Carbon dioxide injection amount adjusting method based on deep learning phase model

Abstract

The invention provides a carbon dioxide injection amount adjusting method based on a deep learning phase model. The current carbon dioxide injection rate of the target well and the formation temperature, pressure and pore radius corresponding to the target well are obtained; determining phase state parameters corresponding to the target well according to a pre-trained parameter prediction model; according to the phase state parameters corresponding to the target well, whether the current carbon dioxide injection amount meets the preset mining requirement or not is judged; and when it is determined that the current carbon dioxide injection amount does not meet the preset mining requirement, the current carbon dioxide injection amount is adjusted. The parameter prediction model is constructed through a physical information neural network algorithm and a preset state equation, and then the parameter prediction model is trained in combination with a preset gas-liquid equilibrium ratio model constructed by a Wilson equation and the preset state equation, so that the parameter prediction model with relatively high precision and relatively good effect can be obtained; and furthermore, the phase state parameters with relatively small errors and relatively high reference values can be accurately determined by utilizing the parameter prediction model.

Description

Technical Field

[0001] This specification belongs to the field of oil extraction, and in particular to a method for adjusting the amount of carbon dioxide injection based on a deep learning phase model. Background Art

[0002] Carbon dioxide is an effective medium for improving crude oil recovery. By determining the physical properties of crude oil-carbon dioxide two-phase fluid through phase analysis, the gas injection design can be optimized to ensure that carbon dioxide can be effectively dissolved in crude oil during the injection process, thereby improving oil recovery efficiency.

[0003] At present, the main methods for phase state calculation include state equation method, machine learning method and empirical formula method. However, the state equation method takes a long time to calculate; although the machine learning method has a faster calculation speed, it is not universal; the empirical formula method has large errors and poor application effect; there is an urgent need for a method that can quickly and accurately predict the phase state parameters of the crude oil-carbon dioxide system and then guide the injection amount of carbon dioxide gas drive.

[0004] Currently, no effective solution has been proposed to the above problems. Summary of the invention

[0005] This specification provides a method for adjusting the carbon dioxide injection amount based on a deep learning phase model, which uses a parameter prediction model to determine the phase parameters of a target well, thereby solving the problems of low efficiency in phase parameter prediction and a large number of iterations. At the same time, the carbon dioxide injection amount is adjusted according to the phase parameters to improve oil recovery efficiency.

[0006] This specification provides a method for adjusting the amount of carbon dioxide injection based on a deep learning phase model, including:

[0007] Obtaining the current carbon dioxide injection volume of the target well, as well as the formation temperature, pressure and pore radius corresponding to the target well;

[0008] According to a pre-trained parameter prediction model, based on the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well, the phase parameters corresponding to the target well are determined; wherein the parameter prediction model is a model constructed based on a physical information neural network algorithm and a preset state equation, the parameter prediction model is trained based on historical sample data and a first gas-liquid balance ratio corresponding to the historical sample data, the first gas-liquid balance ratio is obtained by processing the historical sample data according to a preset gas-liquid balance ratio model, and the preset gas-liquid balance ratio model is constructed by the Wilson equation and the preset state equation;

[0009] According to the phase parameters corresponding to the target well, determining whether the current carbon dioxide injection amount meets the preset production demand;

[0010] When it is determined that the current carbon dioxide injection amount does not meet the preset production demand, the current carbon dioxide injection amount is adjusted.

[0011] In one embodiment, before determining the phase parameters corresponding to the target well according to the pre-trained parameter prediction model based on the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well, the method further includes:

[0012] Acquiring the historical sample data;

[0013] According to the preset gas-liquid equilibrium ratio model, the historical sample data is processed to obtain the first gas-liquid equilibrium ratio;

[0014] Inputting the historical sample data into the parameter prediction model to obtain a second gas-liquid equilibrium ratio;

[0015] According to the preset loss function, based on the first gas-liquid equilibrium ratio and the second gas-liquid equilibrium ratio, determine whether the parameter prediction model has converged. If it is determined that the parameter prediction model has not converged, continue to train the parameter prediction model according to the historical sample data until the parameter prediction model converges to obtain a trained parameter prediction model.

[0016] In one embodiment, the preset loss function includes a first loss function based on data driving and a second loss function based on physical constraints.

[0017] In one embodiment, the historical sample data includes multiple groups of sample data, each group of sample data includes a historical carbon dioxide injection amount, and a historical formation temperature, a historical pressure, and a historical pore radius corresponding to the historical carbon dioxide injection amount.

[0018] The step of inputting the historical sample data into the parameter prediction model to obtain a second gas-liquid equilibrium ratio includes:

[0019] Determine the correlation between each two groups of the historical sample data;

[0020] According to the correlation degree, anomaly detection processing is performed on the historical sample data, and when abnormal data is detected in the historical sample data, correction processing is performed on the abnormal data according to a preset interpolation algorithm to obtain corrected historical sample data;

[0021] The corrected historical sample data is input into the parameter prediction model to obtain the second gas-liquid equilibrium ratio.

[0022] In one embodiment, inputting the corrected historical sample data into the parameter prediction model to obtain the second gas-liquid equilibrium ratio includes:

[0023] Determining a weight value corresponding to the sample data according to a relationship between a historical carbon dioxide injection amount in the sample data and a preset concentration threshold;

[0024] The corrected historical sample data and the weight value corresponding to each group of the sample data are input into the parameter prediction model to obtain the second gas-liquid equilibrium ratio.

[0025] In one embodiment, determining the degree of association between each two groups of sample data includes:

[0026] The correlation between each two groups of the sample data is determined based on the correlation between the historical carbon dioxide injection amounts, the correlation between the historical formation temperatures, and the correlation between the historical pore radii of each two groups of the historical sample data.

[0027] In one embodiment, performing anomaly detection processing on the historical sample data according to the association degree includes:

[0028] According to the association degree, the historical sample data is classified to obtain multiple classes;

[0029] Acquire second sample data in the first category corresponding to the first sample data; wherein the first category is any one of the multiple categories, and the first sample data is any one of the sample data in the first category;

[0030] Acquire a first distance between the first sample data and each of the second sample data, and determine an average distance based on the first distance;

[0031] determining a deviation compression value according to the first distance and the average distance;

[0032] Performing logarithmic transformation on the deviation compression value to obtain a multidimensional deviation compression index;

[0033] Performing standardization processing on the multidimensional deviation compression index to obtain a standardized deviation compression index;

[0034] According to the standardized deviation compression index and the preset index threshold, anomaly detection processing is performed on the historical sample data.

[0035] This specification provides a carbon dioxide injection amount adjustment device based on a deep learning phase model, comprising:

[0036] A data acquisition module, used to acquire the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well;

[0037] A parameter determination module, for determining the phase parameters corresponding to the target well according to a pre-trained parameter prediction model, based on the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well; wherein the parameter prediction model is a model constructed based on a physical information neural network algorithm and a preset state equation, the parameter prediction model is trained based on historical sample data and a first gas-liquid equilibrium ratio corresponding to the historical sample data, the first gas-liquid equilibrium ratio is obtained by processing the historical sample data according to a preset gas-liquid equilibrium ratio model, and the preset gas-liquid equilibrium ratio model is constructed by the Wilson equation and the preset state equation;

[0038] A demand determination module, used to determine whether the current carbon dioxide injection amount meets the preset production demand according to the phase parameters corresponding to the target well;

[0039] The injection amount adjustment module is used to adjust the current carbon dioxide injection amount when it is determined that the current carbon dioxide injection amount does not meet the preset production demand.

[0040] This specification also provides an electronic device, including a processor and a memory for storing instructions executable by the processor, wherein when the processor executes the instructions, a method for adjusting the amount of carbon dioxide injection based on a deep learning phase model is implemented.

[0041] This specification also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed, implement a method for adjusting the amount of carbon dioxide injection based on a deep learning phase model.

[0042] A method for adjusting the carbon dioxide injection amount based on a deep learning phase model provided in this specification is provided, by obtaining the current carbon dioxide injection amount of a target well, and the formation temperature, pressure and pore radius corresponding to the target well; according to a pre-trained parameter prediction model, based on the current carbon dioxide injection amount of the target well, and the formation temperature, pressure and pore radius corresponding to the target well, the phase parameters corresponding to the target well are determined; wherein the parameter prediction model is a model constructed based on a physical information neural network algorithm and a preset state equation, the parameter prediction model is trained based on historical sample data and a first gas-liquid equilibrium ratio corresponding to the historical sample data, the first gas-liquid equilibrium ratio is obtained by processing the historical sample data according to a preset gas-liquid equilibrium ratio model, and the preset gas-liquid equilibrium ratio model is constructed by the Wilson equation and the preset state equation; according to the phase parameters corresponding to the target well, it is judged whether the current carbon dioxide injection amount meets the preset mining demand; when it is determined that the current carbon dioxide injection amount does not meet the preset mining demand, the current carbon dioxide injection amount is adjusted. In this way, firstly, by constructing a parameter prediction model based on a physical information neural network algorithm and a preset state equation, and then combining the preset gas-liquid equilibrium ratio model constructed by the Wilson equation and the preset state equation, the parameter prediction model is trained to obtain a parameter prediction model with high accuracy and good effect, and then the parameter prediction model can be used to accurately determine the phase parameters with small errors and high reference value. Secondly, according to the phase parameters, on the one hand, it is possible to accurately judge whether the current carbon dioxide injection amount meets the preset mining demand; on the other hand, when the current carbon dioxide injection amount does not meet the preset mining demand, the phase parameters can be used as the basis for adjustment to accurately adjust the current carbon dioxide injection amount, thereby effectively improving the oil recovery efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of this specification, the drawings required for use in the embodiments will be briefly introduced below. The drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 is a flow chart of a method for adjusting the amount of carbon dioxide injection based on a deep learning phase model provided by an embodiment of this specification;

[0045] Figure 2 It is a schematic diagram of the overall process provided by an embodiment of this specification;

[0046] Figure 3 It is a schematic diagram of a physical information neural network algorithm flow chart provided by an embodiment of this specification;

[0047] Figure 4 is a schematic diagram of an abnormal data processing strategy flow provided by an embodiment of this specification;

[0048] Figure 5 It is a schematic diagram of the structure of an electronic device provided by an embodiment of this specification;

[0049] Figure 6 It is a schematic diagram of the structural composition of a carbon dioxide injection amount adjustment device based on a deep learning phase model provided by an embodiment of this specification;

[0050] Figure 7 It is a schematic diagram of prediction accuracy of a physical information neural network algorithm provided by an embodiment of this specification;

[0051] Figure 8 It is a schematic diagram comparing the number of iteration steps of a parameter prediction model and a gas-liquid equilibrium model provided by an embodiment of this specification;

[0052] Fig. 9 It is a schematic diagram comparing the operation time of a parameter prediction model and a gas-liquid equilibrium model provided by an embodiment of this specification. DETAILED DESCRIPTION

[0053] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0054] At present, the main methods for calculating phase parameters are state equation method, machine learning method and empirical formula method. The state equation method uses the PR state equation (i.e., Peng-Robinson state equation) to model and uses the equation iteration method for cyclic calculation. However, since the state equation method takes a long time to calculate shale nanopores and crude oil-carbon dioxide systems, and with the increase of carbon dioxide injection or the decrease of pore size, the calculation time will increase significantly, which makes it impossible to quickly analyze the phase data and adjust the carbon dioxide injection amount in real time. The machine learning method has a fast calculation speed, but it is often not universal, and the machine learning method is mainly for crude oil systems and cannot be used for systems with high carbon dioxide content. In addition, the empirical formula method has a fast calculation speed, but the error is large and the application effect is poor. In summary, there is currently a lack of methods that can quickly and accurately predict the phase parameters of crude oil-carbon dioxide systems.

[0055] In view of the root cause of the above problems, this specification considers that first, by constructing a parameter prediction model based on a physical information neural network algorithm and a preset state equation, and then combining the preset gas-liquid equilibrium ratio model constructed by the Wilson equation and the preset state equation, the parameter prediction model is trained to obtain a parameter prediction model with high accuracy and good effect, and then the parameter prediction model can be used to accurately determine the phase parameters with small errors and high reference value. Secondly, according to the phase parameters, on the one hand, it can accurately determine whether the current carbon dioxide injection amount meets the preset mining demand; on the other hand, when the current carbon dioxide injection amount does not meet the preset mining demand, the phase parameters can be used as the basis for adjustment to accurately adjust the current carbon dioxide injection amount, thereby effectively improving the oil recovery efficiency.

[0056] See also Figure 1 As shown, the embodiment of this specification provides a method for adjusting the amount of carbon dioxide injection based on a deep learning phase model, wherein the method is specifically applied to the server side. When implemented specifically, the method may include the following contents:

[0057] S101: Acquire the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well.

[0058] S102: Determine the phase parameters corresponding to the target well according to a pre-trained parameter prediction model, based on the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well; wherein the parameter prediction model is a model constructed based on a physical information neural network algorithm and a preset state equation, the parameter prediction model is trained based on historical sample data and a first gas-liquid equilibrium ratio corresponding to the historical sample data, the first gas-liquid equilibrium ratio is obtained by processing the historical sample data according to a preset gas-liquid equilibrium ratio model, and the preset gas-liquid equilibrium ratio model is constructed by the Wilson equation and the preset state equation.

[0059] S103: judging whether the current carbon dioxide injection amount meets the preset production demand according to the phase parameters corresponding to the target well.

[0060] S104: When it is determined that the current carbon dioxide injection amount does not meet the preset mining demand, the current carbon dioxide injection amount is adjusted.

[0061] Among them, the above-mentioned phase parameters can determine the phase behavior between crude oil and carbon dioxide under different conditions. The phase behavior can be used to characterize the gas-liquid phase distribution, miscibility conditions and critical conditions, etc. At the same time, the phase parameters can also be used to characterize and determine the solubility and miscibility pressure of carbon dioxide in underground reservoirs and to adjust the injection amount of carbon dioxide.

[0062] Furthermore, the injection pressure is adjusted according to the miscibility pressure between carbon dioxide and crude oil, thereby adjusting the injection amount of carbon dioxide. Specifically, in order to achieve maximum oil and gas miscibility, the injection pressure of carbon dioxide can be adjusted to be higher than the miscibility pressure.

[0063] Furthermore, by numerically simulating phase parameters, the development effects under different CO2 injection rates, injection pressures and injection volumes can be simulated.

[0064] In addition, the phase behavior between crude oil and carbon dioxide can be determined through the phase diagram, thereby avoiding the decline in oil and gas production caused by miscible phase separation.

[0065] Among them, the above parameter prediction model is a model constructed based on the physical information neural network algorithm and the preset state equation (ie, the PINN-EOS model).

[0066] In some embodiments, the phase parameters corresponding to the target well may be determined according to the pre-trained parameter prediction model based on the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well in the following manner, including:

[0067] Determine the physical property parameters of the target well according to the current carbon dioxide injection amount of the target well, and the formation temperature, pressure and pore radius corresponding to the target well; wherein the physical property parameters include: critical temperature, critical pressure, centrifugal factor, binary interaction parameter and molar mass;

[0068] According to the pre-trained parameter prediction model, based on the physical property parameters, the phase parameters corresponding to the target well are determined.

[0069] Based on the above embodiments, firstly, by constructing a parameter prediction model based on a physical information neural network algorithm and a preset state equation, and then combining the preset gas-liquid equilibrium ratio model constructed by the Wilson equation and the preset state equation, the parameter prediction model is trained to obtain a parameter prediction model with high accuracy and good effect, and then the parameter prediction model can be used to accurately determine the phase parameters with small errors and high reference value. Secondly, according to the phase parameters, on the one hand, it is possible to accurately judge whether the current carbon dioxide injection amount meets the preset mining demand; on the other hand, when the current carbon dioxide injection amount does not meet the preset mining demand, the phase parameters can be used as the basis for adjustment to accurately adjust the current carbon dioxide injection amount, thereby effectively improving the oil recovery efficiency.

[0070] In some embodiments, see Figure 2As shown, before determining the phase parameters corresponding to the target well according to the pre-trained parameter prediction model based on the current carbon dioxide injection volume of the target well and the formation temperature, pressure and pore radius corresponding to the target well, the method may further include the following contents when implemented:

[0071] S1: Acquire the historical sample data.

[0072] S2: According to the preset gas-liquid equilibrium ratio model, the historical sample data is processed to obtain the first gas-liquid equilibrium ratio.

[0073] S3: Inputting the historical sample data into the parameter prediction model to obtain a second gas-liquid equilibrium ratio.

[0074] S4: According to a preset loss function, based on the first gas-liquid equilibrium ratio and the second gas-liquid equilibrium ratio, determine whether the parameter prediction model has converged; if it is determined that the parameter prediction model has not converged, continue to train the parameter prediction model according to the historical sample data until the parameter prediction model converges, so as to obtain a trained parameter prediction model.

[0075] In some embodiments, according to the preset gas-liquid equilibrium ratio model, the historical sample data is processed to obtain the first gas-liquid equilibrium ratio. When specifically implemented, it may include:

[0076] Determine the historical physical property parameters of the target well based on the historical carbon dioxide injection volume of the target well in the historical sample data, and the historical formation temperature, historical pressure and historical pore radius corresponding to the target well;

[0077] According to the preset gas-liquid equilibrium ratio model, the historical physical property parameters are processed to obtain the first gas-liquid equilibrium ratio.

[0078] The preset gas-liquid equilibrium ratio model is constructed by the Wilson equation and the preset state equation.

[0079] Furthermore, the above historical sample data includes: preset N (for example, 8) components containing carbon dioxide, wherein the content of carbon dioxide in each component is controlled within the range of 1% to 75%, and the lighter the molecular weight of the remaining hydrocarbons, the higher the content in the component, and the sum of the contents of the 8 components can be 1. Specifically, the data of the first 7 components can be generated according to the set interval, and finally the sum of the contents of the first 7 components is subtracted from 1 to obtain the content of the 8th component. If the content of the 8th component is within the set interval, it can be determined that the group of data is valid, otherwise the values ​​of the first 7 components are adjusted until the content of the 8th component is within the set interval.

[0080] Furthermore, the value range of the formation temperature in the historical sample data may be 335-405 K, the value range of the pressure may be 15-45 MPa, and the value range of the pore size may be 1 nm-1 mm.

[0081] Furthermore, the critical temperature Tc (K) in the above physical property parameters can be: [304.2, 190.6, 324.2, 452.8, 607.8, 715.1, 763.4, 857.4].

[0082] Critical pressure Pc(Pa):[7376.46,4600.16,4709.7,3524.6,2325.7,1723.17,1594.92,1435.49].

[0083] Centrifugal factor ω: [0.225, 0.008, 0.112556, 0.225, 0.479, 0.731, 0.913, 1.125].

[0084] Binary interaction coefficient BIP:

[0085] [0,0.0000827,0.00516,0.0212,0,0,0,0],

[0086] [0.0000827,0,0.00398,0.0187,0,0,0,0],

[0087] [0.00516,0.00398,0,0.00646,0,0,0,0],

[0088] [0.0212,0.0187,0.00646,0,0,0,0,0],

[0089] [0,0,0,0,0,0,0,0],

[0090] [0,0,0,0,0,0,0,0],

[0091] [0,0,0,0,0,0,0,0],

[0092] [0,0,0,0,0,0,0,0].

[0093] Molar mass M (g / mol): [44.01, 16.04, 33.85193, 66.5965, 134.21, 219.53, 287.59, 385.63].

[0094] Saturation pressure P0 (Pa): [38515625,34960937,29531250,41406250,34375000,30312500,33750000,41865234,37539062].

[0095] Specifically, the preset state equation may be an A-PR state equation, which may be used for shale nanopores and crude oil-carbon dioxide systems. The first gas-liquid equilibrium ratio may be obtained by performing a phase separation calculation under a state lower than the saturation pressure when the saturation pressure is obtained by iteratively calculating the preset state equation.

[0096] Specifically, the above Wilson equation is obtained according to the following formula:

[0097]

[0098] Among them, P ci is the critical pressure of the sample data component i, P is the pressure of the gas-liquid two-phase equilibrium closed system, ω i is the eccentricity factor of the sample data component i, T ci is the critical temperature of the sample data component i.

[0099] Since the gas-liquid equilibrium constant predicted by the Wilson equation deviates greatly from the true value, has a large number of iteration steps and a long running time, the Wilson equation can be replaced by a physical neural network algorithm with a smaller interpolation value between the predicted gas-liquid equilibrium ratio and the true value, fewer iteration steps and a shorter running time.

[0100] For details, see Figure 3 As shown, the above parameter prediction model can be composed of 5 fully connected layers, wherein the activation function of the model can be a rectified linear unit (ReLU), and the optimizer of the model can be an Adam optimizer (Adaptive Moment Estimation, Adam), which can clean and repair the historical sample data to obtain corrected historical sample data.

[0101] Furthermore, when the carbon dioxide composition, formation temperature, pressure and pore radius are known, the input data passes through the linear transformation and ReLU activation function of each layer in turn. The mathematical expression of the function is:

[0102] f(x)=max(0,x),

[0103] Based on the above method, since ReLU transforms each element of the input, it outputs the value when the input value is greater than zero and outputs the negative value as zero. Therefore, sparse activation can be generated, which helps to reduce overfitting. Using the Adam optimizer, the learning rate is 0.001. Adam can calculate the adaptive learning rate for each parameter separately, can handle sparse gradients and uneven gradients, and has good robustness.

[0104] The corrected historical sample data is forward propagated through the network, and according to each Epoch of model training, the corrected historical sample data is respectively forward propagated, loss calculated, back propagated, and weights updated to obtain a trained parameter prediction model. In this way, data such as the current carbon dioxide injection volume of the target well can be input into the physical information neural network in the trained parameter prediction model to extract features through the physical information neural network, and then the extracted features are standardized using the trained parameter prediction and normalizer, converted into tensors for prediction, and then the prediction results are de-standardized.

[0105] Among them, when the historical sample data is a preset number (such as 40500), the training speed and accuracy of the parameter prediction model reach a good balance. That is, in the stability test, when the input component data is in two phases, the parameter prediction model can be used for prediction.

[0106] In some embodiments, the preset loss function includes a first loss function based on data driving and a second loss function based on physical constraints.

[0107] Specifically, the data-driven first loss function is used to measure the difference between the first gas-liquid equilibrium ratio and the second gas-liquid equilibrium ratio, and can be calculated according to the following formula:

[0108]

[0109] Among them, u i is the first gas-liquid equilibrium ratio of the i-th group of data, u(x i ) is the second gas-liquid equilibrium ratio of the i-th group of data.

[0110] Specifically, the second loss function based on physical constraints can be used to measure the degree of conformity between the second gas-liquid equilibrium ratio output by the parameter prediction model and the physical equation, wherein the second loss function can be calculated according to the following formula:

[0111]

[0112] in, is an operator that contains constraints for the two-phase fluid physics equations, x jIt is the jth group of data in the historical sample data.

[0113] Furthermore, the loss value may be determined according to the preset loss function based on the first gas-liquid equilibrium ratio and the second gas-liquid equilibrium ratio by the following formula:

[0114] L=ω d L data +ω p L physics ,

[0115] Among them, L is the loss value, ω d It is based on the data-driven first loss function L data The weight coefficient of p is the second loss function L of the physical constraint physics The weight coefficient of .

[0116] Furthermore, a verification sample data set can be obtained, and the trained parameter prediction model can be verified based on the verification sample data set to obtain the prediction accuracy of the trained parameter prediction model, and when the prediction accuracy is lower than a preset value (such as 93%), the parameters of the parameter prediction model are adjusted, and the training and prediction steps are repeated until the prediction accuracy is no lower than the preset value.

[0117] Based on the above method, for the two-phase fluid components of crude oil and carbon dioxide, the gas-liquid equilibrium ratio can be predicted using the trained parameter prediction model, and the loss value determined by the first loss function based on data drive and the second loss function based on physical constraints can be combined to make the predicted value of the gas-liquid equilibrium ratio closer to the actual value in the physical sense. At the same time, the accuracy of the gas-liquid equilibrium ratio prediction can be improved according to the preset values ​​under different confidence levels.

[0118] In addition, the use of a preset loss function can minimize both the data fitting error and the physical error, thereby ensuring that the output of the trained parameter prediction model conforms to the specified physical laws. By introducing information from the physical neural network algorithm, the output of the parameter prediction model can still efficiently approximate the solution to the two-phase fluid phase state problem of crude oil and carbon dioxide when the amount of data is insufficient, ensuring the physical consistency and practical significance of the output of the parameter prediction model, making the prediction more reliable.

[0119] In some embodiments, the historical sample data includes multiple groups of sample data, each group of sample data includes a historical carbon dioxide injection amount, and a historical formation temperature, a historical pressure, and a historical pore radius corresponding to the historical carbon dioxide injection amount. The historical sample data is input into the parameter prediction model to obtain a second gas-liquid equilibrium ratio. When the method is specifically implemented, it may also include the following contents:

[0120] S1: Determine the correlation between every two groups of historical sample data.

[0121] S2: performing anomaly detection processing on the historical sample data according to the correlation degree, and when abnormal data is detected in the historical sample data, performing correction processing on the abnormal data according to a preset interpolation algorithm to obtain corrected historical sample data.

[0122] S3: Inputting the corrected historical sample data into the parameter prediction model to obtain the second gas-liquid equilibrium ratio.

[0123] In some embodiments, the method of inputting the corrected historical sample data into the parameter prediction model to obtain the second gas-liquid equilibrium ratio may further include the following when the method is specifically implemented:

[0124] S1: Determine a weight value corresponding to the sample data according to a relationship between a historical carbon dioxide injection amount in the sample data and a preset concentration threshold.

[0125] S2: Inputting the corrected historical sample data and the weight value corresponding to each group of the sample data into the parameter prediction model to obtain the second gas-liquid equilibrium ratio.

[0126] In some embodiments, the method for determining the degree of association between each two groups of sample data may further include the following when implemented:

[0127] S1: Determine the correlation between each two groups of sample data according to the correlation between the historical carbon dioxide injection amounts, the correlation between the historical formation temperatures, and the correlation between the historical pore radii of each two groups of historical sample data.

[0128] In some embodiments, see Figure 4 As shown, the anomaly detection process is performed on the historical sample data according to the correlation degree. When the method is specifically implemented, the following contents may also be included:

[0129] S1: Classify the historical sample data according to the association degree to obtain multiple classes.

[0130] S2: Acquire second sample data in the first category corresponding to the first sample data; wherein the first category is any one of the multiple categories, and the first sample data is any one of the sample data in the first category.

[0131] S3: Obtain a first distance between the first sample data and each of the second sample data, and determine an average distance based on the first distance.

[0132] S4: Determine a deviation compression value according to the first distance and the average distance.

[0133] S5: Perform logarithmic transformation on the deviation compression value to obtain a multi-dimensional deviation compression index.

[0134] S6: performing standardization processing on the multi-dimensional deviation compression index to obtain a standardized deviation compression index.

[0135] S7: Perform anomaly detection processing on the historical sample data according to the standardized deviation compression index and the preset index threshold.

[0136] Specifically, the above-mentioned affinity index may be a comprehensive affinity index (CAI).

[0137] Furthermore, the correlation between the historical carbon dioxide injection amounts of each two groups of the historical sample data can be obtained according to the following formula:

[0138]

[0139] Among them, X i represents the component content of the first sample data, X j represents the component content of the second sample data, and σ is the bandwidth parameter of the component.

[0140] The correlation between the historical formation temperatures of each two groups of sample data can be obtained according to the following formula:

[0141]

[0142] Among them, T i Represents the formation temperature data of the first sample data, T j represents the formation temperature data of the second sample data, σ T is the bandwidth parameter of the formation temperature.

[0143] The correlation between each two groups of historical pore radii can be obtained according to the following formula:

[0144]

[0145] Among them, R i Represents the pore radius data of the first sample data, R j represents the pore radius data of the second sample data, σ R is the bandwidth parameter of the pore radius.

[0146] Furthermore, the comprehensive correlation can be defined as the Frobenius norm of the correlation matrix (i.e., the matrix determined by the correlation between the historical carbon dioxide injection amounts, the correlation between the historical formation temperatures, and the correlation between the historical pore radii), specifically, as follows:

[0147]

[0148] in, It is an element in the correlation matrix (carbon dioxide injection amount, formation temperature, pore radius), which represents the comprehensive correlation between the historical sample data i and j. CAI is the comprehensive correlation.

[0149] In some embodiments, the first distance between the first sample data and each of the second sample data is obtained, and an average distance is determined according to the first distance, and a deviation compression value is determined according to the first distance and the average distance. In specific implementation, it can be obtained according to the following formula:

[0150]

[0151] Among them, C(x i ) is the deviation compression value, N(x i ) is the first sample data x i Neighborhood of x i is the first sample data, x j is the jth second sample data, d(x i ,x j ) is the first distance between the first sample data and the jth second sample data, d(x i ) represents the average distance.

[0152] In some embodiments, the deviation compression value is logarithmically transformed to obtain a multidimensional deviation compression index. In specific implementation, it can be obtained according to the following formula:

[0153] MDCI(x i )=log(1+C(x i )),

[0154] Among them, MDCI(x i ) is a multidimensional deviation compression index. Further, the deviation compression value is logarithmically transformed to highlight the difference between the deviation compression value in sparse and dense distribution.

[0155] In some embodiments, the multi-dimensional deviation compression index is standardized to obtain a standardized deviation compression index. When implemented specifically, it can be obtained according to the following formula:

[0156]

[0157] Among them, Z MDCI (x i ) represents the standardized multidimensional deviation compression index (MDCI) value, μ MDCI Represents the mean value of the multidimensional deviation compression index (MDCI) of the data set, σ MDCI It represents the standard deviation of the multidimensional deviation compression index (MDCI) value of the data set, SDCI (x i ) represents the normalized deviation compression index.

[0158] In some embodiments, performing anomaly detection processing on the historical sample data according to the standardized deviation compression index and the preset index threshold may include:

[0159] For example, taking the preset index threshold as 0.5 as an example, when the SDCI value of the historical sample data is less than 0.5, it can be determined that there is no abnormality in the historical sample data, that is, the historical sample data can be normal data.

[0160] When the SDCI value of the historical sample data is greater than 0.5, it can be determined that the historical sample data is abnormal, that is, the historical sample data may be abnormal data.

[0161] In addition, for abnormal data, the degree of abnormality of the abnormal data can be determined according to the second preset index threshold, and corresponding data processing methods can be used to repair abnormal data of different abnormal degrees.

[0162] For example, taking the second preset index threshold including the second preset index threshold 1 (such as 0.7) and the second preset index threshold 2 (such as 0.75) as an example, when the SDCI value of the historical sample data is not less than 0.5 and less than 0.7, it can be determined that the historical sample data is slightly abnormal data.

[0163] When the SDCI value of the historical sample data is greater than 0.75, it can be determined that the historical sample data is severely abnormal data.

[0164] Furthermore, for the detected mild abnormal data, the multivariate interpolation method can be used to repair it, which can be calculated according to the following formula:

[0165]

[0166] Among them, Z(x0) is the historical sample data after repair, Z(x i ) is the historical sample data, λ i is the weight coefficient.

[0167] Specifically, during the parameter prediction model training process, a lower weight may be assigned to the repaired historical sample data, while a higher weight may be assigned to the normal historical sample data.

[0168] Based on the above method, the parameter prediction model can reduce the negative impact of abnormal sample data on the training results and maintain a certain degree of confidence. For the key abnormal data in the historical sample data, the key abnormal data can be determined as invalid data and deleted to ensure that the parameter prediction model will not be disturbed by serious abnormal sample data.

[0169] In the prediction process, a processing strategy similar to the training data process is adopted. The current carbon dioxide injection volume of the input target well, as well as the formation temperature, pressure and pore radius corresponding to the target well can be firstly subjected to anomaly detection processing, and then processed according to the degree of anomaly.

[0170] Furthermore, considering that the carbon dioxide content will reach a fairly high value during the carbon dioxide injection development process, which is also a notable feature of the shale oil reservoir development process, in this case, when constructing the parameter prediction model, a larger weight can be given to the data with high carbon dioxide content. In this way, when training the parameter prediction model, it can pay more attention to the second loss function based on physical constraints rather than the first loss function based on data-driven, so that the phase state of the high carbon dioxide content fluid in the nanopores is more in line with the real physical conditions, thereby improving the interpretability of the constructed parameter prediction model.

[0171] As can be seen from the above, the embodiment of this specification provides a method for adjusting the carbon dioxide injection amount based on a deep learning phase model, by obtaining the current carbon dioxide injection amount of the target well, and the formation temperature, pressure and pore radius corresponding to the target well; according to the pre-trained parameter prediction model, based on the current carbon dioxide injection amount of the target well, and the formation temperature, pressure and pore radius corresponding to the target well, determine the phase parameters corresponding to the target well; wherein the parameter prediction model is a model constructed based on a physical information neural network algorithm and a preset state equation, and the parameter prediction model is trained based on historical sample data and the first gas-liquid equilibrium ratio corresponding to the historical sample data, and the first gas-liquid equilibrium ratio is obtained by processing the historical sample data according to a preset gas-liquid equilibrium ratio model, and the preset gas-liquid equilibrium ratio model is constructed by the Wilson equation and the preset state equation; according to the phase parameters corresponding to the target well, judge whether the current carbon dioxide injection amount meets the preset mining demand; when it is determined that the current carbon dioxide injection amount does not meet the preset mining demand, adjust the current carbon dioxide injection amount. In this way, firstly, by constructing a parameter prediction model based on a physical information neural network algorithm and a preset state equation, and then combining the preset gas-liquid equilibrium ratio model constructed by the Wilson equation and the preset state equation, the parameter prediction model is trained to obtain a parameter prediction model with high accuracy and good effect, and then the parameter prediction model can be used to accurately determine the phase parameters with small errors and high reference value. Secondly, according to the phase parameters, on the one hand, it is possible to accurately judge whether the current carbon dioxide injection amount meets the preset mining demand; on the other hand, when the current carbon dioxide injection amount does not meet the preset mining demand, the phase parameters can be used as the basis for adjustment to accurately adjust the current carbon dioxide injection amount, thereby effectively improving the oil recovery efficiency.

[0172] See also Figure 5 As shown, an embodiment of this specification also provides a specific electronic device, wherein the electronic device includes a network communication port 501, a processor 502 and a memory 503, and the above structures are connected through internal cables so that each structure can perform specific data interaction.

[0173] The network communication port 501 can be used to obtain the current carbon dioxide injection volume of the target well, as well as the formation temperature, pressure and pore radius corresponding to the target well.

[0174] The processor 502 can be specifically used to determine the phase parameters corresponding to the target well based on the pre-trained parameter prediction model, the current carbon dioxide injection amount of the target well, and the formation temperature, pressure and pore radius corresponding to the target well; wherein the parameter prediction model is a model constructed based on a physical information neural network algorithm and a preset state equation, and the parameter prediction model is trained based on historical sample data and a first gas-liquid equilibrium ratio corresponding to the historical sample data, and the first gas-liquid equilibrium ratio is obtained by processing the historical sample data according to a preset gas-liquid equilibrium ratio model, and the preset gas-liquid equilibrium ratio model is constructed by the Wilson equation and the preset state equation; according to the phase parameters corresponding to the target well, determine whether the current carbon dioxide injection amount meets the preset mining demand; if it is determined that the current carbon dioxide injection amount does not meet the preset mining demand, adjust the current carbon dioxide injection amount.

[0175] The memory 503 may be specifically used to store corresponding instruction programs.

[0176] Based on the above method, the relevant structural performance of electronic equipment can be effectively utilized, the data processing speed of electronic equipment can be improved, and the carbon dioxide injection amount adjustment method based on the deep learning phase model can be efficiently realized.

[0177] In this embodiment, the network communication port 501 can be a virtual port that is bound to different communication protocols so that different data can be sent or received. For example, the network communication port can be a port responsible for web data communication, a port responsible for FTP data communication, or a port responsible for email data communication. In addition, the network communication port can also be a physical communication interface or communication chip. For example, it can be a wireless mobile network communication chip, such as GSM, CDMA, etc.; it can also be a Wifi chip; it can also be a Bluetooth chip.

[0178] In this embodiment, the processor 502 may be implemented in any appropriate manner. For example, the processor may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) executable by the (micro)processor, a logic gate, a switch, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc. This specification does not limit this.

[0179] In this embodiment, the memory 503 may include multiple levels. In a digital system, anything that can store binary data can be a memory; in an integrated circuit, a circuit with a storage function but no physical form is also called a memory, such as RAM, FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory stick, TF card, etc.

[0180] The embodiment of the present specification also provides a computer-readable storage medium based on the above-mentioned method for adjusting the carbon dioxide injection amount based on the deep learning phase model, wherein the computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed, the following are achieved: obtaining the current carbon dioxide injection amount of the target well, and the formation temperature, pressure and pore radius corresponding to the target well; determining the phase state parameters corresponding to the target well based on the current carbon dioxide injection amount of the target well, and the formation temperature, pressure and pore radius corresponding to the target well according to a pre-trained parameter prediction model; wherein the parameter prediction model is a model constructed based on a physical information neural network algorithm and a preset state equation, and the parameter prediction model is trained based on historical sample data and a first gas-liquid equilibrium ratio corresponding to the historical sample data, and the first gas-liquid equilibrium ratio is obtained by processing the historical sample data according to a preset gas-liquid equilibrium ratio model, and the preset gas-liquid equilibrium ratio model is constructed by the Wilson equation and the preset state equation; judging whether the current carbon dioxide injection amount meets the preset production demand according to the phase state parameters corresponding to the target well; and adjusting the current carbon dioxide injection amount when it is determined that the current carbon dioxide injection amount does not meet the preset production demand.

[0181] In this embodiment, the storage medium includes, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a cache, a hard disk (HDD), or a memory card. The memory may be used to store computer program instructions. The network communication unit may be an interface for network connection communication set in accordance with the standard specified by the communication protocol.

[0182] In this embodiment, the functions and effects specifically implemented by the program instructions stored in the computer-readable storage medium can be explained in comparison with other implementations and will not be described in detail here.

[0183] See also Figure 6 At the software level, the embodiment of this specification also provides a device for adjusting the amount of carbon dioxide injection based on a deep learning phase model, which may specifically include the following structural modules:

[0184] The data acquisition module 601 is used to acquire the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well;

[0185] A parameter determination module 602 is used to determine the phase parameters corresponding to the target well according to a pre-trained parameter prediction model, based on the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well; wherein the parameter prediction model is a model constructed based on a physical information neural network algorithm and a preset state equation, the parameter prediction model is trained based on historical sample data and a first gas-liquid equilibrium ratio corresponding to the historical sample data, the first gas-liquid equilibrium ratio is obtained by processing the historical sample data according to a preset gas-liquid equilibrium ratio model, and the preset gas-liquid equilibrium ratio model is constructed by the Wilson equation and the preset state equation;

[0186] The demand determination module 603 is used to determine whether the current carbon dioxide injection amount meets the preset production demand according to the phase parameters corresponding to the target well;

[0187] The injection amount adjustment module 604 is used to adjust the current carbon dioxide injection amount when it is determined that the current carbon dioxide injection amount does not meet the preset production demand.

[0188] In some embodiments, before the above-mentioned parameter determination module 602, during the specific implementation, historical sample data is obtained; according to the preset gas-liquid equilibrium ratio model, the historical sample data is processed to obtain the first gas-liquid equilibrium ratio; a second gas-liquid equilibrium ratio module is used to input the historical sample data into the parameter prediction model to obtain the second gas-liquid equilibrium ratio; a preset loss function module is used to determine whether the parameter prediction model converges based on the first gas-liquid equilibrium ratio and the second gas-liquid equilibrium ratio according to a preset loss function. If it is determined that the parameter prediction model has not converged, the parameter prediction model is continued to be trained according to the historical sample data until the parameter prediction model converges to obtain a trained parameter prediction model.

[0189] In some embodiments, when the preset loss function module is specifically implemented, the preset loss function includes a first loss function based on data driving and a second loss function based on physical constraints.

[0190] In some embodiments, the historical sample data includes multiple groups of sample data, each group of the sample data includes historical carbon dioxide injection volume, and historical formation temperature, historical pressure and historical pore radius corresponding to the historical carbon dioxide injection volume. When the second gas-liquid balance ratio module is specifically implemented, the correlation determination module is used to determine the correlation between every two groups of the historical sample data; the anomaly detection processing module is used to perform anomaly detection processing on the historical sample data according to the correlation, and when abnormal data is detected in the historical sample data, the abnormal data is corrected according to a preset interpolation algorithm to obtain corrected historical sample data; the second gas-liquid balance ratio submodule is used to input the corrected historical sample data into the parameter prediction model to obtain the second gas-liquid balance ratio.

[0191] In some embodiments, the above-mentioned second gas-liquid equilibrium ratio submodule, when implemented, is used to determine the weight value corresponding to the sample data based on the relationship between the historical carbon dioxide injection amount in the sample data and the preset concentration threshold; the corrected historical sample data and the weight value corresponding to each group of the sample data are input into the parameter prediction model to obtain the second gas-liquid equilibrium ratio.

[0192] In some embodiments, the above-mentioned correlation determination module, when implemented, is used to determine the correlation between each two groups of sample data based on the correlation between the historical carbon dioxide injection amounts of each two groups of the historical sample data, the correlation between the historical formation temperatures, and the correlation between the historical pore radii.

[0193] In some embodiments, the above-mentioned anomaly detection processing module, when specifically implemented, is used to classify the historical sample data according to the correlation degree to obtain multiple classes; obtain second sample data corresponding to the first sample data in the first class; wherein the first class is any one of the multiple classes, and the first sample data is any one of the sample data in the first class; obtain a first distance between the first sample data and each of the second sample data, and determine an average distance based on the first distance; determine a deviation compression value based on the first distance and the average distance; perform logarithmic transformation on the deviation compression value to obtain a multidimensional deviation compression index; perform standardization on the multidimensional deviation compression index to obtain a standardized deviation compression index; perform anomaly detection on the historical sample data based on the standardized deviation compression index and a preset index threshold.

[0194] It should be noted that the units, devices or modules described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described separately by functions divided into various modules. Of course, when implementing this specification, the functions of each module can be implemented in the same or more software and / or hardware, or the modules that implement the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0195] As can be seen from the above, a device for adjusting the carbon dioxide injection amount based on a deep learning phase model provided in the embodiment of this specification, firstly, by constructing a parameter prediction model based on a physical information neural network algorithm and a preset state equation, and then combining the preset gas-liquid equilibrium ratio model constructed by the Wilson equation and the preset state equation, the parameter prediction model is trained to obtain a parameter prediction model with high accuracy and good effect, and then the parameter prediction model can be used to accurately determine the phase parameters with small errors and high reference value. Secondly, according to the phase parameters, on the one hand, it is possible to accurately judge whether the current carbon dioxide injection amount meets the preset mining demand; on the other hand, when the current carbon dioxide injection amount does not meet the preset mining demand, the phase parameters can be used as the basis for adjustment to accurately adjust the current carbon dioxide injection amount, thereby effectively improving the oil recovery efficiency.

[0196] In a specific scenario example, a method for adjusting the carbon dioxide injection amount based on a deep learning phase model provided in this specification can be applied, and the phase parameters of the target well are determined using a parameter prediction model, which solves the problems of low efficiency in phase parameter prediction and high number of iterations. At the same time, the carbon dioxide injection amount is adjusted according to the phase parameters to improve the oil recovery efficiency. The specific implementation process may include the following contents.

[0197] S1: Input the preset N (e.g., 8) carbon dioxide-containing components in the historical sample data into a preset gas-liquid equilibrium ratio model to obtain the gas-liquid equilibrium ratio corresponding to each component as shown in Table 1 below:

[0198] Table 1

[0199]

[0200]

[0201] Furthermore, the preset gas-liquid equilibrium ratio model can be used to calculate the phase equilibrium of crude oil and carbon dioxide two-phase fluid to obtain parameters such as the gas-liquid equilibrium ratio. Among them, the preset state equation in the preset gas-liquid equilibrium ratio model has the influence of nanoporous fluid adsorption. In this way, the parameter prediction model trained by the preset gas-liquid equilibrium ratio model can be used for conventional reservoirs, and can also be used for unconventional oil reservoirs rich in nanopores such as tight and shale. Specifically, when the Gibbs (Gibbs Free Energy, Gibbs) free energy is minimum, the fugacity of each carbon dioxide component in the gas phase and liquid phase tends to be equal, and the system is in equilibrium. The Gibbs free energy calculation method is as follows:

[0202]

[0203] Among them, F I and F II are the liquid and gas phase fractions, respectively, and They are composed of liquid and gas phases respectively; and are the liquid and gas phase fugacity, respectively.

[0204] Specifically, the carbon dioxide component z i , balance ratio K i And the gas phase fraction Q satisfies the following formula:

[0205]

[0206] According to the above parameters, the proportion of each component in the gas and liquid phases can be calculated:

[0207] z i =Qy i +(1-Q)x i

[0208]

[0209] Among them, K i is the gas-liquid equilibrium constant of carbon dioxide component i; x i and i are the mole fractions of component i in the liquid and gas phases, respectively.

[0210] Furthermore, considering that the carbon dioxide content has a large change, a step-by-step interval sampling is adopted to generate historical sample data, the fluid components are re-merged, and the temperature, pressure and number of calculation groups are defined to meet the following formula:

[0211]

[0212]

[0213] Among them, C i is the mole fraction of each carbon dioxide component, C iupper is the upper limit of the mole fraction of each carbon dioxide component, C ilower is the lower limit of the mole fraction of each carbon dioxide component.

[0214] S2: Input N (e.g., 8) carbon dioxide-containing components preset in the historical sample data into the parameter prediction model to obtain the gas-liquid equilibrium ratio corresponding to each component as shown in Table 2 below:

[0215] Table 2

[0216] K1 K2 K3 K4 K5 K6 K7 K8 1.061 1.624 0.9290 0.4572 0.1375 0.0374 0.0140 0.0022 1.023 1.381 0.9377 0.5756 0.2550 0.1054 0.0536 0.0149 1.182 2.117 0.9587 0.3432 0.0550 0.0076 0.0018 0.0001 1.049 1.838 0.8763 0.3392 0.0660 0.0111 0.0030 0.0003 1.054 1.912 0.8844 0.3314 0.0596 0.0092 0.0023 0.0002 1.215 2.079 0.9889 0.3765 0.0681 0.0107 0.0029 0.0002 1.098 1.733 0.9325 0.4196 0.1056 0.0237 0.0079 0.0010 1.033 1.392 0.9486 0.5878 0.2699 0.1162 0.0602 0.0172 1.177 1.911 0.9775 0.4102 0.0902 0.0175 0.0054 0.0006

[0217] See also Figure 7 As shown, the above parameter prediction model has a high accuracy, with an average accuracy of 96.527% for 27 groups of data, and in tests with a large number of calculation groups, the accuracy is also at a high level, reaching 98%.

[0218] In some embodiments, 9 test data out of the 27 tests are taken, and the gas-liquid equilibrium ratio obtained by the physical neural network algorithm in the parameter prediction model is used as the initial value of the gas-liquid equilibrium ratio, and is substituted into the preset state equation in the parameter prediction model to obtain the number of iteration steps and the operation time, and compared with the number of iteration steps and the operation time of the gas-liquid equilibrium ratio model, as shown in FIG. Figure 8 and Fig. 9 As shown, the constructed parameter prediction model (i.e., PINN-EOS model) effectively reduces the number of iterations and calculation time compared to the gas-liquid equilibrium ratio model while ensuring 100% calculation accuracy, thereby significantly improving the calculation efficiency.

[0219] Furthermore, the traditional preset state equation method can realize the calculation of any phase state parameters, but the calculation efficiency is low. The parameter prediction model (i.e., PINN-EOS model) can greatly reduce the number of iterations of the state equation method because the estimated value is close to the accurate value, thereby greatly improving the calculation efficiency.

[0220] In addition, the traditional machine learning model has high computational efficiency, but has large errors, and a model can only predict a certain physical property parameter. However, by using the gas-liquid equilibrium constant as the associated parameter, and substituting the gas-liquid equilibrium ratio obtained by the physical neural network algorithm in the parameter prediction model as the initial value of the gas-liquid equilibrium ratio into the preset state equation in the parameter prediction model, the computational efficiency can be greatly improved while achieving the same accuracy as the preset state equation method, and the calculation of any phase state parameter can be realized.

[0221] In addition, the above parameter prediction model can be applied to systems with high CO2 content. In addition to being used in conventional oil reservoirs, it can also consider the effect of pore size on phase state and be used in unconventional oil reservoirs such as tight / shale.

[0222] Although the present specification provides method operation steps as described in the embodiments or flow charts, more or less operation steps may be included based on conventional or non-creative means. The order of steps listed in the embodiments is only one way of executing the order of many steps, and does not represent a unique execution order. When the device or client product in practice is executed, it can be executed in sequence or in parallel according to the method shown in the embodiments or the drawings (for example, a parallel processor or a multi-threaded processing environment, or even a distributed data processing environment). The term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, product or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such a process, method, product or device. In the absence of more restrictions, it is not excluded that there are other identical or equivalent elements in the process, method, product or device including the elements. The first, second, etc. words are used to represent the name, and do not represent any particular order.

[0223] Those skilled in the art also know that, in addition to implementing the controller in a purely computer-readable program code, the controller can be made to implement the same function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered as a hardware component, and the devices for implementing various functions included therein can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules for implementing the method and structures within the hardware component.

[0224] Through the description of the above embodiments, it can be known that those skilled in the art can clearly understand that the present specification can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present specification can essentially be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the methods described in each embodiment of the present specification or some parts of the embodiments.

[0225] Although the present specification is described through embodiments, those skilled in the art will appreciate that there are many modifications and changes to the present specification without departing from the spirit of the present specification, and it is intended that the appended claims include these modifications and changes without departing from the spirit of the present specification.

Claims

1. A method for adjusting the amount of carbon dioxide injection based on a deep learning phase model, characterized in that: include: Obtaining the current carbon dioxide injection volume of the target well, as well as the formation temperature, pressure and pore radius corresponding to the target well; According to a pre-trained parameter prediction model, based on the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well, the phase parameters corresponding to the target well are determined; wherein the parameter prediction model is a model constructed based on a physical information neural network algorithm and a preset state equation, the parameter prediction model is trained based on historical sample data and a first gas-liquid balance ratio corresponding to the historical sample data, the first gas-liquid balance ratio is obtained by processing the historical sample data according to a preset gas-liquid balance ratio model, and the preset gas-liquid balance ratio model is constructed by the Wilson equation and the preset state equation; According to the phase parameters corresponding to the target well, determining whether the current carbon dioxide injection amount meets the preset production demand; When it is determined that the current carbon dioxide injection amount does not meet the preset production demand, the current carbon dioxide injection amount is adjusted.

2. The method according to claim 1, characterized in that Before determining the phase parameters corresponding to the target well according to the pre-trained parameter prediction model based on the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well, the method further includes: Acquiring the historical sample data; According to the preset gas-liquid equilibrium ratio model, the historical sample data is processed to obtain the first gas-liquid equilibrium ratio; Inputting the historical sample data into the parameter prediction model to obtain a second gas-liquid equilibrium ratio; According to the preset loss function, based on the first gas-liquid equilibrium ratio and the second gas-liquid equilibrium ratio, determine whether the parameter prediction model has converged. If it is determined that the parameter prediction model has not converged, continue to train the parameter prediction model according to the historical sample data until the parameter prediction model converges to obtain a trained parameter prediction model.

3. The method according to claim 2, characterized in that The preset loss function includes a first loss function based on data driving and a second loss function based on physical constraints.

4. The method according to claim 2, characterized in that: The historical sample data includes multiple groups of sample data, each group of sample data includes a historical carbon dioxide injection amount, and a historical formation temperature, a historical pressure, and a historical pore radius corresponding to the historical carbon dioxide injection amount. The step of inputting the historical sample data into the parameter prediction model to obtain a second gas-liquid equilibrium ratio includes: Determine the correlation between each two groups of the historical sample data; According to the correlation degree, anomaly detection processing is performed on the historical sample data, and when abnormal data is detected in the historical sample data, correction processing is performed on the abnormal data according to a preset interpolation algorithm to obtain corrected historical sample data; The corrected historical sample data is input into the parameter prediction model to obtain the second gas-liquid equilibrium ratio.

5. The method according to claim 4, characterized in that The step of inputting the corrected historical sample data into the parameter prediction model to obtain the second gas-liquid equilibrium ratio includes: Determining a weight value corresponding to the sample data according to a relationship between a historical carbon dioxide injection amount in the sample data and a preset concentration threshold; The corrected historical sample data and the weight value corresponding to each group of the sample data are input into the parameter prediction model to obtain the second gas-liquid equilibrium ratio.

6. The method according to claim 4, characterized in that The determining the degree of association between each two groups of sample data comprises: The correlation between each two groups of the sample data is determined based on the correlation between the historical carbon dioxide injection amounts, the correlation between the historical formation temperatures, and the correlation between the historical pore radii of each two groups of the historical sample data.

7. The method according to claim 6, characterized in that The performing anomaly detection processing on the historical sample data according to the association degree includes: According to the association degree, the historical sample data is classified to obtain multiple classes; Acquire second sample data in the first category corresponding to the first sample data; wherein the first category is any one of the multiple categories, and the first sample data is any one of the sample data in the first category; Acquire a first distance between the first sample data and each of the second sample data, and determine an average distance based on the first distance; determining a deviation compression value according to the first distance and the average distance; Performing logarithmic transformation on the deviation compression value to obtain a multidimensional deviation compression index; Performing standardization processing on the multidimensional deviation compression index to obtain a standardized deviation compression index; According to the standardized deviation compression index and the preset index threshold, anomaly detection processing is performed on the historical sample data.

8. A carbon dioxide injection amount adjustment device based on a deep learning phase model, characterized in that: include: A data acquisition module, used to acquire the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well; A parameter determination module, for determining the phase parameters corresponding to the target well according to a pre-trained parameter prediction model, based on the current carbon dioxide injection volume of the target well, and the formation temperature, pressure and pore radius corresponding to the target well; wherein the parameter prediction model is a model constructed based on a physical information neural network algorithm and a preset state equation, the parameter prediction model is trained based on historical sample data and a first gas-liquid equilibrium ratio corresponding to the historical sample data, the first gas-liquid equilibrium ratio is obtained by processing the historical sample data according to a preset gas-liquid equilibrium ratio model, and the preset gas-liquid equilibrium ratio model is constructed by the Wilson equation and the preset state equation; A demand determination module, used to determine whether the current carbon dioxide injection amount meets the preset production demand according to the phase parameters corresponding to the target well; The injection amount adjustment module is used to adjust the current carbon dioxide injection amount when it is determined that the current carbon dioxide injection amount does not meet the preset production demand.

9. An electronic device, characterized in that: It comprises a processor and a memory for storing instructions executable by the processor, and when the processor executes the instructions, the steps of the method for adjusting the carbon dioxide injection amount based on the deep learning phase model as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the instructions are executed by a processor, the steps of the method for adjusting the carbon dioxide injection amount based on a deep learning phase model as described in any one of claims 1 to 7 are implemented.

Images