Combustion front identification method based on multi-source data, medium and equipment

Abstract

The invention belongs to the technical field of heavy oil development, and discloses a combustion front identification method based on multi-source data, a medium and equipment. The method comprises the following steps: collecting various data sources; performing data processing and screening feature data; based on a deep neural network technology, a one-dimensional combustion pipe prediction model, a resistivity sub-prediction model and a pH value regression sub-prediction model are respectively constructed, characteristic weighting is performed on combustion front edge prediction results of the sub-prediction models, and a combustion front edge position prediction model under multi-factor coupling is constructed. Training and optimizing the combustion front position prediction model; and the trained and optimized combustion front edge position prediction model is used for actual production, and the fireflooding combustion front edge position is predicted. According to the combustion front position prediction model, multi-source data information is comprehensively considered, real-time monitoring and accurate prediction of the combustion front position can be achieved, the fireflooding oil extraction efficiency is improved, the exploitation cost is reduced, and technical support is provided for oilfield exploitation.

Description

Technical Field

[0001] This invention belongs to the field of heavy oil development technology, specifically relating to a combustion front identification method, medium, and device based on multi-source data. Background Technology

[0002] Fire flooding technology refers to injecting air into the oil reservoir from a gas injection well, igniting the reservoir with an igniter or chemical agent, and generating heat, gas, water vapor, and gaseous hydrocarbons, which create various displacement effects. Fire flooding technology has advantages such as high recovery rate, low energy consumption, and low pollution, and it has long been considered a highly promising method for the extraction of heavy and extra-heavy oil.

[0003] Fire-enhanced oil recovery, as an important oilfield development technology, relies heavily on the accurate prediction of the combustion front location. This prediction is crucial for optimizing combustion dynamics, improving oil recovery efficiency, and ensuring production safety. However, traditional methods have several limitations, such as dependence on limited experimental conditions and data, low accuracy of models built based on empirical formulas, and the inability to achieve real-time monitoring and rapid response at the production site. Therefore, improving the accuracy of combustion front location prediction has become an urgent problem to be solved in the field of fire-enhanced oil recovery. Summary of the Invention

[0004] This invention aims to address the technical problems existing in the background art by providing a combustion leading edge identification method, medium, and device based on multi-source data. It combines multiple data sources and constructs multiple sub-prediction models based on deep neural network technology. The combustion leading edge prediction results of each sub-prediction model are then feature-weighted. Based on deep neural network technology, a combustion leading edge position prediction model under multi-factor coupling is constructed, trained, and optimized. This model comprehensively considers multi-source data information and can achieve real-time monitoring and accurate prediction of the combustion leading edge position.

[0005] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0006] A combustion leading edge identification method based on multi-source data, the method comprising the following steps:

[0007] Step S1. Data selection, including:

[0008] Fire-driven simulation experiments were conducted using a one-dimensional combustion tube to obtain experimental data on combustion front movement under different production conditions.

[0009] Electromagnetic methods were used to obtain resistivity data around the production well, and the location of the combustion front in the electromagnetic method experiment was obtained based on the resistivity data analysis.

[0010] Obtain the pH value of crude oil using crude oil property analysis data;

[0011] Obtain production data using a real-time production daily report database;

[0012] Step S2. Process the data collected in step S1 and filter the feature data;

[0013] Step S3. Construct a combustion front location prediction model under multi-factor coupling and perform model training and optimization; including:

[0014] Step S31: Based on deep neural network technology, construct a one-dimensional combustion tube prediction model, a resistivity prediction model, and a pH value regression prediction model respectively;

[0015] Step S32: The combustion leading edge prediction results of the above sub-prediction models are weighted by features. Based on deep neural network technology, a combustion leading edge position prediction model under multi-factor coupling is constructed, and the combustion leading edge position prediction model is trained and optimized.

[0016] Step S4. Apply the trained and optimized combustion leading edge position prediction model to actual production to predict the position of the fire-driven combustion leading edge.

[0017] Further, step S2 specifically includes:

[0018] For the one-dimensional combustion tube experimental data obtained in step S1, the combustion front position obtained by electromagnetic method, crude oil pH value and production data, the data are screened and features that are not related to the target variable are excluded. Correlation analysis and statistical methods are used to mine relevant feature data.

[0019] Further, step S31 specifically includes:

[0020] Step S311: Based on the experimental data of the one-dimensional combustion tube, a one-dimensional combustion tube prediction model is trained and constructed using deep neural network technology to obtain the combustion velocity at the combustion leading edge, and the position of the first combustion leading edge is calculated based on the combustion leading edge velocity.

[0021] Step S312: Based on the production data and the combustion leading edge position obtained by the electromagnetic method, a resistivity prediction model is trained and constructed using deep neural network technology to obtain the second combustion leading edge position;

[0022] Step S313: Based on the crude oil pH value and production data, a pH value regression prediction model is trained and constructed using deep neural network (DNN) technology to obtain the crude oil pH value A at a certain moment under different production conditions, and the position of the third combustion front is calculated based on the crude oil pH value A at a certain moment.

[0023] Furthermore, in step S311, the inputs to the one-dimensional combustion tube prediction model include: porosity, oil saturation, gas injection pressure, gas injection rate, and experimental data on oxygen utilization; the outputs are: combustion rate at the combustion front and position of the first combustion front.

[0024] The position of the first combustion leading edge is calculated based on the output combustion leading edge velocity using the following formula:

[0025] First combustion leading edge position = combustion rate × ignition time.

[0026] Further, in step S312, the inputs of the resistivity sub-prediction model include: production well number, injection well number, stroke, number of strokes, oil pressure, casing pressure, daily liquid production, daily oil production, daily gas production, water cut, production wellhead temperature, daily gas injection volume, and gas injection pressure; the output is: the position of the second combustion front.

[0027] Further, in step S313, the inputs of the pH value regression sub-prediction model include: production well number, injection well number, stroke, number of strokes, oil pressure, casing pressure, daily liquid production, daily oil production, daily gas production, water cut, production wellhead temperature, daily gas injection volume, and gas injection pressure; the outputs are: crude oil pH value A at a certain moment, and the position of the third combustion front.

[0028] The position of the third combustion front is calculated based on the pH value A of the output crude oil at a certain moment using the following formula:

[0029]

[0030] In the above formula, X is the position of the third combustion leading edge; A is the pH value of the crude oil at a certain moment; A0 is the base pH value of the crude oil, which is a known fixed value; A m is the maximum pH value of crude oil, which is a known fixed value; L is the distance between injection and production wells.

[0031] Furthermore, step S32 specifically includes:

[0032] Step S321: Initially set three self-learning weighted summation weight coefficients, and sum the calculation results of the first combustion leading edge position, the second combustion leading edge position, and the third combustion leading edge position to calculate the fourth combustion leading edge position;

[0033] Step S322: Based on the positions of the first combustion leading edge, the second combustion leading edge, the third combustion leading edge, and the fourth combustion leading edge, a combustion leading edge position prediction model under multi-factor coupling is trained and constructed using deep neural network technology, and the final combustion leading edge position prediction result is output.

[0034] Furthermore, step S322 specifically includes the following process:

[0035] Step S3221: For the three sub-prediction models, use the corresponding feature data processed in step S2 as input to obtain the first combustion leading edge position, the second combustion leading edge position, the third combustion leading edge position and the fourth combustion leading edge position respectively;

[0036] Step S3222: Compare the positions of the first, second, third, and fourth combustion fronts with the actual combustion front positions, respectively. Calculate their respective loss functions using the mean squared error of the mean squared error (MSE) and sum them to obtain the overall loss function of the combustion front prediction model under multi-factor coupling.

[0037] L MSE =L 一维燃烧管 +L 电磁法 +L PH值法 +L 加权预测

[0038] In the above formula, L MSE L represents the overall loss function of the combustion front prediction model. 一维燃烧管 L represents the first loss function corresponding to the one-dimensional combustion tube prediction model; 电磁法 L is the second loss function corresponding to the resistivity sub-prediction model; PH值法 L is the third loss function corresponding to the pH value regressor prediction model; 加权预测 To calculate the fourth loss function corresponding to the fourth combustion leading edge position using weighted summation;

[0039] Step S3223: Based on the overall loss function of the combustion front prediction model, the SGD optimization algorithm is selected to iteratively optimize the weighted summation coefficients in step S321 and the model parameters in the three sub-prediction models. After reducing the overall loss function to a set threshold, the optimized weighted summation coefficients and the model parameters in the three sub-prediction models are obtained, thus obtaining the final optimized combustion front prediction model under multi-factor coupling. The prediction result of the fourth combustion front position output by the final optimized combustion front prediction model is used as the final combustion front position prediction result.

[0040] Furthermore, step S32, during the training optimization process, also includes the following verification step:

[0041] The trained combustion front position prediction model was tested using both experimental data and actual production data. By comparing the model's predicted values ​​of the combustion front position with experimental observations or actual oilfield production data, the effectiveness and reliability of the combustion front prediction model were verified, evaluated, and optimized.

[0042] Additionally, a computer-readable storage medium is provided, which stores a computer program, the computer program including program instructions that, when executed by a computer, cause the computer to perform the method described in any of the preceding claims.

[0043] In addition, an electronic device is provided, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the program, it implements the method as described in any of the preceding claims.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] This invention achieves accurate prediction and real-time dynamic monitoring of the combustion front position during fire-driven oil recovery. Specifically, it combines multiple data sources, including one-dimensional combustion tube experimental data, resistivity data obtained through electromagnetic methods, crude oil pH values ​​obtained from crude oil characteristic analysis data, and production data from a real-time production daily report database. Based on deep neural network (DNN) technology, it constructs a one-dimensional combustion tube sub-prediction model, a resistivity sub-prediction model, and a pH value regression sub-prediction model. The combustion front prediction results of these sub-prediction models are then feature-weighted, and based on DNN technology, a multi-factor coupled combustion front position prediction model is constructed. This model is trained and optimized before being used to predict the combustion front position in actual production. This trained and optimized multi-factor coupled combustion front position prediction model comprehensively considers multi-source data information, enabling accurate prediction of the fire front position. It effectively improves the accuracy and real-time performance of combustion front position prediction, filling a gap in existing technology and providing more reliable technical support and decision-making basis for oilfield development.

[0046] This invention overcomes the shortcomings of traditional methods by comprehensively utilizing multi-source data, enabling more accurate monitoring and control of combustion dynamics during fire flooding. This allows for accurate monitoring and dynamic regulation of the combustion front position during fire flooding, improving the production efficiency of fire flooding oil recovery, reducing extraction costs, and providing an important guarantee for the safe and efficient development of oilfield resources. Attached Figure Description

[0047] Figure 1 This is a flowchart of the combustion leading edge identification method according to an embodiment of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1

[0050] Combination Figure 1 As shown, this embodiment of the invention provides a combustion front identification method based on multi-source data. Based on a deep neural network (DNN), it comprehensively utilizes multiple data sources such as one-dimensional combustion tube experimental data, resistivity data obtained by electromagnetic method, and crude oil pH value obtained by well logging to construct a combustion front position prediction model. Through multiple data fusion methods, it achieves high-precision prediction of the fire front position.

[0051] The method includes the following steps:

[0052] Step S1. Data selection, specifically including:

[0053] Fire-driven simulation experiments were conducted using a one-dimensional combustion tube to obtain experimental data on combustion front movement under different production conditions.

[0054] Electromagnetic method experiments were used to obtain resistivity data around the production well, and the location of the combustion front in the electromagnetic method experiment was obtained by analyzing the resistivity data. In the electromagnetic method, the location of the combustion front is a local minimum where the resistivity changes from high to low and then back to high. By using the local minimum of resistivity, the location of the combustion front can be predicted and used as training data for subsequent models.

[0055] The pH value of crude oil is obtained using crude oil characteristic analysis data. The crude oil pH value is obtained by further analyzing the crude oil characteristic analysis data through complex and expensive crude oil characteristic analysis experiments. It is also used as subsequent model training data.

[0056] Utilize the real-time daily production report database to obtain production data such as daily liquid production, daily oil production, and daily gas production.

[0057] Step S2. Data processing, specifically including:

[0058] For the one-dimensional combustion tube experimental data obtained in step S1, the combustion front position obtained by electromagnetic method, crude oil pH value and production data, the data are screened and features that are not related to the target variable are excluded. Correlation analysis and statistical methods are used to mine relevant feature data.

[0059] Step S3. Construct a combustion front location prediction model under multi-factor coupling, and perform model training and optimization, including:

[0060] Step S31: Based on deep neural network (DNN) technology, construct a one-dimensional combustion tube prediction model, a resistivity prediction model, and a pH regression prediction model respectively;

[0061] Step S32: The combustion leading edge prediction results of the above sub-prediction models are weighted by features. Based on deep neural network (DNN) technology, a combustion leading edge position prediction model under multi-factor coupling is constructed, and the combustion leading edge position prediction model is trained and optimized.

[0062] This optimized, multi-factor coupled combustion leading edge position prediction model takes into account multi-source data information and can achieve accurate prediction of the fire line position.

[0063] Step S31 specifically includes:

[0064] Step S311: Based on the experimental data of the one-dimensional combustion tube, a one-dimensional combustion tube prediction model is trained and constructed using deep neural network (DNN) technology to obtain the combustion velocity at the combustion leading edge, and the position of the first combustion leading edge is calculated based on the combustion leading edge velocity.

[0065] The inputs to the one-dimensional combustion tube prediction model include: porosity, oil saturation, gas injection pressure, gas injection rate, and experimental data on oxygen utilization; the outputs are: combustion rate at the combustion front and position of the first combustion front.

[0066] The position of the first combustion leading edge is calculated based on the output combustion leading edge velocity using the following formula:

[0067] First combustion leading edge position = combustion rate × ignition time.

[0068] Step S312: Based on the production data and the combustion leading edge position obtained by the electromagnetic method, a resistivity prediction model is trained and constructed using deep neural network (DNN) technology to obtain the second combustion leading edge position. It should be noted that the production data here is the input of the model, and the combustion leading edge position obtained by the electromagnetic method is the model training data.

[0069] The inputs to the resistivity sub-prediction model include: production well number, injection well number, stroke, number of strokes, oil pressure, casing pressure, daily liquid production, daily oil production, daily gas production, water cut, production wellhead temperature, daily gas injection volume, and gas injection pressure; the output is: the position of the second combustion leading edge.

[0070] Step S313: Based on the crude oil pH value and production data, a pH value regression prediction model is trained and constructed using deep neural network (DNN) technology to obtain the pH value A of crude oil at a certain moment under different production conditions, and the position of the third combustion front is calculated based on the pH value A of crude oil at a certain moment; it should be noted that the production data here is the input of the model, and the crude oil pH value is the training data of the model.

[0071] The inputs to the pH regression sub-prediction model include: production well number, injection well number, stroke, number of strokes, oil pressure, casing pressure, daily liquid production, daily oil production, daily gas production, water cut, production wellhead temperature, daily gas injection volume, and gas injection pressure; the outputs are: the pH value A of crude oil at a certain moment and the position of the third combustion front.

[0072] Since there is a linear relationship between the combustion front position and the crude oil pH value, the combustion front position can be predicted by calculating the relationship between the crude oil pH value and the injection-production well spacing. Therefore, the third combustion front position is calculated based on the output crude oil pH value A at a certain moment using the following formula:

[0073]

[0074] In the above formula, X is the position of the third combustion leading edge; A is the pH value of the crude oil at a certain moment; A0 is the base pH value of the crude oil, which is a known fixed value; A m is the maximum pH value of crude oil, which is a known fixed value; L is the distance between injection and production wells.

[0075] Step S32 specifically includes:

[0076] Step S321: Initially set three self-learning weighted summation weight coefficients, and sum the calculation results of the first combustion leading edge position, the second combustion leading edge position, and the third combustion leading edge position to calculate the fourth combustion leading edge position;

[0077] Step S322: Based on the positions of the first combustion leading edge, the second combustion leading edge, the third combustion leading edge, and the fourth combustion leading edge, a combustion leading edge position prediction model under multi-factor coupling is trained and constructed using deep neural network (DNN) technology, and the final combustion leading edge position prediction result is output.

[0078] In step S322, the combustion leading edge position prediction model is trained and optimized to ensure that the trained combustion leading edge position prediction model can accurately capture the key information and corresponding prediction modes in the data (i.e., one-dimensional combustion tube experimental method prediction mode, electromagnetic method prediction mode and pH value method prediction mode), and has good generalization ability.

[0079] Step S322 specifically includes:

[0080] Step S3221: For the three sub-prediction models, use the corresponding feature data processed in step S2 as input to obtain the first combustion leading edge position, the second combustion leading edge position, the third combustion leading edge position and the fourth combustion leading edge position respectively;

[0081] Step S3222: Compare the positions of the first, second, third, and fourth combustion fronts with the actual combustion front positions, respectively. Calculate their respective loss functions using the mean squared error of the mean squared error (MSE) and sum them to obtain the overall loss function of the combustion front prediction model under multi-factor coupling.

[0082] L MSE =L 一维燃烧管 +L 电磁法 +L PH值法 +L 加权预测

[0083] In the above formula, L MSE L represents the overall loss function of the combustion front prediction model. 一维燃烧管 L represents the first loss function corresponding to the one-dimensional combustion tube prediction model; 电磁法 L is the second loss function corresponding to the resistivity sub-prediction model; PH值法 L is the third loss function corresponding to the pH value regressor prediction model; 加权预测 To calculate the fourth loss function corresponding to the fourth combustion leading edge position using weighted summation;

[0084] Step S3223: Based on the overall loss function of the combustion front prediction model, the SGD (Stochastic Gradient Descent) optimization algorithm is selected to iteratively optimize the weighted summation coefficients in step S321 and the model parameters in the three sub-prediction models. After reducing the overall loss function to a set threshold, the optimized weighted summation coefficients and the model parameters in the three sub-prediction models are obtained, thus obtaining the final optimized combustion front prediction model under multi-factor coupling. The prediction result of the fourth combustion front position output by the final optimized combustion front prediction model is close to the true value and is used as the final combustion front position prediction result output.

[0085] Step S32, during the training optimization process, also includes the following verification step:

[0086] The trained combustion front position prediction model was tested using both experimental data and actual production data. By comparing the model's predicted values ​​of the combustion front position with experimental observations / actual oilfield production data, the effectiveness and reliability of the combustion front prediction model were verified, evaluated, and optimized to ensure that the combustion front prediction model has good predictive performance in practical applications.

[0087] Step S4. Apply the trained and optimized combustion leading edge position prediction model to actual production to predict the position of the fire-driven combustion leading edge.

[0088] The combustion front position prediction model constructed by the above method in this embodiment of the invention can obtain the current combustion front position by inputting real-time production data into the model. This model not only considers various production conditions and experimental data, but also realizes dynamic monitoring and real-time prediction of the combustion front position based on the real-time input production data, providing strong technical support for the efficient development of fire-driven oil recovery.

[0089] Example 2

[0090] This invention provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the method described in Embodiment 1.

[0091] Example 3

[0092] This invention provides an electronic device, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the method described in Embodiment 1.

[0093] The above description is merely an embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the scope of the present invention should be included within the protection scope of the present invention.

Claims

1. A combustion leading edge identification method based on multi-source data, characterized in that, The method includes the following steps: Step S1. Data selection, including: Fire-driven simulation experiments were conducted using a one-dimensional combustion tube to obtain experimental data on combustion front movement under different production conditions. Electromagnetic methods were used to obtain resistivity data around the production well, and the location of the combustion front in the electromagnetic method experiment was obtained based on the resistivity data analysis. Obtain the pH value of crude oil using crude oil property analysis data; Obtain production data using a real-time production daily report database; Step S2. Process the data collected in step S1 and filter the feature data; Step S3. Construct a combustion front location prediction model under multi-factor coupling and perform model training and optimization; including: Step S31: Based on deep neural network technology, construct a one-dimensional combustion tube prediction model, a resistivity prediction model, and a pH value regression prediction model respectively; Step S32: The combustion leading edge prediction results of the above sub-prediction models are weighted by features. Based on deep neural network technology, a combustion leading edge position prediction model under multi-factor coupling is constructed, and the combustion leading edge position prediction model is trained and optimized. Step S4. Apply the trained and optimized combustion leading edge position prediction model to actual production to predict the position of the fire-driven combustion leading edge.

2. The method according to claim 1, characterized in that, Step S2 specifically includes: For the one-dimensional combustion tube experimental data obtained in step S1, the combustion front position obtained by electromagnetic method, crude oil pH value and production data, the data are screened and features that are not related to the target variable are excluded. Correlation analysis and statistical methods are used to mine relevant feature data.

3. The method according to claim 1, characterized in that, Step S31 specifically includes: Step S311: Based on the experimental data of the one-dimensional combustion tube, a one-dimensional combustion tube prediction model is trained and constructed using deep neural network technology to obtain the combustion velocity at the combustion leading edge, and the position of the first combustion leading edge is calculated based on the combustion leading edge velocity. Step S312: Based on the production data and the combustion leading edge position obtained by the electromagnetic method, a resistivity prediction model is trained and constructed using deep neural network technology to obtain the second combustion leading edge position; Step S313: Based on the crude oil pH value and production data, a pH value regression prediction model is trained and constructed using deep neural network (DNN) technology to obtain the crude oil pH value A at a certain moment under different production conditions, and the position of the third combustion front is calculated based on the crude oil pH value A at a certain moment.

4. The method according to claim 3, characterized in that, In step S311, the inputs to the one-dimensional combustion tube prediction model include: porosity, oil saturation, gas injection pressure, gas injection rate, and experimental data on oxygen utilization; the outputs are: combustion rate at the combustion front and position of the first combustion front. The position of the first combustion leading edge is calculated based on the output combustion leading edge velocity using the following formula: First combustion leading edge position = combustion rate × ignition time.

5. The method according to claim 3, characterized in that, In step S312, the inputs of the resistivity sub-prediction model include: production well number, injection well number, stroke, number of strokes, oil pressure, casing pressure, daily liquid production, daily oil production, daily gas production, water cut, production wellhead temperature, daily gas injection volume, and gas injection pressure; the output is: the position of the second combustion leading edge.

6. The method according to claim 3, characterized in that, In step S313, the inputs of the pH value regression sub-prediction model include: production well number, injection well number, stroke, number of strokes, oil pressure, casing pressure, daily liquid production, daily oil production, daily gas production, water cut, production wellhead temperature, daily gas injection volume, and gas injection pressure; the outputs are: crude oil pH value A at a certain moment and the position of the third combustion front. The position of the third combustion front is calculated based on the pH value A of the output crude oil at a certain moment using the following formula: In the above formula, X is the position of the third combustion leading edge; A is the pH value of the crude oil at a certain moment; A0 is the base pH value of the crude oil, which is a known fixed value; A m is the maximum pH value of crude oil, which is a known fixed value; L is the distance between injection and production wells.

7. The method according to claim 3, characterized in that, Step S32 specifically includes: Step S321: Initially set three self-learning weighted summation weight coefficients, and sum the calculation results of the first combustion leading edge position, the second combustion leading edge position, and the third combustion leading edge position to calculate the fourth combustion leading edge position; Step S322: Based on the positions of the first combustion leading edge, the second combustion leading edge, the third combustion leading edge, and the fourth combustion leading edge, a combustion leading edge position prediction model under multi-factor coupling is trained and constructed using deep neural network technology, and the final combustion leading edge position prediction result is output.

8. The method according to claim 7, characterized in that, Step S322 specifically includes: Step S3221: For the three sub-prediction models, use the corresponding feature data processed in step S2 as input to obtain the first combustion leading edge position, the second combustion leading edge position, the third combustion leading edge position and the fourth combustion leading edge position respectively; Step S3222: Compare the positions of the first, second, third, and fourth combustion fronts with the actual combustion front positions, respectively. Calculate their respective loss functions using the mean squared error of the mean squared error (MSE) and sum them to obtain the overall loss function of the combustion front prediction model under multi-factor coupling. L MSE L 一维燃烧管 +L 电磁法 +L PH值法 +L 加权预测 In the above formula, L MSE L represents the overall loss function of the combustion front prediction model. 一维燃烧管 L represents the first loss function corresponding to the one-dimensional combustion tube prediction model; 电磁法 L is the second loss function corresponding to the resistivity sub-prediction model; PH值法 L is the third loss function corresponding to the pH value regressor prediction model; 加权预测 To calculate the fourth loss function corresponding to the fourth combustion leading edge position using weighted summation; Step S3223: Based on the overall loss function of the combustion front prediction model, the SGD optimization algorithm is selected to iteratively optimize the weighted summation coefficients in step S321 and the model parameters in the three sub-prediction models. After reducing the overall loss function to a set threshold, the optimized weighted summation coefficients and the model parameters in the three sub-prediction models are obtained, thus obtaining the final optimized combustion front prediction model under multi-factor coupling. The prediction result of the fourth combustion front position output by the final optimized combustion front prediction model is used as the final combustion front position prediction result.

9. The method according to claim 3, characterized in that, Step S32, during the training optimization process, also includes the following verification step: The trained combustion front position prediction model was tested using both experimental data and actual production data. By comparing the model's predicted values ​​of the combustion front position with experimental observations or actual oilfield production data, the effectiveness and reliability of the combustion front prediction model were verified, evaluated, and optimized.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a computer, cause the computer to perform the method as described in any one of claims 1-9.

11. An electronic device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-9.

Images