A method and system for predicting the production of polymer flooding reservoirs based on a double attention mechanism and a storage medium

Abstract

The application discloses a kind of based on dual attention mechanism's polymer flooding reservoir production prediction method and system and storage medium, it is related to oilfield development field, to solve the problem that existing method is difficult to simultaneously capture the bidirectional dependent relationship and local feature of long time series data, and the effect of feature weight distribution and time series information focus is insufficient.The technical points of the present application include: S1, the field development data of polymer flooding reservoir is collected and optimized features are screened;S2, construct CNN-BiLSTM-DAM fusion model, the function realization process of model is as follows: based on convolution layer, the local feature of input data is extracted, the dynamic weight of each time step local feature is allocated by attention mechanism layer, the forward and backward dependent relationship of weighted time series feature data is captured by the bidirectional recurrent structure of BiLSTM layer, the weight adjustment of time series feature output by BiLSTM layer is carried out by attention mechanism layer, and then the global time dependence is captured by BiLSTM layer;S3, the optimized features are input into the trained CNN-BiLSTM-DAM fusion model, and the production of polymer flooding reservoir is predicted.

Description

Technical Field

[0001] This invention relates to the field of oilfield development technology, and more specifically, to a method, system, and storage medium for predicting the production of polymer flooding reservoirs based on a dual attention mechanism. Background Technology

[0002] Polymer flooding reservoir development includes different stages. In the early stage of production, only water injection is carried out and there is no polymer injection data. When the amount of polymer injection reaches a certain level and the oilfield production indicators do not improve significantly, based on professional knowledge in the oilfield field, the block can stop polymer injection and enter the later stage. In this stage, no more polymer is injected, and development is carried out by relying on the remaining pressure of polymer injection or by continuing water injection. There are no longer any polymer injection characteristics.

[0003] Existing methods for predicting production in polymer flooded reservoirs partially rely on polymer injection parameters to explore changes in crude oil production, but these methods fail due to a lack of polymer injection data during non-polymer injection phases. Traditional time-series forecasting methods, such as univariate time-series forecasting (e.g., ARIMA, SARIMA models), only focus on the temporal relationships of a single variable, failing to integrate multivariate features such as the number of wells opened and the amount of fluid injected. Single recurrent neural networks (RNNs) suffer from the vanishing gradient problem, while long short-term memory networks (LSTMs), although capable of handling long-term time-series data, struggle to capture bidirectional dependencies simultaneously. Convolutional neural networks (CNNs), while capable of extracting local features, neglect long-term dependencies in time-series data. Furthermore, single attention mechanisms have limited effectiveness in feature weight allocation and temporal information focusing, failing to simultaneously optimize the accuracy of feature extraction and time-series processing.

[0004] Therefore, for the production prediction of polymer flooding reservoirs in the non-polymer injection stage, there is an urgent need for a prediction method that can integrate multivariate characteristics, capture long-term bidirectional dependence, and dynamically optimize characteristics and time series weights. Summary of the Invention

[0005] The technical problem to be solved by this invention is:

[0006] Existing time-series reservoir production methods integrate multivariate features, making it difficult to simultaneously capture the bidirectional dependencies and local features of long-term time-series data. They also lack the ability to effectively allocate feature weights and focus on time-series information, resulting in an inability to accurately predict production trends during this period.

[0007] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0008] This invention provides a method for predicting production in polymer flooding reservoirs based on a dual attention mechanism, comprising the following steps:

[0009] S1. Collect field development data of polymer flooding reservoirs, clean the data, and screen and optimize features through relevant analysis to obtain an optimized feature set;

[0010] S2. Construct a CNN-BiLSTM-DAM fusion model. The model's functional implementation process is as follows: extract local features of the input data based on the convolutional layer, assign dynamic weights to the local features at each time step through the attention mechanism layer, capture the forward and backward dependencies of the weighted temporal feature data through the bidirectional recurrent structure of the BiLSTM layer, adjust the weights of the temporal features output by the BiLSTM layer through the attention mechanism layer, and then capture the global temporal dependencies through the BiLSTM layer.

[0011] S3. Input the optimized feature set into the trained CNN-BiLSTM-DAM fusion model to predict the production of polymer flooding reservoirs.

[0012] Furthermore, the CNN-BiLSTM-DAM fusion model includes: a multivariate temporal feature input layer, a convolutional layer, a first attention mechanism layer, a first BiLSTM layer, a second attention mechanism layer, a second BiLSTM layer, a Dropout layer, and an output layer. The convolutional layer is used to extract local features from the input data. The first attention mechanism layer is used to assign dynamic weights to the local features at each time step. The first BiLSTM layer is used to capture the forward and backward dependencies of the temporal data through a bidirectional recurrent structure. The second attention mechanism layer is used to adjust the weights of the temporal features output by the first BiLSTM layer, focusing on temporal information that significantly affects the prediction results. The second BiLSTM layer is used to capture global temporal dependencies.

[0013] Furthermore, the convolutional layer adopts one-dimensional convolution (Conv1D), with the number of filters=64, the kernel size=1, the activation function='relu', and the padding mode='same'.

[0014] Furthermore, the number of hidden units in the BiLSTM layer is BiLSTM_UNITS=64.

[0015] Furthermore, the output layer is a fully connected layer with an output dimension of 1 and an activation function of linear, used to output the predicted production value of polymer flooding reservoirs.

[0016] Furthermore, the feature selection and optimization described in S1 through correlation analysis specifically involves using Pearson correlation analysis to select and optimize features. The formula for calculating the Pearson correlation coefficient is as follows:

[0017]

[0018] in, and Let be the values ​​of the two variables for the i-th observation. and These are the means of the two variables.

[0019] Furthermore, the data cleaning described in S1 includes filling in missing values ​​in the data and interpolating outliers in the data.

[0020] Furthermore, during the training process of the CNN-BiLSTM-DAM fusion model, the root mean square error (RMSE), root mean square percentage error (MAPE), and coefficient of determination (COP) are used. As a model evaluation index, the calculation formulas are as follows:

[0021]

[0022]

[0023]

[0024] in, This is the actual value. For predicted values, For actual production, To predict production, Let n be the mean of the actual output, and n be the sample size.

[0025] This invention provides a polymer flooding reservoir production prediction system based on a dual attention mechanism. The system has a program module corresponding to the steps of any of the above-described technical solutions, and executes the steps in the polymer flooding reservoir production prediction method based on the dual attention mechanism when running.

[0026] The present invention also provides a computer-readable storage medium storing a computer program configured to, when invoked by a processor, implement the steps in the polymer flooding reservoir production prediction method based on the dual attention mechanism described in any of the above technical solutions.

[0027] Compared with the prior art, the beneficial effects of the present invention are:

[0028] This invention proposes a CNN-BiLSTM-DAM fusion model based on historical time-series data of polymer flooding reservoirs since their exploitation. CNN effectively extracts local features, while BiLSTM captures long-term bidirectional dependencies, enhancing the processing capability for complex time-series data and enabling accurate reservoir production prediction covering the entire development cycle of the oilfield. By dynamically adjusting feature and time-series weights through the dual attention mechanism, the model focuses more on key information and reduces noise impact. Compared to single models or single-layer attention mechanisms, this significantly reduces prediction error (RMSE reduced to 5218, MAPE reduced to 0.09), increases the R² value (up to 0.97), and enhances prediction accuracy and robustness. Attached Figure Description

[0029] Figure 1 This is a polymer flooding reservoir production prediction method based on a dual attention mechanism in an embodiment of the present invention;

[0030] Figure 2 This is a diagram of the CNN-BiLSTM-DAM fusion model architecture in an embodiment of the present invention;

[0031] Figure 3 This is a trend diagram of oilfield time-series characteristics in an embodiment of the present invention;

[0032] Figure 4 This is a Pearson correlation heatmap of time-series data in an embodiment of the present invention;

[0033] Figure 5 Loss index graph for training CNN-BiLSTM-DAM fusion models under different Epochs in embodiments of the present invention;

[0034] Figure 6 This is a fitting effect diagram of the CNN-BiLSTM-DAM fusion model in an embodiment of the present invention. Detailed Implementation

[0035] To enable those skilled in the art to better understand the present invention, exemplary embodiments or examples of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described embodiments or examples are merely some, not all, of the embodiments or examples of the present invention. All other embodiments or examples obtained by those skilled in the art based on the embodiments or examples of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0036] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0037] Specific Implementation Plan 1: Combining Figures 1 to 2 As shown, this invention provides a method for predicting the production of polymer flooding reservoirs based on a dual attention mechanism, comprising the following steps:

[0038] S1. Collect field development data of polymer flooding reservoirs, clean the data, and screen and optimize features through relevant analysis to obtain an optimized feature set;

[0039] S2. Construct a CNN-BiLSTM-DAM fusion model. The model's functional implementation process is as follows: extract local features of the input data based on the convolutional layer, assign dynamic weights to the local features at each time step through the attention mechanism layer, capture the forward and backward dependencies of the weighted temporal feature data through the bidirectional recurrent structure of the BiLSTM layer, adjust the weights of the temporal features output by the BiLSTM layer through the attention mechanism layer, and then capture the global temporal dependencies through the BiLSTM layer.

[0040] S3. Input the optimized feature set into the trained CNN-BiLSTM-DAM fusion model to predict the production of polymer flooding reservoirs.

[0041] This invention is based on the Dual Attention Mechanism (DAM), which combines the Channel Attention Mechanism (CAM) and the Spatial Attention Mechanism (SAM). By adding the attention mechanism between the CNN and the two BiLSTM layers, the model's attention to key time steps is increased, and the model's ability to identify important features is enhanced. This helps the model to better handle data with noisy or uneven information distribution, thereby improving the model's stability and accuracy.

[0042] Specific Implementation Scheme Two: The CNN-BiLSTM-DAM fusion model includes: a multivariate temporal feature input layer, a convolutional layer, a first attention mechanism layer, a first BiLSTM layer, a second attention mechanism layer, a second BiLSTM layer, a Dropout layer, and an output layer. The convolutional layer is used to extract local features from the input data. The first attention mechanism layer is used to assign dynamic weights to the local features at each time step. The first BiLSTM layer is used to capture the forward and backward dependencies of the temporal data through a bidirectional recurrent structure. The second attention mechanism layer adjusts the weights of the temporal features output by the first BiLSTM layer, focusing on temporal information that significantly affects the prediction results. The second BiLSTM layer is used to capture global temporal dependencies. Other aspects of this implementation scheme are the same as in Specific Implementation Scheme One.

[0043] Specific implementation plan three: The CNN-BiLSTM-DAM fusion model includes:

[0044] Multivariate time series feature input layer: Set the time step TIME_STEPS=24, that is, use the feature data of the previous 24 months to predict the output at the next time point, and the input feature dimension INPUT_DIMS=6;

[0045] Convolutional layer: One-dimensional convolution operation (Conv1D) is used, with the number of filters set to 64, kernel size set to 1, activation function set to 'relu', and padding mode set to 'same', to extract local features of the input data; the kernel size is set to 1 to ensure that the continuity of temporal data is not disrupted.

[0046] First attention mechanism layer: Located between the convolutional layer and the first BiLSTM layer, it strengthens the influence of key features by assigning dynamic weights to local features at each time step;

[0047] First BiLSTM layer: Set the number of hidden units to BiLSTM_UNITS=64, and capture the forward and backward dependencies of timing data through a bidirectional loop structure;

[0048] The second attention mechanism layer is located between the two BiLSTM layers. It adjusts the weights of the temporal features output by the first BiLSTM layer, focuses on the temporal information that has a significant impact on the prediction results, and reduces noise interference.

[0049] The second BiLSTM layer: Based on the output of the first layer, it further captures global time-dependent patterns, with the same number of hidden units as the first layer;

[0050] Dropout layer: A dropout layer is set after both the convolutional layer and the BiLSTM layer to prevent overfitting of the model by randomly deactivating some neurons; this ensures that the model still has high prediction accuracy and robustness in the non-polymer injection stage (without polymer injection data) of polymer flooding reservoirs.

[0051] Output layer: A fully connected (Dense) layer with an output dimension of 1 and an activation function of linear, outputting the predicted monthly production value of polymer flooding reservoirs. This implementation scheme is otherwise identical to specific implementation scheme two.

[0052] Specific Implementation Plan Four: As described in S1, features are selected for optimization through correlation analysis. Specifically, Pearson correlation analysis is used to select features for optimization. The formula for calculating the Pearson correlation coefficient is:

[0053]

[0054] in, and Let be the values ​​of the two variables for the i-th observation. and These are the means of the two variables.

[0055] This implementation plan is otherwise the same as Implementation Plan 3.

[0056] Specific Implementation Scheme 5: The data cleaning described in S1 includes filling in missing values ​​and interpolating outliers. All other aspects of this implementation scheme are the same as Specific Implementation Scheme 4.

[0057] Specific Implementation Plan Six: S1 also includes trend analysis, periodic analysis, seasonality analysis, and autocorrelation analysis of the data. All other aspects of this implementation plan are the same as Specific Implementation Plan Five.

[0058] Specific implementation plan seven: During the training process of the CNN-BiLSTM-DAM fusion model, the root mean square error (RMSE), root mean square percentage error (MAPE), and coefficient of determination (COP) are used. As a model evaluation index, the calculation formulas are as follows:

[0059]

[0060]

[0061]

[0062] in, This is the actual value. For predicted values, For actual production, To predict production, Let n be the mean of the actual output, and n be the sample size.

[0063] In time series analysis, each variable represents an observation at different time points, and the Pearson correlation coefficient measures the strength and direction of the correlation between these variables over time. This implementation scheme is otherwise identical to specific implementation scheme six.

[0064] The invention proposes a polymer flooding reservoir production prediction method (algorithm) based on a dual attention mechanism, which is the underlying technical core of the invention. Various products can be derived from the algorithm.

[0065] Based on the method proposed in this invention, a polymer flooding reservoir production prediction system based on a dual attention mechanism is developed using a programming language. This system has program modules corresponding to the steps of the above-mentioned technical solution, and executes the steps in the above-mentioned polymer flooding reservoir production prediction method based on a dual attention mechanism when running.

[0066] The developed system (software) computer program is stored on a computer-readable storage medium, and the computer program is configured to implement the steps of the above-described polymer flooding reservoir production prediction method based on a dual attention mechanism when called by a processor. In other words, the invention is materialized on a carrier, becoming a computer program product.

[0067] Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, application-specific integrated circuits (ASICs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0068] The computational programs (also referred to as programs, software, software applications, or code) of this invention include machine instructions of a programmable processor and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device PLD) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0069] The beneficial effects of the present invention will be described below with reference to specific embodiments.

[0070] Example 1

[0071] Historical development data from 1974 to 2020 was collected from the polymer flooding reservoir field development database, yielding a total of 553 records. Each record includes 6 input features (total number of oil wells, number of oil wells in operation, total number of injection wells, number of injection wells in operation, verified water production, and injection volume) and 1 output feature (monthly production of the polymer flooding reservoir). To adapt to the data characteristics of different polymer flooding reservoir blocks, flexible collection time ranges and data screening criteria were set according to the block development cycle and data record completeness, balancing data coverage and effectiveness.

[0072] like Figure 3As shown, oil production increases continuously from the initial stage of extraction, with a relatively stable production period. After reaching its peak, production begins to decline non-linearly. When oilfield production decreases to a certain level, driven by economic benefits, the polymer flooding reservoir development stage begins. The injection volume continuously increases, and the injection of polymers alters the physical properties of the reservoir, making the crude oil more fluid and thus increasing well production. During this process, the number of water wells (i.e., injection wells) continuously increases, especially during the polymer flooding stage. Simultaneously, new wells are continuously generated during the ongoing extraction process, leading to a continuous increase in the number of oil wells. However, after the mid-stage of extraction, the formation reserves decrease, and the number of oil wells also shows a downward trend. These characteristics have a crucial impact on the production output of this polymer flooding oilfield block.

[0073] For missing values ​​in the data, imputation is performed using the mean of data with the same characteristics from the adjacent three months (when the number of missing features in a single record is ≤2) or the record is directly removed (when the number of missing features in a single record is >2). For outliers, identification is based on the 3σ principle (exceeding the range of "feature mean ± 3 × feature standard deviation"), and replacement is performed using linear interpolation of data from the preceding and following six months. After preprocessing, the correlation between input and output features is screened based on the Pearson correlation coefficient formula, and the results are as follows. Figure 4 As shown, features with an absolute correlation coefficient greater than 0.3 are retained, and the feature data is finally obtained.

[0074] Feature data was divided into training and validation sets in a 9:1 ratio. The CNN-BiLSTM-DAM fusion model was trained on the training set, and its performance was tested on the validation set. The training epochs were set to 90, and the batch size was set to 30 samples per batch. RMSE, MAPE, and R² were used as evaluation metrics to ensure the model achieved optimal performance in both error control and data fit. The training completion standard was defined as "validation set R² value ≥ 0.97, MAPE ≤ 0.09". At this point, the model's prediction error was significantly reduced, accurately capturing the production change trend in polymer flooding reservoirs during the non-polymer injection phase. When inputting prediction data, the data format was consistent with the training data. If the data dimensions did not match, adjustments were made through data completion and feature matching to avoid abnormal model input. If new development data for polymer flooding reservoirs were obtained, the new data could be added to the historical dataset for repeated training, enabling model updates and iterations to adapt to the dynamic changes in reservoir development.

[0075] like Figure 5As shown, the loss metric for training CNN-BiLSTM-DAM models with different Epochs values ​​is analyzed. When Epochs=70, the model has fewer training epochs, resulting in significant oscillations in the loss function and exhibiting instability. Setting Epochs to 80 and 90 increases the number of training epochs and reduces oscillations. The figure shows that the model with 90 training epochs achieves better loss fitting. Finally, while Epochs=100 improves the model's fitting, overfitting occurs in the later stages. Furthermore, due to the excessive number of training epochs, the model exhibits oscillations in the later stages, indicating overlearning of the training set and incorporating noise and random fluctuations from the data, thus demonstrating model instability.

[0076] A model ablation comparison method was adopted. First, the performance prediction of basic RNN, LSTM, and BiLSTM models was performed. Then, a Convolutional Neural Network (CNN) model was added to enhance the feature extraction performance of the time series model, and training was conducted. Next, an attention mechanism was added to the CNN-enhanced model, and training continued. Finally, a dual attention mechanism was added to the better-performing model, and the main network structure of the model was continuously adjusted. The comparative analysis of the results of each model is shown in Table 1.

[0077] Table 1 Comparison of Ablation Experiment Prediction Results

[0078]

[0079] As shown in Table 1, the models based on the original RNN, LSTM, and BiLSTM have a maximum prediction error of over 25,000 tons, a root mean square error of over 7,700 tons, and a percentage error greater than 30%, indicating a generally poor model fit. In contrast, the LSTM and BiLSTM models... The values ​​are all greater than 0.8. Compared to the advantages of Recurrent Neural Networks (RNN) models in processing long-term time-series data, the gating method enhances the prediction performance of the model based on long-term time-series data. After adding a Convolutional Neural Network (CNN) to the model, the maximum error of the RNN, LSTM, and BiLSTM models all decreased, but the decrease was not significant. This may be because some noise and random fluctuations in the data after enhanced feature extraction were also learned, resulting in little change in extreme value prediction. However, the root mean square error and root mean square percentage error both decreased significantly, with CNN-BiLSTM reaching below 7000. This indicates that the overall error of the model decreased through convolutional feature extraction, and the model's fitting effect was better than the original model. After adding a single-layer attention mechanism, the root mean square error and maximum error decreased significantly, and the model's fitting effect was better. The value is higher than 0.9, while the error is reduced to around 6000 tons, and the percentage error is below 0.14. This demonstrates the improvement in model fitting effect brought about by the attention mechanism. Due to the precise weight allocation in the time series model, the model performance continuously improves. Finally, introducing the dual attention mechanism into the CNN-BiLSTM model significantly reduces the maximum error value, indicating that the addition of the attention mechanism weakens the extraction of some noise and random fluctuations in the data. The prediction error of special values ​​is significantly reduced, with the root mean square error reaching 5218 and the root mean square percentage error reaching 0.09. Figure 6 As shown, the model fits well. The value reached 0.97, indicating that the CNN-BiLSTM-DAM model achieves high prediction accuracy with a small root mean square error, demonstrating excellent fitting performance and meeting the requirements for practical applications.

[0080] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A method for predicting production in polymer flooding reservoirs based on a dual attention mechanism, characterized in that: Includes the following steps: S1. Collect field development data of polymer flooding reservoirs, clean the data, and screen and optimize features through relevant analysis to obtain an optimized feature set; S2. Construct a CNN-BiLSTM-DAM fusion model, which includes: a multivariate temporal feature input layer, a convolutional layer, a first attention mechanism layer, a first BiLSTM layer, a second attention mechanism layer, a second BiLSTM layer, a Dropout layer, and an output layer. The convolutional layer is used to extract local features of the input data. The first attention mechanism layer is used to assign dynamic weights to the local features at each time step. The first BiLSTM layer is used to capture the forward and backward dependencies of the weighted temporal data through a bidirectional recurrent structure. The second attention mechanism layer is used to adjust the weights of the temporal features output by the first BiLSTM layer, focusing on temporal information that has a significant impact on the prediction results. The second BiLSTM layer is used to capture global temporal dependencies. S3. Input the optimized feature set into the trained CNN-BiLSTM-DAM fusion model to predict the production of polymer flooding reservoirs.

2. The method for predicting polymer flooding reservoir production based on a dual attention mechanism according to claim 1, characterized in that: The convolutional layer uses one-dimensional convolution (Conv1D), with 64 filters, a kernel size of 1, an activation function of 'relu', and a padding mode of 'same'.

3. The method for predicting polymer flooding reservoir production based on a dual attention mechanism according to claim 2, characterized in that: The number of hidden units in the BiLSTM layer is BiLSTM_UNITS=64.

4. The method for predicting polymer flooding reservoir production based on a dual attention mechanism according to claim 3, characterized in that: The output layer is a fully connected layer with an output dimension of 1 and an activation function of linear, used to output the predicted production value of polymer flooding reservoirs.

5. The method for predicting polymer flooding reservoir production based on a dual attention mechanism according to claim 4, characterized in that: The feature selection for optimization described in S1 specifically involves using Pearson correlation analysis. The formula for calculating the Pearson correlation coefficient is as follows: in, and Let be the values ​​of the two variables for the i-th observation. and These are the means of the variables.

6. The method for predicting polymer flooding reservoir production based on a dual attention mechanism according to claim 5, characterized in that: The data cleaning described in S1 includes filling in missing values ​​in the data and interpolating outliers in the data.

7. The method for predicting polymer flooding reservoir production based on a dual attention mechanism according to claim 6, characterized in that: During the training of the CNN-BiLSTM-DAM fusion model, the root mean square error (RMSE), root mean square percentage error (MAPE), and coefficient of determination (COP) are used. As a model evaluation index, the calculation formulas are as follows: in, This is the actual value. For predicted values, For actual production, To predict production, Let n be the mean of the actual output, and n be the sample size.

8. A polymer flooding reservoir production prediction system based on a dual attention mechanism, characterized in that: The system has a program module corresponding to the steps of the method described in any one of claims 1 to 7, and executes the steps in the above-described polymer flooding reservoir production prediction method based on dual attention mechanism when running.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program configured to, when invoked by a processor, implement the steps of the polymer flooding reservoir production prediction method based on the dual attention mechanism as described in any one of claims 1 to 7.

Images