CN111946311B - Separated injection and production simulation method - Google Patents

Abstract

The invention discloses a split-injection and split-production simulation method, which comprises the following steps: firstly, starting from the same zero moment, acquiring the historical water injection amount w of the N-port water injection well_i,k,tAnd historical oil production po of M production wells_j,k,tAnd then calculating the total water injection amount W of all the layers of the water injection wells at the t-th time point according to the acquired data_tSecond, calculate a ratio matrix Ro that eliminates oil production in the producing well_j,k,tThirdly, trend elimination is carried out on each production well at each moment; and establishing an injection-production relation model, and finally, formulating a water injection scheme, giving the water injection amount of each water injection well, and predicting and optimizing the oil production amount by combining the injection-production relation model.

Description

Separated injection and production simulation method

Technical Field

The invention relates to petroleum collection, in particular to a separate injection and separate production method of petroleum.

Background

In the past, the reasonable injection allocation of a water injection well is an important content of oilfield water injection development and management work. In recent years, along with the rapid development of a separate injection and separate production downhole tool and a matched process technology, an oil field development field gradually changes from a combined injection and combined production mode to a more efficient separate injection and separate production mode.

The existing method for determining the stratified injection allocation quantity of the displacement flux equalization is a method for quantitatively representing displacement degree difference of each stratum by adopting the displacement flux on the basis of analyzing the long-term water injection development characteristics of a multi-layer commingled oil reservoir, and is used for establishing a method for determining the injection allocation quantity of each stratum in the longitudinal direction of a water injection well based on the displacement flux equalization idea by taking the equalization displacement as a target and comprehensively considering the influences of factors such as the thickness, the porosity, the water injection history, the regulation and control period of the reservoir. In addition, the determination method of the layered injection allocation of the current common water injection well further comprises a grey correlation analysis method, a multiple regression method, a thickness method, a formation coefficient method, a residual oil method and the like. However, these methods all have certain disadvantages, such as using a grey correlation analysis method, and in practical application of a multiple regression method, because of the variable and complex underground conditions, the correlation of injection and production data may be weak, and it is difficult to establish a reliable water distribution model from the injection and production data; the method of applying the thickness method and the formation coefficient method needs less geological parameters, but the model considers less conditions, so the effect is not obvious in practical application. Residual oil methods and displacement fluxes can achieve good results, but the methods require more geophysical parameters that are not readily available.

Disclosure of Invention

In order to improve the oil yield by optimally configuring the injection water yield and research the optimal configuration problem of the water injection quantity of different layers of a water injection well, the invention provides a method for simulating injection and separate production, which generates an injection and production relation simulator according to historical production data and a deep learning model learning injection and production rule and then generates an optimal injection and production scheme by utilizing the injection and production relation simulator generation model; the technical scheme is as follows:

a kind of injection and production simulation method, including the following steps:

step one, starting from the same zero moment, acquiring historical water injection amount w of N water injection wells_i,k,tAnd historical oil production po of M production wells_j,k,tWherein:

1,2,3 …, N; n is the total number of water injection wells;

j ═ 1,2,3 …, M; m is the total number of production wells;

k is 1,2,3 …, K; k is the number of injection and production layers, which corresponds to the number of oil production layers;

t is 0,1,2 …, T; t is the data acquisition end point moment;

step two, calculating the total water injection amount W of all the layers of the water injection wells at the tth time point according to the acquired data_tAs shown in equation (1):

step three, calculating a ratio matrix Ro of oil production in the production well_j,k,tAs shown in equation (2):

step four, eliminating the trend of each production well at each moment according to the ratio matrix of the oil production and the equation (3);

obtaining the value of the oil production amount at the k layer time t ═ a of the j production well converted to the time t ═ 0

Wherein, s-t 0 represents the initial t-0 time;

po_j,k,t＝athe oil production at the kth layer t ═ a of the jth production well is represented;

Ro_j,k,t＝0representing the ratio of water injection to oil at the moment when the kth layer t of the jth production well is 0;

Ro_j,k,t＝arepresenting the ratio of water injection to oil at the moment of the kth layer t of the jth production well to a;

step five, establishing an injection-production relation model

O＝G(Weights²(s(Weights¹X+b¹))+b²) Equation (4)

Wherein:

x represents the water injection amount of each layer of different water injection wells;

Weights^his a connection weight matrix between layers, h is 1, 2;

b^his an offset vector, h is 1, 2;

o is the oil production;

s and G are activation functions;

s＝S(y)；

G＝S(y)；

with w_i,k,tAs X input equation (4) to

Training as the oil production O output of equation (4) yields Weights^hAnd b^h；

Step six, formulating a water injection scheme, giving the total water injection amount of all water injection wells, obtaining a detailed water injection scheme through a distribution method, and taking the detailed water injection scheme as an input X of equation (4) to obtain the estimated oil production O_g；

Repeating the sixth step until the estimated oil production O is obtained_gAt the maximum, the water injection pattern at this time is the optimum dispensing water injection pattern.

As a preferred technical solution, in the sixth step, an optimal water injection solution is generated based on the gray wolf algorithm.

Drawings

FIG. 1 is a historical co-injection/co-production versus time curve;

FIG. 2 (including 2a/2b/2c/2d/2e/2f/2g/2h/2i/2j/2k/2m/2n/2P/2q/2r/2s/2t/2u/2v/2 w/2 x/2y/2z) plots of the fitted and historical true oil production (_ po) and water production (_ pw) over time for each well (P1/P2/P3/P4) for each layer (_1/_2/_3/_4) at predicted time points 21-175;

FIG. 3 (including 3a/3b/3c/3d/3e/3f/3g/3h/3i/3j/3k/3m/3n/3P/3q/3r/3s/3t/3u/3v/3 w/3 x/3y/3z) is a graph of the predicted time point 175-205 of the fitting and historical true oil production (. quadrature.) and water production (. quadrature.) over time for each well (P1/P2/P3/P4);

fig. 4 is a historical co-injection/estimated co-production/historical co-production-time curve.

Detailed Description

The present invention will be further described with reference to the following examples and the accompanying drawings.

A kind of injection and production simulation method, characterized by including the following steps:

step one, starting from the same zero moment, acquiring historical water injection amount w of N water injection wells_i,k,tAnd historical oil production po of M production wells_j,k,tAnd historical water production rate pw_j,k,tWherein:

1,2,3 …, N; n is the total number of water injection wells;

j ═ 1,2,3 …, M; m is the total number of production wells;

k is 1,2,3 …, K; k is the number of injection and production layers, which corresponds to the number of oil production layers;

t is 0,1,2 …, T; t is the data acquisition end point moment;

step two, calculating the total water injection amount W of all the layers of the water injection wells at the tth time point according to the acquired data_tAs shown in equation (1):

step three, calculating a ratio matrix Ro of oil production in the production well_j,k,tAs shown in equation (2):

calculating a ratio matrix Rw of water production in a producing well_j,k,tAs shown in equation (2 a):

step four, eliminating the trend of each production well at each moment according to the ratio matrix of the oil production and the equation (3);

obtaining the value of the oil production amount at the k layer time t ═ a of the j production well converted to the time t ═ 0

Wherein, s-t 0 represents the initial t-0 time;

po_j,k,t＝athe oil production at the kth layer t ═ a of the jth production well is represented;

Ro_j,k,t＝0representing the ratio of water injection to oil at the moment when the kth layer t of the jth production well is 0;

Ro_j,k,t＝arepresenting the ratio of water injection to oil at the moment of the kth layer t of the jth production well to a;

according to the ratio matrix of the water yield and the equation (3a), trend elimination is carried out on each production well at each moment;

obtaining the value of the water yield conversion of the kth layer of the jth production well at the moment t-a to the moment t-0

Wherein, s-t 0 represents the initial t-0 time;

pw_j,k,t＝arepresenting the water yield of the jth production well at the kth layer t ═ a moment;

Rw_j,k,t＝0indicating the time when the kth layer t of the jth production well is equal to 0The ratio of water injection to water;

Rw_j,k,t＝arepresenting the ratio of water injection converted into water at the moment that the kth layer t of the jth production well is equal to a;

step five, establishing an injection-production relation model

O＝G(Weights²(s(Weights¹X+b¹))+b²) Equation (4)

Wherein:

x represents the water injection amount of each well layer;

Weights^his a connection weight matrix between layers, h is 1, 2;

b^his an offset vector, h is 1, 2;

o is the oil production;

s and G are activation functions;

s＝S(y)；

G＝S(y)；

another expression of equation (4) is:

with w_i,k,tInputting equation (4) as X, in correspondence with

Training as the oil production O output of equation (4) yields Weights^hAnd b^h；

Wherein:

w is water yield;

is a connection weight matrix from layer to layer

Is a bias vector;

another expression of equation (4a) is:

with w_i,k,tInputting equation (4a) as X, in correspondence with

As the water yield W output of equation (4a), is trained to obtain

Step six, under the same total water injection amount, different water injection schemes are formulated, the water injection amount of each water injection well is given and is used as the input X of the equation (4), and the estimated oil production O is obtained_g；

Repeating the sixth step until the estimated oil production O is obtained_gTaking the water injection scheme at the moment as the optimal separate injection water injection scheme; inputting the optimal separate injection water injection scheme into equation (4a), thereby obtaining the estimated water yield W_g；

In the sixth step, a plurality of groups of water injection schemes can be randomly formulated according to the total water injection amount so as to seek the maximum estimated oil production O_gSo as to determine the optimal water injection scheme;

to more efficiently seek the optimal water-filling scheme, the optimal water-filling scheme may also be generated based on the wolf algorithm (GWO); GWO is a parameter optimization algorithm inspired by the social level and hunting behavior of the gray wolf group, in nature, the social level of the gray wolf group is divided into 4 levels, from top to bottom, alpha wolf, beta wolf, delta wolf and omega wolf, the alpha wolf, beta wolf and delta wolf belong to the three-headed wolf, the positions of all omega wolfs are updated depending on the positions of the three-headed wolf, and the water injection amount of each layer of the water injection scheme is recorded in the position of the wolf in this example; the specific iterative optimization process is as follows:

randomly initializing R wolfs, inputting the R wolfs as X in sequence to an equation (4) of the injection-production relation model to obtain oil production, and comparing the oil production to obtain positions Y of alpha, beta and delta of the wolfs_α、 Y_β、Y_δAnd the corresponding oil production quantity value, then the iteration iter is equal to 0, each wolf is traversed in sequence, whether the value of each dimensionality of the r wolf falls into the feasible region interval of water injection of each historical layer is judged, an upper boundary or a lower boundary is assigned outside the interval, and otherwise, the processing is not carried out; the position Y of the r-th wolf_rInputting the X into an equation (4) of the injection-production relation model to obtain the sum of the oil production values of all layers calculated by the current Rth wolf; judging whether the oil production of the r-th wolf is higher than the oil production of the wolf alpha, the wolf beta and the wolf delta, if so, updating the latest positions of the three wolfs, otherwise, not updating the positions of the three wolfs; updating the latest position of the r-th gray wolf through an equation set (5) and an equation (6);

A_z＝[2-iter*(2/Max_iteration)](2*rand(0,1)-1)

C_z＝2*rand(0,1),(z＝1,2,3)

D_α＝|C₁Y_α-Y_r|

D_β＝|C₂Y_β-Y_r|

D_δ＝|C₃Y_δ-Y_requation set (5);

wherein:

Y_αindicates the position of the wolf α;

Y_βindicates the position of the wolf β;

Y_δindicates the position of the wolf delta;

Y_rindicates the r-th root of wolf's republic of ChinaA forward position;

Y_r(d +1) represents the position of the next time the r-th wolf;

A_zrepresenting a vector of co-ordinates;

C_zis [0, 2 ]]A random number of intervals;

D_αrepresents the distance between the current r-th wolf and the wolf alpha;

D_βrepresents the distance between the current r-th wolf and the wolf beta;

D_δrepresents the distance between the current r-th wolf and the wolf delta;

when | A in equation 6_zWhen | A is greater than or equal to 1, the gray wolves are dispersed in each area as much as possible and search for the hunting targets_zIf | < 1, the gray wolf will search for a prey in a certain area or a certain area in the centralized child; whether iter reaches the maximum iteration number Max _ iteration is judged, if yes, Y at the moment is returned_αIf not, traversing each wolf to update the position through equation set (5) and equation (6), if not, repeating the steps to obtain the final Y_αAs an optimal water injection scheme.

Test example:

from the same zero moment, acquiring water injection data of 5 water injection wells with 3 layers of positive rhythm types, and acquiring oil production data and water production data of 4 production wells with 3 layers of positive rhythm types, wherein the acquisition time range is 0-205; that is, N is 5, M is 4, K is 3, and T is 205, and the water injection amount/oil production amount/water production amount at each time point of each well layer is obtained.

Adding the water injection amount/oil production amount/water production amount of each layer of each well at the same moment respectively to obtain historical combined water injection amount/combined oil production amount/combined water production amount, and drawing an obtained historical combined water injection amount/combined oil production amount/combined water production amount-time curve as shown in a figure 1;

then, learning and training are carried out by using the separate injection and production data (water injection amount/oil production amount/water production amount) from the time point 21 to the time point 175, and an injection and production relation model is established (parameters: input layer 15, hidden layer 1 15, hidden layer 2 10, and output layer 24); keeping the total water injection quantity of each well layer constant at each moment, and combining GWO algorithm (parameter: grey wolf number is 1)0, maximum iteration is 200) to give an optimal water injection scheme at each time point, then the water injection amount at each time point from 21 to 205 is sequentially input into the model, the oil production amount of each well layer at each time point is respectively obtained, and the estimated oil production amount O is obtained_gAnd estimate water production W_gDrawing a yield-time curve of each layer of each well; will estimate the oil production O_gAnd estimate water production W_gAs an estimate E_estimateAt the same time (time points 21 to 205) with the historical oil production po_j,k,tAnd historical water production rate pw_j,k,tAs true value E_ture(ii) a The average absolute error MAE is calculated using equation (7), and the result is shown in fig. 2 and 3, where f is 155 in fig. 2 and 31 in fig. 3.

Wherein FIG. 2 is a plot of production versus time series for each well layer at time points 21 through 175, and FIG. 3 is a plot of production versus time series for each well layer at time points 175 through 205;

in fig. 2 and 3, the upper left symbol of each figure means:

p1, P2, P3 and P4 respectively represent the 1 st, 2 nd, 3 th and 4 th production wells;

a 1, a 2 and a 3 respectively represent the 1 st, the 2 nd and the 3 rd oil production layers;

po and pw respectively represent oil production and water production;

in order to evaluate the effect and estimation capability of the above optimal dispensing water distribution method, the model can be evaluated by using an average absolute error MAE, and the calculation formula of the MAE is shown in equation 7:

wherein:

E_turerepresenting the true value;

E_estimaterepresenting an estimated value;

f represents the number of observation points;

from fig. 2 and 3, it can be concluded that the above method has relatively good performance in both fitting and estimation, and the MAE error of oil production is not large, and can be applied to engineering practice.

And summing the estimated oil production of each production well at the same moment to obtain estimated oil production, and drawing an estimated oil production-time curve, as shown in fig. 4. The predicted oil production/historical oil production versus time curves at times 21-205 are shown in fig. 4.

As can be seen from FIG. 4, the estimated oil yield obtained by the above scheme is higher than the historical oil yield.

To quantify how much the predicted oil production is improved relative to the historical true oil production, a lift index is defined herein, as shown in equation (8):

wherein:

O₁a summary of the estimated oil production at time points 21 to 205 for each well layer;

O₂a summary of the historical true oil production at time points 21 to 205 for each well layer.

The results obtained by calculation are shown in table 1:

TABLE 1 summary of predicted oil production and historical true oil production at time points 21-205 for each well layer

As can be seen from table 1, the summary of the estimated oil production at time points 21 to 205 for each well layer is higher than the summary of the historical real oil production, which is an average improvement of 12.21%, and the improvement distribution for each layer is more uniform. The method can achieve the effect of increasing the oil yield by optimizing the water injection scheme under the condition that the total water injection amount is not changed.

A water injection scheme is given through an GWO algorithm and applied to engineering practice, the obtained real oil production is close to the estimated oil production under the same water injection scheme, and the estimated result is close to the real result.

The invention has the advantages that the optimization of the water injection scheme is beneficial to improving the oil production of separate injection and separate extraction, the estimated oil production is close to the real oil production, the optimal water injection scheme is given, the oil production is improved, the practicability is strong, and the simulation degree is high.

Finally, it should be noted that the above-mentioned description is only a preferred embodiment of the present invention, and those skilled in the art can make various similar representations without departing from the spirit and scope of the present invention.

Claims (2)

1. A kind of injection and production simulation method, characterized by including the following steps:

step one, starting from the same zero moment, acquiring historical water injection amount w of N water injection wells_i,k,tAnd historical oil production po of M production wells_j,k,tWherein:

1,2,3 …, N; n is the total number of water injection wells;

j ═ 1,2,3 …, M; m is the total number of production wells;

k is 1,2,3 …, K; k is the number of injection and production layers, which corresponds to the number of oil production layers;

t is 0,1,2 …, T; t is the data acquisition end point moment;

step two, calculating the total water injection amount W of all the layers of the water injection wells at the tth time point according to the acquired data_tAs shown in equation (1):

step three, calculating a ratio matrix Ro of oil production in the production well_j,k,tAs shown in equation (2):

step four, eliminating the trend of each production well at each moment according to the ratio matrix of the oil production and the equation (3);

obtaining the value of the oil production amount at the k layer time t ═ a of the j production well converted to the time t ═ 0

Wherein, s-t 0 represents the initial t-0 time;

po_j,k,t＝athe oil production at the kth layer t ═ a of the jth production well is represented;

Ro_j,k,t＝0representing the ratio of water injection to oil at the moment when the kth layer t of the jth production well is 0;

Ro_j,k,t＝arepresenting the ratio of water injection to oil at the moment of the kth layer t of the jth production well to a;

step five, establishing an injection-production relation model

O＝G(Weights²(s(Weights¹X+b¹))+b²) Equation (4)

Wherein:

x represents the water injection amount of each layer of different water injection wells;

Weights^his a connection weight matrix between layers, h is 1, 2;

b^his an offset vector, h is 1, 2;

o is the oil production;

s and G are activation functions;

s＝S(y)；

G＝S(y)；

with w_i,k,tInputting equation (4) as X, in correspondence with

Training as the oil production O output of equation (4) yields Weights^hAnd b^h；

Step six, formulating a water injection scheme, giving the water injection amount of each water injection well, and taking the water injection amount as the input X of the equation (4) to obtain the estimated oil production O_g；

Repeating the steps until the estimated oil production O is obtained_gAt the maximum, the water injection pattern at this time is the optimum dispensing water injection pattern.

2. The injection and production division simulation method according to claim 1, wherein:

in the sixth step, an optimal water injection scheme is generated based on the gray wolf algorithm, specifically:

randomly initializing R wolfs, inputting the R wolfs as X in sequence to an equation (4) of the injection-production relation model to obtain oil production, and comparing the oil production to obtain positions Y of alpha, beta and delta of the wolfs_α、Y_β、Y_δAnd the corresponding oil production quantity value enters an iteration iter which is equal to 0, each gray wolf is traversed in sequence, whether the value of each dimensionality of the r gray wolf falls into a feasible region interval of water injection of each historical layer is judged, an upper boundary or a lower boundary is assigned outside the interval, and otherwise, the process is not carried out; the position Y of the r-th wolf_rInputting the X into an equation (4) of the injection-production relation model to obtain the sum of the oil production values of all layers calculated by the current Rth wolf; judging whether the oil production of the r-th wolf is higher than the oil production of the wolf alpha, the wolf beta and the wolf delta, if so, updating the latest positions of the three wolfs, otherwise, not updating the positions of the three wolfs; updating the latest position of the r-th gray wolf through an equation set (5) and an equation (6);

A_z＝[2-iter*(2/Max_iteration)](2*rand(0,1)-1)

C_z＝2*rand(0,1),(z＝1,2,3)

D_α＝|C₁Y_α-Y_r|

D_β＝|C₂Y_β-Y_r|

D_δ＝|C₃Y_δ-Y_requation set (5);

wherein:

Y_αindicates the position of the wolf α;

Y_βindicates the position of the wolf β;

Y_δindicates the position of the wolf delta;

Y_rrepresents the current position of the r-th wolf;

Y_r(d +1) represents the position of the next time the r-th wolf;

A_zrepresenting a vector of co-ordinates;

C_zis [0, 2 ]]A random number of intervals;

D_αrepresents the distance between the current r-th wolf and the wolf alpha;

D_βrepresents the distance between the current r-th wolf and the wolf beta;

D_δrepresents the distance between the current r-th wolf and the wolf delta;

when | A in equation 6_zWhen | A is greater than or equal to 1, the gray wolves are dispersed in each area as much as possible and search for the hunting targets_zIf | < 1, the gray wolf will search for a prey in a certain area or a certain area in the centralized child; whether iter reaches the maximum iteration number Max _ iteration is judged, if yes, Y at the moment is returned_αIf not, traversing each wolf to update the position through equation set (5) and equation (6), if not, repeating the steps to obtain the final Y_αAs an optimal water injection scheme.

Images