is BIC information criterion evaluation of estimate, and HQ is
HQ information criterion evaluation of estimate, FPE are FPE information criterion evaluation of estimate, and k is model parameter number, by comprehensively considering above four
Information criterion, selection make the lesser lag order of the above criterion value be suitable lag order p.
In conclusion by adopting the above-described technical solution, the beneficial effects of the present invention are:
1. a kind of water-drive pool seepage field evaluation method based on Vector Autoression Models, raw by fitting extraction well history
Data are produced, machine learning model is established, capture any extraction well oil production that may be present, can be to multiple with the relationship of historical data
Miscellaneous data relationship modeling, precision is high, it is short to calculate the time, and can recover the oil by modeling injection well augmented injection effect assessment injection well
Contribution amount, efficiently avoiding numerical simulation fitting under complex geological condition has that convergence is poor.With existing seepage flow
Field evaluation method is compared, and method subjectivity of the present invention is less, without that can be evaluated by expertise, and it is accurate
Degree is higher, and prediction can carry out analysis of uncertainty, it is ensured that the safety and accuracy of prediction result.
2. being adopted in May, 2017 in March, 2018 with the present invention in May, the 2017 pervious data prediction of the oil field the M area X oil reservoir
Well oil production out, precision of prediction evaluate the oil field the M area X numerical value reservoir model according to injection well up to 86.92%, and with the present invention
As a result waterflooding adjustment is carried out, numerical simulation recovery ratio promotes 0.3112% in 2 years, and oil production has more 34770m3, practice have shown that
Show that the accuracy of evaluation result is higher with the present invention.
3. Vector Autoression Models are good at excavating the data relationship between multiple wells in the present invention, it is good at from multiple time sequences
Extracted in column interaction rule, Model suitability is strong, compared with existing seepage field evaluation method, not only calculate cost compared with
It is low, and precision of prediction is higher.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be to needed in the embodiment attached
Figure is briefly described, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not construed as pair
The restriction of range for those of ordinary skill in the art without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings, in which:
Fig. 1 is a kind of flow chart of water-drive pool seepage field evaluation method based on Vector Autoression Models；
Fig. 2 is association's Correlation Moment system of battle formations between the injection-production well of the embodiment of the present invention one；
Fig. 3 is that the embodiment of the present invention one is the highest two pairs of injection-production wells of correlation；
Fig. 4 is the prediction verification result figure of the different extraction wells of the embodiment of the present invention one；
Fig. 5 is the inherent extraction well prediction error condition produced of the verifying collection of the embodiment of the present invention one；
Fig. 6 is the result figure that analysis of uncertainty is carried out to verifying collection prediction result of the embodiment of the present invention one；
Fig. 7 is that the different injection wells of the embodiment of the present invention one influence broken line to the single time step of entirety extraction well oil production
Figure；
Fig. 8 is that the different injection wells of the embodiment of the present invention one produce well oil production accumulated time step-size influences broken line to entirety
Figure；
Fig. 9 is the evaluation result figure of the injection well of the embodiment of the present invention one；
Figure 10 be the embodiment of the present invention one machine learning module be used for produce well production prediction and seepage field evaluation answer
Use interface；
Figure 11 is the lag order analysis module interface of the embodiment of the present invention one；
Figure 12 is the prediction result drafting module interface and impulse response module interface of the embodiment of the present invention one.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention, i.e. described embodiment is a part of the embodiments of the present invention, instead of all the embodiments.It is logical
It is often described herein as to arrange and designing with a variety of different configurations with the component of the embodiment of the present invention shown in the accompanying drawings.
Therefore, the detailed description of the embodiment of the present invention provided in the accompanying drawings is not intended to limit below claimed
The scope of the present invention, but be merely representative of selected embodiment of the invention.Based on the embodiment of the present invention, those skilled in the art
Member's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
It should be noted that the relational terms of term " first " and " second " or the like be used merely to an entity or
Operation is distinguished with another entity or operation, and without necessarily requiring or implying between these entities or operation, there are any
This actual relationship or sequence.Moreover, the terms "include", "comprise" or its any other variant be intended to it is non-exclusive
Property include so that include a series of elements process, method, article or equipment not only include those elements, but also
Further include other elements that are not explicitly listed, or further include for this process, method, article or equipment it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including described
There is also other identical elements in the process, method, article or equipment of element.
A kind of water-drive pool seepage field evaluation method based on Vector Autoression Models, solves existing seepage field evaluation side
Method, which exists, to be calculated higher cost and there is a problem of that convergence is poor under accuracy is lower and complex geological condition.
A kind of water-drive pool seepage field evaluation method based on Vector Autoression Models, comprising the following steps:
Step 1: collecting the historical production data and/or history water filling data of target reservoir；
Step 2: historical production data and/or history water filling data are pre-processed；
Step 3: data model of fit is filled the water according to pretreated historical production data and/or history；
Step 4: according to the produced quantity in the model prediction extraction well future after fitting, and the prediction result of model being carried out not
Deterministic parsing, while the oil recovery contribution amount according to the model parameter evaluation injection well after fitting.
The present invention establishes machine learning model by fitting extraction well historical production data, captures that may be present any
Well oil production is produced with the relationship of historical data, it is short that the time can be calculated, and can pass through to complex data relationship modeling, precision height
Modeling injection well augmented injection effect assessment injection well oil recovery contribution amount, efficiently avoids numerical simulation under complex geological condition
Fitting has that convergence is poor.Compared with existing seepage field evaluation method, method subjectivity of the present invention is less,
Without that can be evaluated by expertise, and accuracy is higher, and prediction can carry out analysis of uncertainty, it is ensured that prediction knot
The safety and accuracy of fruit.
Feature and performance of the invention are described in further detail below with reference to embodiment.
Embodiment one
Presently preferred embodiments of the present invention provides a kind of based on vector auto regression mould by taking the oil field the M area X fault block oil reservoir as an example
The water-drive pool seepage field evaluation method of type, as shown in Figure 1；
Vector auto regression (VAR) is a kind of random process model, mutual between multiple time series datas for capturing
Linear dependence.Aleatory variable can pass through lagged value, certainty variable and the error of its own and its dependent variable in model
The equation of item composition indicates, and deduces to its future value.Model does not need to pass through simultaneous equations using numerical simulation mode
Structural model and the mode that needs variable specifically to act in flow event.Unique priori knowledge needed for VAR method is can
To assume the interactional variable list on across the phase, that is, the optional influence factor of target variable is provided, therefore change need to be observed
Correlation between correlation between amount, i.e. injection well and extraction well (abbreviation injection-production well)
The association's correlation matrix being illustrated in figure 2 between injection-production well, abscissa are injection well, and ordinate is extraction well, face in figure
Color is shallower to represent that correlation is higher, and color is relatively deep then on the contrary, going out to produce the note of the oil production and injection well of well from the figure observable
Enter the correlation between amount, correlation matrix can be seen that from the association, and the flow between most of injection-production well is simultaneously uncorrelated, however small part is infused
The correlation adopted between well is stronger；The highest two pairs of injection-production wells of correlation are illustrated in figure 3, as seen from the figure, part injection-production well is presented
Out with increasing with subtracting the case where, and then may be assumed that injection-production well flow can be characterized by linear relationship, and then can pointedly adopt to note
Well-pattern system modeling, therefore the present invention is using different extraction well productions and injection well injection rate as the time series mistake being relative to each other
Journey establishes vector auto regression (VAR) model, captures yield dependence between well, simulates future production, and locally seep to injection well
It is evaluated in flow field；
It the described method comprises the following steps:
Step 1: collecting the historical production data and/or history water filling data of target reservoir, specially collect extraction well
The history of historical production data and/or injection well fills the water data, arranges as the file of " * .xlsx " or " * .csv " format, table
In every row include at least days, pound sign, daily oil production data and/or daily water-injection rate data；
Step 2: to historical production data and/or history water filling data pre-process, specifically to daily oil production data and/
Or daily water-injection rate data carry out sliding window smoothing techniques and normalized, can increase the consistent level of data, and it is difficult to reduce fitting
Degree；
Step 3: data model of fit being filled the water according to pretreated historical production data and/or history, and verifies model；
Step 3.1: by pretreated daily oil production data and/or daily water-injection rate data configuration at time series format,
I.e. each column data indicate that different wells, each row of data indicate the data of different moments；
Step 3.2: data being filled the water according to historical production data and/or history by vector auto regression method, are fitted history
Creation data, and be that model chooses reasonable lag order by the method that lag order is chosen；
Construct vector are as follows:
Wherein, yi,tFor i-th mouthful of extraction well produced quantity of t moment, unit is m3/ the moon；ei,tFor i-th mouthful of injection well note of t moment
Enter amount, unit is m3/ the moon；kPFor extraction well sum；kIFor injection well sum；YtWell produced quantity vector is produced for t moment,EtFor t moment injection well injection rate vector,
It is as follows that extraction well yield prediction model is established by the correlation of flow between well:
Wherein, AiWell parameter matrix is produced for the i-th rank, for describing the discharge relation between extraction well, BiFor the i-th rank injection well parameter matrix, for describing interactive relation between injection-production well,P is lag order, and characterizing at most previous p months extraction well yields can be to present extraction
Well yield has an impact, pIFor injection well lag order, at most previous p is characterizedIA month injection well yield can be to present extraction
Well yield has an impact, and c is deviation factors,utFor the residual values of t moment,
Model is using of that month and previous p month extraction well daily oil production as Yt+…+Yt-p+1, and with 1 month and its preceding p- following
The diurnal injection in January is as inputPredict lower monthly oil production estimated valueAnd iteration carries out, it may be assumed that
Wherein, the symbol for taking triangle such as estimates parameter matrixIndicate that the symbol is estimated value,
Estimate that parameter matrix is solved by simultaneous linear equations, if n represents Observable data volume, solution procedure are as follows:
Y=DZ+U,
U=[u1, u2..., un],
Wherein, Y is total dependent variable matrix, and Z is total independent variable matrix, and U is total residual matrix, and D is total parameter matrix；
It is solved again by least square method:
Because when well quantity is excessive or lag order is excessive, there may be over-fittings for obtained result, therefore need to be in side
Journey group adds regularization term；
In the present embodiment, the lag order for choosing target data set according to lag order selection method is 6；
Step 3.3: choose historical data end some months data as verifying collection, using it before data as train
Collect training pattern, if the longest time of data is 2018, is arranged using pervious data in 2017 as training set, 2017
Later data are verifying collection；
In May, 2017 pervious data are chosen in the oil field the M area X historical production data in the present embodiment as training set,
In May, 2017 and data later verify model prediction result as verifying collection；
It is illustrated in figure 4 the prediction verification result of the different extraction wells of the present embodiment, Fig. 4 (a) is western 43-6-4 and western 44-
The training set data of 5-2, Fig. 4 (b) are the training set data of western 44-8-2 and western 46-5-1, and Fig. 4 (c) is western 1-9-2 and western 42-
The training set data of 8-1, Fig. 4 (d) are the training set data of western 49-5-3 and western 50-5-1, and figure dotted line indicates training set number
According to difference label represents verifying collection data, and solid line represents the verifying collection data of prediction, it is seen then that original training set data fluctuation is larger,
Certain challenge is brought to prediction, further analysis verifying collects the extraction well prediction case that interior yield is greater than 0；
Step 3.4: verifying is collected by the model after training and carries out prediction effect, then the effect of prediction is evaluated,
If the prediction effect for verifying collection is better, show that the prediction effect that well production is produced to future time instance is better, and recover the oil to injection well
Contribution amount precision of prediction is higher；
Injection well is evaluated by parameter, Appreciation gist is that any injection well yield is increased fixed numbers (usually
For 1), if evaluation extraction well is then the extraction well yield for increasing fixed numbers, observing it influences total system, further it is
Injection well or extraction well can be evaluated, injection well judgement schematics are as follows:
Wherein,For diagonal unit matrix,ziIt is injection well in the following i moment oil recovery contribution amount, nothing
Dimension；
It is illustrated in figure 5 the inherent extraction well prediction error condition produced of verifying collection, it is to test that yellow, which is histogram, in histogram
Card collects interior actual average yield, and blue is prediction average product, and lower section linear graph is Relative Error value, it is seen then that removes few portion
Due to implementing well stimulation in the recent period, prediction result is relatively low compared with actual result for point extraction well, other extraction wells be fitted compared with
It is good, wherein it is 0.513392 that verifying, which concentrates on the extraction well mean absolute error value produced, average relative error 0.130811, table
Bright fitting effect is preferable；
Step 4: according to the produced quantity in the model prediction extraction well future after fitting, and the prediction result of model being carried out not
Deterministic parsing, while the oil recovery contribution amount according to the model parameter evaluation injection well after fitting；
Since given data volume is smaller, prediction result still has certain error, and uncertainty analysis can be provided because previous
Model error size caused by data, and model is provided for the certainty size of prediction, safer prediction is carried out, that is, is existed
Therefore careful decision when uncertainty in traffic is larger carries out analysis of uncertainty to model prediction and evaluation result and is necessary,
The derivation of uncertainty in traffic size is based primarily upon, the time series process of script is thought of as random process, succinct square
Matrix representation is as follows:
Wherein,For average forecasting error matrix, n is Observable data volume；
Consider the long lasting effect of parameter:
Wherein,For diagonal unit matrix,ΦiFor the following i-th hours cumulative affecting parameters matrix,∑yIt (h) is future h hours cumulative influence matrix,σjIt (h) will be jth mouth well in future
The h moment is estimated to predict error, wherein σj(h) corresponding extraction well j walks estimation when predicting in h and predicts error；
For VAR model, wherein important hyper parameter is maximum lagged value p, which represent assume most Yt-pAnd
Et-p+1The size of value will affect Yt, the selection of the value completed by information criterion, firstly, computation model is in lagged value p
Likelihood estimator L, and the size by considering likelihood function value and model parameter, provide information criterion, formula is as follows:
AIC=-2ln (L)+2k,
BIC=-2ln (L)+ln (n) k,
HQ=-2ln (L)+ln (ln (n)) k,
Wherein, L is likelihood estimator, and AIC is AIC information criterion evaluation of estimate, and BIC is BIC information criterion evaluation of estimate, and HQ is
HQ information criterion evaluation of estimate, FPE are FPE information criterion evaluation of estimate, and k is model parameter number, by comprehensively considering above four
Information criterion, selection make the lesser lag order of the above criterion value be suitable lag order p；
It is illustrated in figure 6 and analysis of uncertainty result figure is carried out to verifying collection prediction result, put as training data, pitch to test
Card collection data, solid line are to the prediction result of verifying collection data, and area is range of indeterminacy, it is seen that the prediction of difference extraction well
Range of indeterminacy contains practical situation substantially；Meanwhile being mentioned by uncertain formula, uncertainty in traffic size master
To predict that error and parameter determine by training set, therefore uncertainty in traffic is larger when training set prediction error is larger, such as Fig. 6
(c) and shown in (d).
Seepage field evaluation based on machine learning method is then by assuming that certain injection well increases when monthly average day fluence
1m3, observing it influences other extraction wells, further, calculates different injection wells to whole oil reservoir yield by this method
It influences, line chart is influenced on entirety extraction well oil production for different injection wells as shown in Figure 7, Figure 8, Fig. 7 is single time step
It influences, Fig. 8 is accumulated time step-size influences, and in figure, ordinate is to influence, and unit is m3/ d increases entirety if 1 representative
Produce the total 1m of well day oil production3/ d, abscissa is the time, and unit is the moon, it is seen that single time step influences increase at any time
It is gradually decreasing.The figure that Fig. 8 is designated as 2 gives the cumulative effect of different extraction wells, it is seen then that most of to infuse when step-length is 10
The influence for entering well converges to 0 or so, therefore cumulative effect when being 10 herein according to step-length comments the value of injection well
Valence is illustrated in figure 9 the evaluation result of injection well, indicates being brought to different injection wells injection 1m3 injection water for model prediction
Oil recovery contribution amount, foundation can be provided according to the result for seepage field Adjusted Option.
Method of the invention is realized into machine learning functions of modules by Python programming language, guarantees the practicability of algorithm
With convenience, theory support can be provided for seepage field structure adjusting by the evaluation result of software, and further be promoted complicated high
Aqueous oil reservoir waterflood efficiency and development degree；
Machine learning module is used to produce the application of well production prediction and seepage field evaluation, interface is as shown in Figure 10,
Whether the application may be selected to consider to inject in machine learning model by reading the extraction well weekly or monthly magazine data in oil field
Well influences, and assumes that target reservoir is not injected into well if not considering.Injection well influence is considered herein.Further,
It needs to choose model suitable lag order, is as shown in figure 11 the module interfaces analyzed lag order, illustrates target
Reservoir Data corresponds to four information criterions, it is seen that when lag order is 6, wherein three information criterion values are minimum value, therefore
Choosing lag order is 6 relatively reasonable, further, chooses the pervious production oil field the M area X oil reservoir in May, 2017 and water filling number
According to as training set, in May, 2017 and data later are verified model prediction result, the module is such as verifying collection
Shown in Figure 12.
The present invention has innovatively used vector auto regression method to predict extraction well oil production, and can tie to prediction
Fruit carries out analysis of uncertainty, improves the accuracy and convergence of method prediction.Method can calculate injection 1m simultaneously3Inject water
Shi Butong injection well oil recovery contribution amount provides foundation to evaluate the exploitation potential of injection well control area for seepage field adjustment.
Well historical production data is produced by fitting, establishes machine learning model, it is same to capture any extraction well oil production that may be present
The relationship of historical data, can be to complex data relationship modeling, and precision is high, it is short to calculate the time, and can pass through modeling injection well
Augmented injection effect assessment injection well oil recovery contribution amount, efficiently avoiding numerical simulation fitting under complex geological condition, there are convergences
Poor problem.Compared with existing seepage field evaluation method, method subjectivity of the present invention is less, without passing through by expert
Testing can be evaluated, and accuracy is higher, and prediction can carry out analysis of uncertainty, it is ensured that the safety and standard of prediction result
True property.In May, 2017 in March, 2018 well is produced in May, the 2017 pervious data prediction of the oil field the M area X oil reservoir with the present invention
Oil production, precision of prediction up to 86.92%, and with the present invention to the oil field the M area X numerical value reservoir model according to injection well evaluation result
Waterflooding adjustment is carried out, numerical simulation recovery ratio promotes 0.3112% in 2 years, and oil production has more 34770m3, practice have shown that with this
Invention show that the accuracy of evaluation result is higher.
It should be noted that since Figure of description must not colour and alter, so present invention middle part subregion is not apparent
Place is relatively difficult to show, if necessary, can provide color image.
The foregoing is merely illustrative of the preferred embodiments of the present invention, the protection scope being not intended to limit the invention, any
Those skilled in the art within the spirit and principles in the present invention made by any modifications, equivalent replacements, and improvements etc.,
It should all be included in the protection scope of the present invention.
Claims (7)
1. a kind of water-drive pool seepage field evaluation method based on Vector Autoression Models, which comprises the following steps:
The historical production data and/or history for collecting target reservoir fill the water data；
Historical production data and/or history water filling data are pre-processed；
Data model of fit is filled the water according to pretreated historical production data and/or history, and verifies model；
Uncertain point is carried out according to the produced quantity in the model prediction extraction well future after fitting, and to the prediction result of model
Analysis, while the oil recovery contribution amount according to the model parameter evaluation injection well after fitting.
2. a kind of water-drive pool seepage field evaluation method based on Vector Autoression Models according to claim 1, special
Sign is that the historical production data for collecting target reservoir and/or history fill the water data, specially collects the history of extraction well
The history of creation data and/or injection well fills the water data, arranges as the file of " * .xlsx " or " * .csv " format, every in table
Row includes at least days, pound sign, daily oil production data and/or daily water-injection rate data.
3. a kind of water-drive pool seepage field evaluation method based on Vector Autoression Models according to claim 2, special
Sign is, described to pre-process to historical production data and/or history water filling data, comprising: to daily oil production data and/or
Daily water-injection rate data carry out sliding window smoothing techniques and normalized.
4. a kind of water-drive pool seepage field evaluation method based on Vector Autoression Models according to claim 1, special
Sign is, described to fill the water data model of fit according to pretreated historical production data and/or history, and verifies model and include
Following steps:
By pretreated daily oil production data and/or daily water-injection rate data configuration at time series format, i.e. each column tables of data
Show that different wells, each row of data indicate the data of different moments；
Data are filled the water according to historical production data and/or history by vector auto regression method, are fitted historical production data, and lead to
The method for crossing lag order selection is that model chooses reasonable lag order；
The data of historical data end some months are chosen as verifying collection, using it before data as training set training pattern；
Verifying is collected by the model after training and carries out prediction effect, then the effect of prediction is evaluated, if verifying collection is pre-
Survey effect is better, shows that the prediction effect that well production is produced to future time instance is better, and predict essence to injection well oil recovery contribution amount
Du Genggao.
5. a kind of water-drive pool seepage field evaluation method based on Vector Autoression Models according to claim 4, special
Sign is that described to fill the water data according to historical production data and/or history by vector auto regression method, fitting history produces number
According to, and be that model chooses reasonable lag order by the method that lag order is chosen, comprising:
Construct vector are as follows:
Wherein, yi,tFor i-th mouthful of extraction well produced quantity of t moment, unit is m3/ the moon；ei,tFor i-th mouthful of injection well injection rate of t moment,
Unit is m3/ the moon；kPFor extraction well sum；kIFor injection well sum；YtWell produced quantity vector is produced for t moment,
EtFor t moment injection well injection rate vector,
It is as follows that extraction well yield prediction model is established by the correlation of flow between well:
Wherein, AiWell parameter matrix is produced for the i-th rank, for describing the discharge relation between extraction well, BiFor the i-th rank injection well parameter matrix, for describing interactive relation between injection-production well,P
For lag order, present extraction well yield can be had an impact by characterizing at most previous p months extraction well yields, pIFor injection
Well lag order characterizes at most previous pIA month injection well yield can have an impact present extraction well yield, and c is deviation
The factor,utFor the residual values of t moment,
Model is using of that month and previous p month extraction well daily oil production as Yt+…+Yt-p+1, and with following 1 month and its preceding p-1 month
Diurnal injection as inputPredict lower monthly oil production estimated valueAnd iteration carries out, it may be assumed that
Wherein, the symbol for taking triangle such as estimates parameter matrixIndicate that the symbol is estimated value, estimation
Parameter matrix is solved by simultaneous linear equations, if n represents Observable data volume, solution procedure are as follows:
Y=DZ+U,
U=[u1,u2,…,un],
Wherein, Y is total dependent variable matrix, and Z is total independent variable matrix, and U is total residual matrix, and D is total parameter matrix；
It is solved again by least square method:
Because when well quantity is excessive or lag order is excessive, there may be over-fittings for obtained result, therefore need to be in equation group
Add regularization term；
The lag order of target data set is chosen according to lag order selection method again.
6. a kind of water-drive pool seepage field evaluation method based on Vector Autoression Models according to claim 4, special
Sign is that the model by after training, which collects verifying, carries out prediction effect, then evaluates the effect of prediction, comprising:
Injection well is evaluated by parameter, Appreciation gist is that any injection well yield is increased fixed numbers, if evaluation
Extraction well is then the extraction well yield for increasing fixed numbers, and observing it influences total system, further can to injection well or
Extraction well is evaluated, injection well judgement schematics are as follows:
Wherein,For diagonal unit matrix,ziIt is injection well in the following i moment oil recovery contribution amount, it is immeasurable
Guiding principle.
7. a kind of water-drive pool seepage field evaluation method based on Vector Autoression Models according to claim 1, special
Sign is that the prediction result to model carries out analysis of uncertainty, while evaluating injection according to the model parameter after fitting
The oil recovery contribution amount of well, comprising:
Based on the time series process of script is thought of as random process, succinct matrix is expressed as follows:
Wherein,For average forecasting error matrix, n is Observable data volume；
Consider the long lasting effect of parameter:
Wherein,For diagonal unit matrix,ΦiFor the following i-th hours cumulative affecting parameters matrix,∑yIt (h) is future h hours cumulative influence matrix,σjIt (h) will be jth mouth well in future
The h moment is estimated to predict error, wherein σj(h) corresponding extraction well j walks estimation when predicting in h and predicts error；
For VAR model, wherein important hyper parameter is maximum lagged value p, which represent assume most Yt-pAnd Et-p+1's
The size of value will affect Yt, the selection of the value is completed by information criterion, firstly, likelihood of the computation model in lagged value p is estimated
Evaluation L, and the size by considering likelihood function value and model parameter, provide information criterion, formula is as follows:
AIC=-2ln (L)+2k,
BIC=-2ln (L)+ln (n) k,
HQ=-2ln (L)+ln (ln (n)) k,
Wherein, L is likelihood estimator, and AIC is AIC information criterion evaluation of estimate, and BIC is BIC information criterion evaluation of estimate, and HQ is HQ letter
Criterion evaluation of estimate is ceased, FPE is FPE information criterion evaluation of estimate, and k is model parameter number, by comprehensively considering above four information
Criterion, selection make the lesser lag order of the above criterion value be suitable lag order p.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910636131.1A CN110348137B (en) | 2019-07-15 | 2019-07-15 | Water-drive reservoir seepage field evaluation method based on vector autoregressive model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910636131.1A CN110348137B (en) | 2019-07-15 | 2019-07-15 | Water-drive reservoir seepage field evaluation method based on vector autoregressive model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110348137A true CN110348137A (en) | 2019-10-18 |
| CN110348137B CN110348137B (en) | 2022-04-22 |
Family
ID=68176265
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910636131.1A Active CN110348137B (en) | 2019-07-15 | 2019-07-15 | Water-drive reservoir seepage field evaluation method based on vector autoregressive model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110348137B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111236901A (en) * | 2020-01-09 | 2020-06-05 | 西南石油大学 | Water drive reservoir seepage field heterogeneity evaluation method based on flow field diagnosis |
| CN111749687A (en) * | 2020-07-23 | 2020-10-09 | 中海石油国际能源服务(北京)有限公司 | Method, device and equipment for determining main force horizon of multilayer oil reservoir and storage medium |
| CN111832231A (en) * | 2020-07-07 | 2020-10-27 | 中海石油（中国）有限公司天津分公司 | Injection and production well correlation analysis method and system |
| CN112282714A (en) * | 2020-11-30 | 2021-01-29 | 河海大学 | Full well network water injection development optimization method based on deep learning and graph theory |
| CN112412444A (en) * | 2020-11-20 | 2021-02-26 | 中国石油天然气股份有限公司 | Injection-production communication strength determination method and device, computer equipment and storage medium |
| CN112464552A (en) * | 2020-10-28 | 2021-03-09 | 山东大学 | Method and system suitable for predicting seepage field time sequence of underground water-sealed oil storage |
| CN113128734A (en) * | 2019-12-30 | 2021-07-16 | 北京国双科技有限公司 | Method and device for predicting oil field yield |
| CN115935834A (en) * | 2023-02-23 | 2023-04-07 | 中国石油大学(华东) | History fitting method based on deep autoregressive network and continuous learning strategy |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070016389A1 (en) * | 2005-06-24 | 2007-01-18 | Cetin Ozgen | Method and system for accelerating and improving the history matching of a reservoir simulation model |
| US20090043555A1 (en) * | 2007-08-06 | 2009-02-12 | Daniel Busby | Method for Evaluating an Underground Reservoir Production Scheme Taking Account of Uncertainties |
| US20090125288A1 (en) * | 2005-11-26 | 2009-05-14 | The University Court Of The University Of Edinburgh | Hydrocarbon Recovery From a Hydrocarbon Reservoir |
| CN109447532A (en) * | 2018-12-28 | 2019-03-08 | 中国石油大学(华东) | A kind of oil reservoir inter well connectivity based on data-driven determines method |
| CN109948272A (en) * | 2019-03-27 | 2019-06-28 | 长江大学 | Tune based on inter well connectivity blocks up dynamic prediction method and system |
-
2019
- 2019-07-15 CN CN201910636131.1A patent/CN110348137B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070016389A1 (en) * | 2005-06-24 | 2007-01-18 | Cetin Ozgen | Method and system for accelerating and improving the history matching of a reservoir simulation model |
| US20090125288A1 (en) * | 2005-11-26 | 2009-05-14 | The University Court Of The University Of Edinburgh | Hydrocarbon Recovery From a Hydrocarbon Reservoir |
| US20090043555A1 (en) * | 2007-08-06 | 2009-02-12 | Daniel Busby | Method for Evaluating an Underground Reservoir Production Scheme Taking Account of Uncertainties |
| CN109447532A (en) * | 2018-12-28 | 2019-03-08 | 中国石油大学(华东) | A kind of oil reservoir inter well connectivity based on data-driven determines method |
| CN109948272A (en) * | 2019-03-27 | 2019-06-28 | 长江大学 | Tune based on inter well connectivity blocks up dynamic prediction method and system |
Non-Patent Citations (2)
| Title |
|---|
| ZHANG RUI 等: "Production performance forecasting method based onmultivariate time series and vector autoregressive machine learning model for waterflooding reservoirs", 《PETROLEUM EXPLORATION AND DEVELOPMENT》 * |
| 张瑞 等: "基于多变量时间序列及向量自回归机器学习模型的水驱油藏产量预测方法", 《石油勘探与开发》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113128734A (en) * | 2019-12-30 | 2021-07-16 | 北京国双科技有限公司 | Method and device for predicting oil field yield |
| CN111236901A (en) * | 2020-01-09 | 2020-06-05 | 西南石油大学 | Water drive reservoir seepage field heterogeneity evaluation method based on flow field diagnosis |
| CN111832231A (en) * | 2020-07-07 | 2020-10-27 | 中海石油（中国）有限公司天津分公司 | Injection and production well correlation analysis method and system |
| CN111832231B (en) * | 2020-07-07 | 2024-03-26 | 中海石油（中国）有限公司天津分公司 | Injection well correlation analysis method and system |
| CN111749687A (en) * | 2020-07-23 | 2020-10-09 | 中海石油国际能源服务(北京)有限公司 | Method, device and equipment for determining main force horizon of multilayer oil reservoir and storage medium |
| CN111749687B (en) * | 2020-07-23 | 2023-11-21 | 中海石油国际能源服务(北京)有限公司 | Multi-layer oil reservoir principal force horizon determination method, device, equipment and storage medium |
| CN112464552A (en) * | 2020-10-28 | 2021-03-09 | 山东大学 | Method and system suitable for predicting seepage field time sequence of underground water-sealed oil storage |
| CN112464552B (en) * | 2020-10-28 | 2022-10-25 | 山东大学 | Method and system suitable for predicting seepage field time sequence of underground water-sealed oil storage |
| CN112412444B (en) * | 2020-11-20 | 2023-07-25 | 中国石油天然气股份有限公司 | Injection and production communication strength determining method and device, computer equipment and storage medium |
| CN112412444A (en) * | 2020-11-20 | 2021-02-26 | 中国石油天然气股份有限公司 | Injection-production communication strength determination method and device, computer equipment and storage medium |
| CN112282714A (en) * | 2020-11-30 | 2021-01-29 | 河海大学 | Full well network water injection development optimization method based on deep learning and graph theory |
| CN112282714B (en) * | 2020-11-30 | 2022-03-25 | 河海大学 | Full well network water injection development optimization method based on deep learning and graph theory |
| CN115935834A (en) * | 2023-02-23 | 2023-04-07 | 中国石油大学(华东) | History fitting method based on deep autoregressive network and continuous learning strategy |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110348137B (en) | 2022-04-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110348137A (en) | A kind of water-drive pool seepage field evaluation method based on Vector Autoression Models | |
| CN103380424B (en) | For at the system and method using artificial neural network to simulate pipeline hydraulics in reservoir simulation | |
| US8793111B2 (en) | Automated field development planning | |
| CN103917743A (en) | Statistical reservoir model based on detected flow events | |
| EP1592989A1 (en) | Method for computing complexity, confidence and technical maturity indices for reservoir evaluations | |
| CN114154427B (en) | Volume fracturing fracture expansion prediction method and system based on deep learning | |
| Du et al. | The connectivity evaluation among wells in reservoir utilizing machine learning methods | |
| CN114638401A (en) | Residual oil distribution prediction method and device based on history and prediction oil reservoir knowledge | |
| US8311743B2 (en) | Method for generating an estimation of incremental recovery from a selected enhanced oil recovery (EOR) process | |
| Khazaeni et al. | Intelligent production modeling using full field pattern recognition | |
| CN103628816A (en) | Seismic attribute guidance horizontal-well-steering method and device thereof | |
| RU2590278C1 (en) | Methods and systems for controlling negative mobility of components in reservoir simulation | |
| Fenik et al. | Criteria for ranking realizations in the investigation of SAGD reservoir performance | |
| Barros et al. | Quantitative decision analysis for CO2 storage conformance management: A synthetic case study at Smeaheia, North Sea | |
| Tømmerås et al. | Prewell and postwell predictions of oil and gas columns using an iterative Monte Carlo technique with three-dimensional petroleum systems modeling | |
| Lee | Investigating statistical modeling approaches for reservoir characterization in waterfloods from rates fluctuations | |
| CN113032996B (en) | Water channeling channel identification method for hypotonic fractured reservoir horizontal well | |
| Liu et al. | Physics-Coupled Machine Learning Toolset for Geological Carbon Storage Evaluation and Performance Analysis | |
| US20230323772A1 (en) | Generating a reservoir performance forecast | |
| Leng et al. | A physics-based method for forecasting interwell connectivity using injection data | |
| Mittermeir et al. | Automated determination of aquifer properties from field production data | |
| Shaker | Closed-Loop Reservoir Management for Thermal Recovery Processes | |
| Chang | Decision-making for well placement optimization in oil field development | |
| Al-Husseini et al. | History Matching of Reservoir Simulation Model: A Case Study from the Mishrif Reservoir, Buzurgan Oilfield, Iraq | |
| Wanderley De Holanda | Material Balance Reservoir Models Derived from Production Data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |