CN110348137B - Water-drive reservoir seepage field evaluation method based on vector autoregressive model - Google Patents

Water-drive reservoir seepage field evaluation method based on vector autoregressive model Download PDF

Info

Publication number
CN110348137B
CN110348137B CN201910636131.1A CN201910636131A CN110348137B CN 110348137 B CN110348137 B CN 110348137B CN 201910636131 A CN201910636131 A CN 201910636131A CN 110348137 B CN110348137 B CN 110348137B
Authority
CN
China
Prior art keywords
data
production
well
model
historical
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.)
Active
Application number
CN201910636131.1A
Other languages
Chinese (zh)
Other versions
CN110348137A (en
Inventor
贾虎
邓力珲
张瑞
卞小强
王宏申
石端胜
刘常清
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southwest Petroleum University
Original Assignee
Southwest Petroleum University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Southwest Petroleum University filed Critical Southwest Petroleum University
Priority to CN201910636131.1A priority Critical patent/CN110348137B/en
Publication of CN110348137A publication Critical patent/CN110348137A/en
Application granted granted Critical
Publication of CN110348137B publication Critical patent/CN110348137B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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 invention discloses a water drive reservoir seepage field evaluation method based on a vector autoregressive model, which belongs to the field of oilfield flooding development and adjustment and is used for collecting historical production data and historical flooding data of a target reservoir; preprocessing historical production data and historical water injection data; fitting a model according to the preprocessed historical production data and the preprocessed historical water injection data, and verifying the model; the future production amount of the production well is predicted according to the fitted model, uncertainty analysis is carried out on the prediction result of the model, and meanwhile the oil production contribution amount of the injection well is evaluated according to the fitted model parameters.

Description

Water-drive reservoir seepage field evaluation method based on vector autoregressive model
Technical Field
The invention belongs to the field of oilfield flooding development and adjustment, and relates to a water drive reservoir seepage field evaluation method based on a vector autoregressive model.
Background
Water flooding development is used as a main means for improving the recovery ratio of oil reservoirs, and has extremely wide application. However, most of the oil reservoirs are disorderly and scattered in distribution after long-term water drive development, and the water drive reservoir utilization rule is difficult to be effectively known, so that the adjustment difficulty of a seepage field is large, and the water drive development efficiency is influenced.
The domestic scholars evaluate the seepage field by determining the seepage field influence parameters to provide support for the adjustment decision of the seepage field, but the method needs to evaluate by depending on expert experience, so that the subjectivity is high, and the accuracy of the evaluation result is low. Foreign scholars mostly predict and optimize the water flooding oil reservoir water injection system by methods such as streamline simulation, but the method has the defect of poor convergence under complex geological conditions.
For the seepage field, because the flow condition of underground fluid cannot be directly observed in the prior art, researchers simulate the flow of the underground fluid by solving a substance conservation equation based on Darcy or non-Darcy law so as to further characterize the underground seepage field. And after an oil reservoir engineer completes oil reservoir modeling on the geological characteristics and performs historical fitting through numerical simulation, the oil reservoir model is further closer to the actual geological condition, and a water injection adjustment method for a seepage field still needs to be determined based on the model completed by fitting so as to further improve sweep efficiency and recovery ratio. For small reservoirs, the well operation mode and thus the flow field can be adjusted by an empirical method, however, considering that most reservoirs have complex geological conditions and exploitation modes, the determination of reservoir optimization schemes becomes extremely challenging.
Therefore, aiming at the problems, the invention provides a water drive reservoir seepage field evaluation method based on a vector autoregressive model.
Disclosure of Invention
The invention aims to: the method for evaluating the seepage field of the water-drive reservoir based on the vector autoregressive model solves the problems that the existing seepage field evaluation method is high in calculation cost and low in accuracy, and the convergence is poor under complex geological conditions.
The technical scheme adopted by the invention is as follows:
a water drive reservoir seepage field evaluation method based on a vector autoregressive model comprises the following steps:
collecting historical production data and historical water injection data of a target oil reservoir;
preprocessing historical production data and historical water injection data;
fitting a model according to the preprocessed historical production data and the preprocessed historical water injection data, and verifying the model;
and predicting the future production amount of the production well according to the fitted model, carrying out uncertainty analysis on the prediction result of the model, and evaluating the oil production contribution amount of the injection well according to the fitted model parameters.
Further, the historical production data and the historical waterflood data of the target oil reservoir are collected, specifically, the historical production data of the production well and the historical waterflood data of the injection well are collected and arranged into files in the format of 'xlsx' or 'csv', and each row in the table at least comprises year and month, well number, daily oil production data and daily waterflood data.
Further, the preprocessing the historical production data and the historical water injection data comprises the following steps: and performing sliding window smoothing and normalization processing on the daily oil production data and the daily water injection data.
Further, the fitting a model according to the preprocessed historical production data and the preprocessed historical water injection data, and the verifying the model comprises the following steps:
constructing the preprocessed daily oil yield data and daily water injection data into a time series format, namely each line of data represents different wells, and each line of data represents data at different moments;
fitting the historical production data according to the historical production data and the historical water injection data by a vector autoregressive method, and selecting a reasonable hysteresis order for the model by a hysteresis order selection method;
selecting data of several months at the end of the historical data as a verification set, and taking the previous data as a training set to train a model;
and predicting the effect of the verification set through the trained model, and evaluating the predicted effect, wherein if the predicted effect of the verification set is better, the better the predicted effect of the yield of the production well at the future moment is, and the prediction precision of the contribution to the oil production of the injection well is higher.
Furthermore, the fitting the historical production data according to the historical production data and the historical water injection data by the vector autoregressive method, and selecting a reasonable hysteresis order for the model by a hysteresis order selection method comprises the following steps:
the construction vector is:
Figure GDA0003528301320000021
wherein, yi,tThe yield of the ith extraction well at the moment t is m3A month; e.g. of the typei,tThe injection quantity of the ith injection well at the t moment is m3A month; k is a radical ofPThe total number of the producing wells; k is a radical ofITotal number of injection wells; y istIs the vector of the produced quantity of the producing well at the moment t,
Figure GDA0003528301320000022
Etfor the injection well injection quantity vector at time t,
Figure GDA0003528301320000023
the flow prediction model of the produced well is established according to the mutual relation of the flow among wells as follows:
Figure GDA0003528301320000024
wherein A isiIs a production well parameter matrix of the ith order and is used for describing the flow relation among production wells,
Figure GDA0003528301320000025
Figure GDA0003528301320000026
Biis an ith-order injection well parameter matrix used for describing the interaction relationship among injection wells and extraction wells,
Figure GDA0003528301320000027
p is a hysteresis order, representing that the production well flow of at most P months in the past can have an effect on the current production well flow, PICharacterization of Up to past p for injector well hysteresis orderIMonthly injection well flow rates can be made to the current production well flow rateThe influence, c is a deviation factor,
Figure GDA0003528301320000031
utfor the residual value at the time t,
Figure GDA0003528301320000032
the model takes the daily oil production of the production well in the current month and the past p months as Yt+…+Yt-p+1And taking daily injection amount of 1 month in the future and p-1 month before the future as input
Figure GDA0003528301320000033
Estimate value for predicting oil production in next month
Figure GDA0003528301320000034
And are performed iteratively, i.e.:
Figure GDA0003528301320000035
wherein the symbols with triangles are estimated as parameter matrices
Figure GDA0003528301320000036
The symbol is expressed as an estimated value, the estimation parameter matrix is solved through a simultaneous linear equation set, n is assumed to represent observable data quantity, and the solving process is as follows:
Y＝DZ+U，
Figure GDA0003528301320000037
Figure GDA0003528301320000038
U＝[u1,u2,…,un]，
Figure GDA0003528301320000039
Figure GDA00035283013200000310
Figure GDA00035283013200000311
Figure GDA00035283013200000312
wherein Y is a total dependent variable matrix, Z is a total independent variable matrix, U is a total residual error matrix, and D is a total parameter matrix;
and solving by a least square method:
Figure GDA0003528301320000041
because the obtained result may generate overfitting when the number of wells is too large or the hysteresis order is too large, a regularization term needs to be added to the equation set;
and selecting the lag order of the target data set according to a lag order selection method.
Furthermore, the predicting the effect of the trained model on the verification set and then evaluating the predicted effect includes:
the injection well is evaluated through parameters, the evaluation basis is to increase the flow rate of any injection well by a fixed value, if the evaluation is carried out on a production well, the flow rate of the production well with the fixed value is increased, the influence of the flow rate on the whole system is observed, and the injection well or the production well can be further evaluated, wherein the injection well evaluation formula is as follows:
Figure GDA0003528301320000042
wherein the content of the first and second substances,
Figure GDA0003528301320000043
in the form of a diagonal identity matrix,
Figure GDA0003528301320000044
ziand the contribution quantity of oil production at the future moment i is taken as an injection well, and the method is dimensionless.
Further, the performing uncertainty analysis on the prediction result of the model and simultaneously evaluating the oil recovery contribution of the injection well according to the fitted model parameters includes:
based on considering the original time series process as a random process, the compact matrix thereof is expressed as follows:
Figure GDA0003528301320000045
wherein the content of the first and second substances,
Figure GDA0003528301320000046
is an average prediction error matrix, and n is the observable data volume;
consider the constant influence of the parameters:
Figure GDA0003528301320000047
Figure GDA0003528301320000048
wherein the content of the first and second substances,
Figure GDA0003528301320000049
in the form of a diagonal identity matrix,
Figure GDA00035283013200000410
Φithe matrix of impact parameters is accumulated for the ith future time,
Figure GDA0003528301320000051
∑y(h) the impact matrix is accumulated for the h-th time in the future,
Figure GDA0003528301320000052
σj(h) estimating a prediction error for the jth well at a future time instant h, where σj(h) Corresponding to the estimation prediction error of the producing well j in the h-th prediction;
for the VAR model, where the important hyper-parameter is the maximum hysteresis value p, which represents the assumption of at most Yt-pAnd Et-p+1Will influence YtThe selection of the value is completed by an information criterion, firstly, a likelihood estimation value L of the model at the lag value p is calculated, and the information criterion is given by considering the likelihood function value and the size of the model parameter, and the formula is as follows:
Figure GDA0003528301320000053
AIC＝-2ln(L)+2k，
BIC＝-2ln(L)+ln(n)k，
HQ＝-2ln(L)+ln(ln(n))k，
Figure GDA0003528301320000054
wherein, L is a likelihood estimation value, AIC is an AIC information criterion evaluation value, BIC is a BIC information criterion evaluation value, HQ is an HQ information criterion evaluation value, FPE is an FPE information criterion evaluation value, and k is the number of model parameters.
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
1. a water drive reservoir seepage field evaluation method based on a vector autoregressive model is characterized in that a machine learning model is built by fitting historical production data of a production well, the relation between the oil production amount of any production well and the historical data which may exist is captured, the complex data relation can be modeled, the precision is high, the calculation time is short, the injection well oil production contribution amount can be evaluated by simulating the injection well injection increasing effect through the model, and the problem of poor convergence in numerical simulation fitting under the complex geological condition is effectively solved. Compared with the existing seepage field evaluation method, the method related by the invention has less subjectivity, can be evaluated without depending on expert experience, has higher accuracy, can carry out uncertainty analysis on prediction, and ensures the safety and the accuracy of the prediction result.
2. The invention predicts the oil recovery of the oil extraction well from 2017 to 2018 in the period of 5 months before 2017 of the oil deposit in the X region of the M oil field, the prediction precision can reach 86.92%, and the invention carries out water injection adjustment on the numerical oil deposit model in the X region of the M oil field according to the evaluation result of the injection well, the numerical simulation recovery rate is improved by 0.3112% in 2 years, and the oil production is increased by 34770M3Practice shows that the accuracy of the evaluation result obtained by the method is higher.
3. The vector autoregressive model is good at mining the data relation among multiple wells, is good at extracting the interaction rule from multiple time sequences, has strong model adaptability, and has lower calculation cost and higher prediction precision compared with the existing seepage field evaluation method.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and that for those skilled in the art, other relevant drawings can be obtained according to the drawings without inventive effort, wherein:
FIG. 1 is a flow chart of a method for evaluating a seepage field of a water-flooding reservoir based on a vector autoregressive model;
FIG. 2 is a graph of a co-correlation matrix between injection wells and production wells according to a first embodiment of the present invention;
FIG. 3 shows two pairs of injection and production wells with the highest correlation according to the first embodiment of the present invention;
FIG. 4 is a graph of the predicted validation results of different production wells according to a first embodiment of the present disclosure;
FIG. 5 is a prediction error condition of a production well produced in a validation set according to a first embodiment of the present disclosure;
FIG. 6 is a result diagram of uncertainty analysis of validation set prediction results according to a first embodiment of the present invention;
FIG. 7 is a line graph showing the effect of different injection wells on the oil production of a totality of production wells for a single time step according to a first embodiment of the present invention;
FIG. 8 is a line graph showing the effect of different injection wells on the cumulative time step of oil production from a totality of production wells in accordance with a first embodiment of the present invention;
fig. 9 is a graph showing evaluation results of an injection well according to a first embodiment of the present invention;
fig. 10 is an application interface of the machine learning module for production well production prediction and seepage field evaluation according to the first embodiment of the present invention;
FIG. 11 is a block diagram of a hysteresis order analysis module according to a first embodiment of the present invention;
fig. 12 is a predicted result drawing module interface and an impulse response module interface according to the first embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the detailed description and specific examples, while indicating the preferred embodiment of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention. The components of embodiments of the present invention generally described herein and illustrated in the figures may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
It is noted that relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
A water drive reservoir seepage field evaluation method based on a vector autoregressive model solves the problems that an existing seepage field evaluation method is high in calculation cost and low in accuracy, and convergence is poor under complex geological conditions.
A water drive reservoir seepage field evaluation method based on a vector autoregressive model comprises the following steps:
step 1: collecting historical production data and historical water injection data of a target oil reservoir;
step 2: preprocessing historical production data and historical water injection data;
and step 3: fitting a model according to the preprocessed historical production data and the preprocessed historical water injection data;
and 4, step 4: and predicting the future production amount of the production well according to the fitted model, carrying out uncertainty analysis on the prediction result of the model, and evaluating the oil production contribution amount of the injection well according to the fitted model parameters.
According to the method, the historical production data of the extraction well is fitted, the machine learning model is established, the relation between the oil extraction amount of any extraction well and the historical data which may exist is captured, the complex data relation can be modeled, the precision is high, the calculation time is short, the injection well injection increasing effect can be simulated through the model to evaluate the oil extraction contribution amount of the injection well, and the problem of poor convergence in numerical simulation fitting under the complex geological conditions is effectively solved. Compared with the existing seepage field evaluation method, the method related by the invention has less subjectivity, can be evaluated without depending on expert experience, has higher accuracy, can carry out uncertainty analysis on prediction, and ensures the safety and the accuracy of the prediction result.
The features and properties of the present invention are described in further detail below with reference to examples.
Example one
In the preferred embodiment of the invention, an X-region fault block oil reservoir of an M oil field is taken as an example, and a water drive oil reservoir seepage field evaluation method based on a vector autoregressive model is provided, as shown in FIG. 1;
vector Autoregressive (VAR) is a random process model used to capture the mutual linear dependence between multiple time series data. Any variable in the model can be represented by an equation consisting of a hysteresis value of the variable, a deterministic variable and an error term of the variable, and a future value of the variable is deduced. The model does not need to use a numerical simulation mode, passes through a structural model of simultaneous equations and needs a mode that variables specifically act in the seepage process. The only a priori knowledge required for the VAR method is a list of variables that can be assumed to affect each other over the term, i.e. to give optional influencing factors for the target variable, so that the correlation between the variables, i.e. between the injection well and the production well (injection-production well for short), needs to be observed
As shown in fig. 2, the co-correlation matrix of injection wells and production wells is shown, the abscissa is an injection well, the ordinate is a production well, the lighter color in the graph represents higher correlation, and the darker color is opposite, the correlation between the oil production of the production well and the injection amount of the injection well can be observed from the graph, and as can be seen from the co-correlation matrix, the flow rates of most of the injection wells are not correlated, but the correlation of a small part of the injection wells and the production wells is stronger; as shown in fig. 3, two pairs of injection and production wells with the highest correlation are shown, and it can be known from the figure that some injection and production wells show the same increase and decrease, so that it can be assumed that the flow rate of the injection and production wells can be characterized by a linear relationship, and further the injection and production well network system can be modeled in a targeted manner, so that the invention takes the yields of different production wells and the injection amount of the injection well as the time series processes related to each other, establishes a Vector Autoregressive (VAR) model, captures the yield dependency relationship between wells, simulates the future yield, and evaluates the local seepage field of the injection well;
the method comprises the following steps:
step 1: collecting historical production data and historical water injection data of a target oil reservoir, specifically collecting historical production data of an extraction well and historical water injection data of an injection well, and arranging the historical production data and the historical water injection data into files in a format of 'x.xlsx' or 'x.csv', wherein each row in a table at least comprises year and month, well number, daily oil production data and daily water injection data;
step 2: the method comprises the following steps of preprocessing historical production data and historical water injection data, specifically, performing sliding window smoothing and normalization processing on daily oil production data and daily water injection data, so that the stability of the data can be improved, and the fitting difficulty can be reduced;
and step 3: fitting a model according to the preprocessed historical production data and the preprocessed historical water injection data, and verifying the model;
step 3.1: constructing the preprocessed daily oil yield data and daily water injection data into a time series format, namely each line of data represents different wells, and each line of data represents data at different moments;
step 3.2: fitting the historical production data according to the historical production data and the historical water injection data by a vector autoregressive method, and selecting a reasonable hysteresis order for the model by a hysteresis order selection method;
the construction vector is:
Figure GDA0003528301320000081
wherein, yi,tThe yield of the ith extraction well at the moment t is m3A month; e.g. of the typei,tThe injection quantity of the ith injection well at the t moment is m3A month; k is a radical ofPThe total number of the producing wells; k is a radical ofITotal number of injection wells; y istIs the vector of the produced quantity of the producing well at the moment t,
Figure GDA0003528301320000082
Etfor the injection well injection quantity vector at time t,
Figure GDA0003528301320000083
the flow prediction model of the produced well is established according to the mutual relation of the flow among wells as follows:
Figure GDA0003528301320000084
wherein A isiIs a production well parameter matrix of the ith order and is used for describing the flow relation among production wells,
Figure GDA0003528301320000085
Figure GDA0003528301320000086
Biis an ith-order injection well parameter matrix used for describing the interaction relationship among injection wells and extraction wells,
Figure GDA0003528301320000087
p is a hysteresis order, representing that the production well flow of at most P months in the past can have an effect on the current production well flow, PICharacterization of Up to past p for injector well hysteresis orderIThe monthly injector well flow rate may have an effect on the current production well flow rate, c is a bias factor,
Figure GDA0003528301320000088
utfor the residual value at the time t,
Figure GDA0003528301320000089
the model takes the daily oil production of the production well in the current month and the past p months as Yt+…+Yt-p+1And taking daily injection amount of 1 month in the future and p-1 month before the future as input
Figure GDA00035283013200000810
Estimate value for predicting oil production in next month
Figure GDA00035283013200000811
And are performed iteratively, i.e.:
Figure GDA00035283013200000812
wherein the symbols with triangles are estimated as parameter matrices
Figure GDA0003528301320000091
The symbol is expressed as an estimated value, the estimation parameter matrix is solved through a simultaneous linear equation set, n is assumed to represent observable data quantity, and the solving process is as follows:
Y＝DZ+U，
Figure GDA0003528301320000092
Figure GDA0003528301320000093
U＝[u1,u2,…,un]，
Figure GDA0003528301320000094
Figure GDA0003528301320000095
Figure GDA0003528301320000096
Figure GDA0003528301320000097
wherein Y is a total dependent variable matrix, Z is a total independent variable matrix, U is a total residual error matrix, and D is a total parameter matrix;
and solving by a least square method:
Figure GDA0003528301320000098
because the obtained result may generate overfitting when the number of wells is too large or the hysteresis order is too large, a regularization term needs to be added to the equation set;
in this embodiment, the hysteresis order of the target data set is selected to be 6 according to the hysteresis order selection method;
step 3.3: selecting data of several months at the end of historical data as a verification set, using previous data as a training set training model, and if the longest time of the data is 2018, setting the data before 2017 as the training set and the data after 2017 as the verification set;
in the embodiment, data before 2017 and 5 months in historical production data of an X area of an M oil field are selected as a training set, data after 2017 and 5 months are selected as a verification set, and a model prediction result is verified;
as shown in fig. 4, the predicted verification results of different produced wells of the embodiment are shown, fig. 4(a) is training set data of west 43-6-4 and west 44-5-2, fig. 4(b) is training set data of west 44-8-2 and west 46-5-1, fig. 4(c) is training set data of west 1-9-2 and west 42-8-1, fig. 4(d) is training set data of west 49-5-3 and west 50-5-1, dotted lines in the figure represent training set data, difference marks represent verification set data, solid lines represent predicted verification set data, and it can be seen that the original training set data has large fluctuation, which brings certain challenge to prediction, and further analyzes the prediction situation of produced wells with yield greater than 0 in the verification set;
step 3.4: the trained model is used for predicting the effect of the verification set, then the prediction effect is evaluated, and if the prediction effect of the verification set is better, the prediction effect on the yield of the production well at the future moment is better, and the prediction precision on the oil production contribution of the injection well is higher;
the injection well is evaluated according to the parameters, the evaluation is based on the principle that the flow rate of any injection well is increased by a fixed value (generally 1), if the evaluation is carried out on the production well, the flow rate of the production well with the fixed value is increased, the influence of the flow rate on the whole system is observed, and the injection well or the production well can be further evaluated, wherein the injection well evaluation formula is as follows:
Figure GDA0003528301320000101
wherein the content of the first and second substances,
Figure GDA0003528301320000102
in the form of a diagonal identity matrix,
Figure GDA0003528301320000103
zithe contribution amount of oil production at the future moment i is taken as an injection well, and the method is dimensionless;
as shown in fig. 5, the prediction error condition of the produced wells in the verification set is shown, yellow in the histogram is the actual average yield in the verification set, blue is the predicted average yield, and the lower linear graph is the predicted relative error value, so that the predicted result is lower than the actual result except for a small number of the produced wells due to the recent implementation of the production increasing measures, and the other produced wells are better fitted, wherein the average absolute error value of the produced wells in the verification set is 0.513392, and the average relative error is 0.130811, which indicates that the fitting effect is better;
and 4, step 4: predicting future production amount of the production well according to the fitted model, carrying out uncertainty analysis on a prediction result of the model, and evaluating oil production contribution amount of the injection well according to the fitted model parameters;
because the given data volume is small, the prediction result still has a certain error, and the uncertain analysis can give the magnitude of model error caused by the previous data and give the magnitude of the model's reliability on the prediction, so that safer prediction can be performed, i.e. a cautious decision is made when the prediction uncertainty is large, therefore, it is necessary to perform uncertainty analysis on the model prediction and evaluation results, and the derivation of the prediction uncertainty magnitude is mainly based on the consideration of the original time series process as a random process, and a compact matrix thereof is represented as follows:
Figure GDA0003528301320000111
wherein the content of the first and second substances,
Figure GDA0003528301320000112
is an average prediction error matrix, and n is the observable data volume;
consider the constant influence of the parameters:
Figure GDA0003528301320000113
Figure GDA0003528301320000114
wherein the content of the first and second substances,
Figure GDA0003528301320000115
in the form of a diagonal identity matrix,
Figure GDA0003528301320000116
Φithe matrix of impact parameters is accumulated for the ith future time,
Figure GDA0003528301320000117
∑y(h) the impact matrix is accumulated for the h-th time in the future,
Figure GDA0003528301320000118
σj(h) estimating a prediction error for the jth well at a future time instant h, where σj(h) Corresponding to the estimation prediction error of the producing well j in the h-th prediction;
for the VAR model, where the important hyper-parameter is the maximum hysteresis value p, which represents the assumption of at most Yt-pAnd Et-p+1Will influence YtThe selection of the value is done by means of an information criterion, first of all, the modulus is calculatedAnd (3) setting a likelihood estimation value L of the model at a lag value p, and giving an information criterion by considering the likelihood function value and the size of the model parameter, wherein the formula is as follows:
Figure GDA0003528301320000119
AIC＝-2ln(L)+2k，
BIC＝-2ln(L)+ln(n)k，
HQ＝-2ln(L)+ln(ln(n))k，
Figure GDA00035283013200001110
wherein, L is a likelihood estimation value, AIC is an AIC information criterion evaluation value, BIC is a BIC information criterion evaluation value, HQ is an HQ information criterion evaluation value, FPE is an FPE information criterion evaluation value, and k is the number of model parameters, and the lag order which makes the criterion values smaller is selected as the proper lag order p by comprehensively considering the four information criteria;
as shown in fig. 6, a result diagram of uncertainty analysis performed on the verification set prediction result is shown, where points are training data, crosses are verification set data, a solid line is the prediction result on the verification set data, and an area is an uncertainty range, so that it can be seen that the prediction uncertainty ranges of different production wells basically include actual conditions; meanwhile, as mentioned in the uncertainty formula, the prediction uncertainty is mainly determined by the prediction error and parameters of the training set, so that the prediction uncertainty is large when the prediction error of the training set is large, as shown in fig. 6(c) and (d).
The seepage field evaluation based on the machine learning method is carried out by assuming that the average daily fluence of a certain injection well in the current month is increased by 1m3Observing the influence on other production wells, further calculating the influence of different injection wells on the whole oil deposit yield by the method, as shown in figures 7 and 8, the influence of different injection wells on the oil production of the whole production wells is broken line graph, figure 7 is the influence of single time step, figure 8 is the influence of accumulated time step, in the graph, the ordinate is the influence, and the unit is m3If it is 1, generationThe daily oil recovery of the whole production well is increased by 1m3And/d, with the abscissa as time in months, it can be seen that the effect of a single time step is decreasing with time. The graph marked as 2 in fig. 8 shows the cumulative effect of different production wells, and it can be seen that when the step size is 10, the effect of most injection wells converges to about 0, so the value of the injection well is evaluated according to the cumulative effect when the step size is 10, and as shown in fig. 9, the evaluation result of the injection well represents the oil recovery contribution amount of the model prediction for injecting water of 1m3 into different injection wells, and the result can provide a basis for the seepage field adjustment scheme.
The method realizes the function of a machine learning module through a Python programming language, ensures the practicability and convenience of the algorithm, can provide theoretical support for comprehensive adjustment of a seepage field through the evaluation result of software, and further improves the water drive efficiency and the utilization degree of a complex high-water-cut oil reservoir;
the machine learning module is used for the application of production well yield prediction and seepage field evaluation, the interface of the machine learning module is shown in fig. 10, the application reads the data of production well weekly reports or monthly reports of an oil field, and meanwhile, whether the influence of an injection well is considered in the machine learning model can be selected, and if not, the target oil deposit is assumed to have no injection well. Injector well effects are considered herein. Further, a suitable hysteresis order needs to be selected for the model, for example, as shown in fig. 11, a module interface for analyzing the hysteresis order shows that four information criteria corresponding to target oil deposit data are displayed, it can be seen that when the hysteresis order is 6, three information criteria values are all minimum values, so that it is reasonable to select the hysteresis order to be 6, further, production and water injection data before 2017 and 5 months old of an X-region oil deposit in an M oil field are selected as a training set, data after 2017 and 5 months old are selected as a verification set, and a model prediction result is verified, and the module is shown in fig. 12.
The method provided by the invention innovatively uses a vector autoregression method to predict the oil production of the production well, and can carry out uncertainty analysis on the prediction result, so that the prediction accuracy and convergence of the method are improved. At the same time, the method can calculate the injection 1m3The contribution amount of oil production of different injection wells when water is injected is evaluated so as to evaluate the injectionThe development potential of the well control area provides a basis for adjusting the seepage field. The machine learning model is established by fitting the historical production data of the production well, the relation between the oil production amount of any production well and the historical data which possibly exists is captured, the complex data relation can be modeled, the precision is high, the calculation time is short, the injection well augmented injection effect can be simulated through the model to evaluate the oil production contribution amount of the injection well, and the problem of poor convergence in numerical simulation fitting under the complex geological condition is effectively solved. Compared with the existing seepage field evaluation method, the method related by the invention has less subjectivity, can be evaluated without depending on expert experience, has higher accuracy, can carry out uncertainty analysis on prediction, and ensures the safety and the accuracy of the prediction result. The invention predicts the oil recovery of the oil extraction well from 2017 to 2018 in the period of 5 months before 2017 of the oil deposit in the X region of the M oil field, the prediction precision can reach 86.92%, and the invention carries out water injection adjustment on the numerical oil deposit model in the X region of the M oil field according to the evaluation result of the injection well, the numerical simulation recovery rate is improved by 0.3112% in 2 years, and the oil production is increased by 34770M3Practice shows that the accuracy of the evaluation result obtained by the method is higher.
It should be noted that, since the drawings in the specification should not be colored or modified, it is difficult to display a portion where a part of the distinction is obvious in the present invention, and if necessary, a color picture can be provided.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and should not be taken as limiting the scope of the present invention, and any modifications, equivalents and improvements made by those skilled in the art within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (6)

1. A water drive reservoir seepage field evaluation method based on a vector autoregressive model is characterized by comprising the following steps:
collecting historical production data and/or historical water injection data of a target oil reservoir;
preprocessing historical production data and/or historical water injection data;
fitting a model according to the preprocessed historical production data and/or historical water injection data, and verifying the model;
predicting future production amount of the production well according to the fitted model, carrying out uncertainty analysis on a prediction result of the model, and evaluating oil production contribution amount of the injection well according to the fitted model parameters;
the uncertainty analysis is carried out on the prediction result of the model, and the oil recovery contribution of the injection well is evaluated according to the fitted model parameters, and the method comprises the following steps:
based on considering the original time series process as a random process, the compact matrix thereof is expressed as follows:
Figure FDA0003537517470000011
wherein the content of the first and second substances,
Figure FDA0003537517470000012
is an average prediction error matrix, and n is the observable data volume;
consider the constant influence of the parameters:
Figure FDA0003537517470000013
Figure FDA0003537517470000014
wherein the content of the first and second substances,
Figure FDA0003537517470000015
in the form of a diagonal identity matrix,
Figure FDA0003537517470000016
Φithe matrix of impact parameters is accumulated for the ith future time,
Figure FDA0003537517470000017
∑y(h) the impact matrix is accumulated for the h-th time in the future,
Figure FDA0003537517470000018
σj(h) estimating a prediction error for the jth well at a future time instant h, where σj(h) Corresponding to the estimation prediction error of the producing well j in the h-th prediction;
for the VAR model, where the important hyper-parameter is the maximum hysteresis value p, which represents the assumption of at most Yt-pAnd Et-p+1Will influence YtThe selection of the value is completed by an information criterion, firstly, a likelihood estimation value L of the model at the lag value p is calculated, and the information criterion is given by considering the likelihood function value and the size of the model parameter, and the formula is as follows:
Figure FDA0003537517470000019
AIC＝-2ln(L)+2k，
BIC＝-2ln(L)+ln(n)k，
HQ＝-2ln(L)+ln(ln(n))k，
Figure FDA0003537517470000021
wherein, L is a likelihood estimation value, AIC is an AIC information criterion evaluation value, BIC is a BIC information criterion evaluation value, HQ is an HQ information criterion evaluation value, FPE is an FPE information criterion evaluation value, and k is the number of model parameters.
2. The method for evaluating the seepage field of the water-flooding reservoir based on the vector autoregressive model according to claim 1, wherein the historical production data and/or the historical waterflood data of the target reservoir are collected, specifically, the historical production data of a production well and/or the historical waterflood data of an injection well are collected and are arranged into files in the format of ". xlsx" or ". csv", and each row in the table at least comprises year and month, well number, daily oil production data and/or daily waterflood data.
3. The method for evaluating the seepage field of the water-drive reservoir based on the vector autoregressive model according to claim 2, wherein the preprocessing of the historical production data and/or the historical waterflood data comprises: and performing sliding window smoothing and normalization processing on the daily oil production data and/or the daily water injection data.
4. The method for evaluating the seepage field of the water-drive reservoir based on the vector autoregressive model according to claim 1, wherein the step of fitting the model according to the preprocessed historical production data and/or historical waterflooding data and verifying the model comprises the following steps:
constructing the preprocessed daily oil yield data and/or daily water injection data into a time series format, namely each line of data represents different wells, and each line of data represents data at different moments;
fitting the historical production data according to the historical production data and/or the historical water injection data by a vector autoregressive method, and selecting a reasonable hysteresis order for the model by a hysteresis order selection method;
selecting data of several months at the end of the historical data as a verification set, and taking the previous data as a training set to train a model;
and predicting the effect of the verification set through the trained model, and evaluating the predicted effect, wherein if the predicted effect of the verification set is better, the better the predicted effect of the yield of the production well at the future moment is, and the prediction precision of the contribution to the oil production of the injection well is higher.
5. The method for evaluating the seepage field of the water-drive reservoir based on the vector autoregressive model according to claim 4, wherein the step of fitting the historical production data according to the historical production data and/or the historical waterflooding data by the vector autoregressive method and selecting a reasonable hysteresis order for the model by a hysteresis order selection method comprises the following steps:
the construction vector is:
Figure FDA0003537517470000031
wherein, yi,tThe yield of the ith extraction well at the moment t is m3A month; e.g. of the typei,tThe injection quantity of the ith injection well at the t moment is m3A month; k is a radical ofPThe total number of the producing wells; k is a radical ofITotal number of injection wells; y istIs the vector of the produced quantity of the producing well at the moment t,
Figure FDA0003537517470000032
Etfor the injection well injection quantity vector at time t,
Figure FDA0003537517470000033
the flow prediction model of the produced well is established according to the mutual relation of the flow among wells as follows:
Figure FDA0003537517470000034
wherein A isiIs a production well parameter matrix of the ith order and is used for describing the flow relation among production wells,
Figure FDA0003537517470000035
Figure FDA0003537517470000036
Biis an ith-order injection well parameter matrix used for describing the interaction relationship among injection wells and extraction wells,
Figure FDA0003537517470000037
p is a hysteresis order, and the characteristic of the production well flow of P months at most can generate the current production well flowInfluence, pICharacterization of Up to past p for injector well hysteresis orderIThe monthly injector well flow rate may have an effect on the current production well flow rate, c is a bias factor,
Figure FDA0003537517470000038
utfor the residual value at the time t,
Figure FDA0003537517470000039
the model takes the daily oil production of the production well in the current month and the past p months as Yt+…+Yt-p+1And taking daily injection amount of 1 month in the future and p-1 month before the future as input
Figure FDA00035375174700000310
Estimate value for predicting oil production in next month
Figure FDA00035375174700000311
And are performed iteratively, i.e.:
Figure FDA00035375174700000312
wherein the symbols with triangles are estimated as parameter matrices
Figure FDA00035375174700000313
The symbol is expressed as an estimated value, the estimation parameter matrix is solved through a simultaneous linear equation set, n is assumed to represent observable data quantity, and the solving process is as follows:
Y＝DZ+U，
Figure FDA00035375174700000314
Figure FDA00035375174700000315
U＝[u1,u2,…,un]，
Figure FDA00035375174700000316
Figure FDA0003537517470000041
Figure FDA0003537517470000042
Figure FDA0003537517470000043
wherein Y is a total dependent variable matrix, Z is a total independent variable matrix, U is a total residual error matrix, and D is a total parameter matrix;
and solving by a least square method:
Figure FDA0003537517470000044
because the obtained result may generate overfitting when the number of wells is too large or the hysteresis order is too large, a regularization term needs to be added to the equation set;
and selecting the lag order of the target data set according to a lag order selection method.
6. The method for evaluating the seepage field of the water-drive reservoir based on the vector autoregressive model according to claim 4, wherein the step of predicting the effect of the verification set through the trained model and then evaluating the predicted effect comprises the following steps:
the injection well is evaluated through parameters, the evaluation basis is to increase the flow rate of any injection well by a fixed value, if the evaluation is carried out on a production well, the flow rate of the production well with the fixed value is increased, the influence of the flow rate on the whole system is observed, and the injection well or the production well can be further evaluated, wherein the injection well evaluation formula is as follows:
Figure FDA0003537517470000045
wherein the content of the first and second substances,
Figure FDA0003537517470000046
in the form of a diagonal identity matrix,
Figure FDA0003537517470000047
ziand the contribution quantity of oil production at the future moment i is taken as an injection well, and the method is dimensionless.
CN201910636131.1A 2019-07-15 2019-07-15 Water-drive reservoir seepage field evaluation method based on vector autoregressive model Active CN110348137B (en)

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 CN110348137A (en) 2019-10-18
CN110348137B true 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)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
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
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
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
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
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
CN115935834B (en) * 2023-02-23 2023-06-09 中国石油大学(华东) Historical fitting method based on deep autoregressive network and continuous learning strategy

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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

Family Cites Families (3)

* Cited by examiner, † Cited by third party
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
GB0524134D0 (en) * 2005-11-26 2006-01-04 Univ Edinburgh Improvements in and relating to hydrocarbon recovery from a hydrocarbon reservoir
FR2919932B1 (en) * 2007-08-06 2009-12-04 Inst Francais Du Petrole METHOD FOR EVALUATING A PRODUCTION SCHEME FOR UNDERGROUND GROWTH, TAKING INTO ACCOUNT UNCERTAINTIES

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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)

* Cited by examiner, † Cited by third party
Title
Production performance forecasting method based onmultivariate time series and vector autoregressive machine learning model for waterflooding reservoirs;ZHANG Rui 等;《PETROLEUM EXPLORATION AND DEVELOPMENT》;20210228;第48卷(第1期);201-211 *
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;张瑞 等;《石油勘探与开发》;20200925;第48卷(第1期);175-184 *

Also Published As

Publication number Publication date
CN110348137A (en) 2019-10-18

Similar Documents

Publication Publication Date Title
CN110348137B (en) Water-drive reservoir seepage field evaluation method based on vector autoregressive model
CN109447532B (en) Oil reservoir inter-well connectivity determination method based on data driving
CN111144542B (en) Oil well productivity prediction method, device and equipment
CN103336906B (en) The sampling Gaussian process regression model that in the image data stream of environmental sensor, continuous abnormal detects
CN111612651B (en) Abnormal electric quantity data detection method based on long-term and short-term memory network
EA026086B1 (en) Statistical reservoir model based on detected flow events
Du et al. The connectivity evaluation among wells in reservoir utilizing machine learning methods
CN109426672B (en) Oil reservoir injection-production parameter optimization method based on uncertain geological model
CN111639882B (en) Deep learning-based electricity risk judging method
CN103745002A (en) Method and system for recognizing hidden paid posters on basis of fusion of behavior characteristic and content characteristic
CN113496302B (en) Method and system for carrying out intelligent identification and early warning on drilling risk
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
Sen et al. Inter-well connectivity detection in CO2 WAG projects using statistical recurrent unit models
Chu et al. A deep-learning approach for reservoir evaluation for shale gas wells with complex fracture networks.
CN116542429A (en) Oil reservoir production index machine learning prediction method integrating space-time characteristics
CN109779622B (en) Method and device for characterizing low-efficiency water injection zone of oil reservoir in ultrahigh water cut period
CN113077091A (en) Machine learning-based leakage rate probability distribution prediction method and system
CN106530109A (en) Oilfield development appraisal well decision method based on information value
CN110486009A (en) A kind of automatic reverse method of parameter and system on infinity stratum
Zhou et al. Application of statistical methods to predict production from liquid-rich shale reservoirs
CN113627585A (en) Single well index prediction method and system based on long-time and short-time memory neural network
CN111625525A (en) Environmental data repairing/filling method and system
CN116976146B (en) Fracturing well yield prediction method and system coupled with physical driving and data driving
CN115422840B (en) Daily-scale runoff estimation method based on physical model hybrid deep learning model

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