A method and device for predicting the production performance of a multi-stage fractured horizontal well

Abstract

The present application discloses a method and device for predicting the production dynamics of a multi-stage fractured horizontal well, which relates to the technical field of production dynamics prediction. The method includes: extracting and normalizing the characteristics of the multi-stage fracture information of the target well to obtain fracture statistical characteristics and fracture spatial characteristics, wherein the spatial characteristics of the multi-stage fracture information are extracted through a convolutional neural network combined with an attention mechanism; and splicing the reservoir geological parameter information, fracture statistical characteristics and fracture spatial characteristics of the target well to obtain a static feature vector and inputting it into a trained production dynamics prediction model to obtain the production dynamics prediction result of the target well. The solution of the present application can quickly and accurately obtain the production data at several time points during the entire life cycle of the target well from the start of production to the final shutdown, and can well overcome the problems of large consumption of computing resources and strong dependence on historical production data of the traditional numerical simulation method.

Description

Technical Field

[0001] The present application relates to the technical field of production dynamic prediction, and particularly to a method and device for predicting the production dynamics of a multi-stage fractured horizontal well. Background Art

[0002] Unconventional oil and gas reservoirs constitute an important energy resource base. In particular, the multi-stage fractured horizontal well technology has become a key technology for developing these reservoirs due to its excellent stimulation effect and high economic efficiency. During the development process, accurately predicting the production dynamics of multi-stage fractured horizontal wells is crucial. This not only provides a scientific basis for well placement, fracturing parameter optimization, and drainage production system formulation, supporting the optimization of the development plan; at the same time, it also provides decision-making support for reserve dynamic adjustment, economic benefit evaluation, and production management. This helps to improve resource recovery rate, reduce development costs, and effectively control economic risks.

[0003] Currently, the methods for predicting the production dynamics of multi-stage fractured horizontal wells in unconventional oil and gas reservoirs mainly include empirical formula method, numerical simulation method, and machine learning method. The empirical formula method uses statistical regression and historical data to construct a simple model. Its advantages are fast calculation speed and easy application, suitable for preliminary estimation. However, its disadvantages are that it is difficult to accurately capture complex non-linear relationships, highly dependent on data, and its applicability is limited by the specific conditions during formula development. Although the numerical simulation method has high prediction accuracy and wide applicability, it requires a large amount of computing resources and professional software support. In contrast, the application of the machine learning method in the field of unconventional oil and gas reservoirs is increasing. However, most current studies mainly predict future production changes based on historical production data and related influencing factors.

[0004] Therefore, solving the problems of large computing resource consumption and strong dependence on historical production data of the numerical simulation method is of great significance for achieving rapid and accurate prediction of the production dynamics of multi-stage fractured horizontal wells in unconventional oil and gas reservoirs throughout the entire life cycle from production input to final production cessation. Summary of the Invention

[0005] The purpose of the present application is to provide a method and device for predicting the production dynamics of a multi-stage fractured horizontal well, which can quickly and accurately predict the production dynamic data of the target well.

[0006] To achieve the above purpose, the present application provides the following solutions:

[0007] In a first aspect, the present application provides a method for predicting the production dynamics of a multi-stage fractured horizontal well, including:

[0008] Obtaining the reservoir geological parameter information and multi-stage fracture information of the target well.

[0009] Feature extraction and normalization processing are performed on the multi-segment fracture information of the target well, and fracture statistical features and fracture spatial features are extracted; the spatial features of the multi-segment fracture information are extracted through a convolutional neural network combined with an attention mechanism.

[0010] The reservoir geological parameter information, the fracture statistical features, and the fracture spatial features of the target well are feature-stitched to obtain a static feature vector.

[0011] The static feature vector is input into the trained production dynamic prediction model to obtain the production dynamic prediction result of the target well; the production dynamic prediction model is a model constructed based on the BiLSTM time series prediction model; the production dynamic prediction result is the production data at several time points during the entire life cycle of the target well from the start of production to the final shutdown.

[0012] Optionally, feature extraction and normalization processing are performed on the multi-segment fracture information of the target well, and fracture statistical features and fracture spatial features are extracted, specifically including:

[0013] Each row of data of the multi-segment fracture information of the target well is parsed to extract the initial fracture statistical features.

[0014] The multi-segment fracture information of the target well is converted into a picture form, and the initial fracture spatial features are extracted through a convolutional neural network combined with an attention mechanism.

[0015] Normalization processing is respectively performed on the initial fracture statistical features and the initial fracture spatial features to obtain fracture statistical features and fracture spatial features.

[0016] Optionally, it further includes: constructing an initial production dynamic prediction model, and training it according to the reservoir geological parameters and multi-segment fracture information of several well groups, and optimizing the initial production dynamic prediction model to obtain the trained production dynamic prediction model; specifically including:

[0017] Based on the BiLSTM time series prediction model, an initial production dynamic prediction model is constructed.

[0018] The reservoir geological parameters and multi-segment fracture information of several historical well groups are obtained.

[0019] For any historical well group, reservoir numerical simulation is performed according to the reservoir geological parameters and multi-segment fracture information of the historical well group to obtain the simulated production dynamic data of the historical well group.

[0020] Feature extraction and normalization processing are performed on the multi-segment fracture information of the historical well group, and fracture statistical features and fracture spatial features of the historical well group are extracted.

[0021] Taking the reservoir geological parameters, fracture statistical characteristics, and fracture spatial characteristics of historical well groups as the input of the initial production dynamic prediction model, and taking the simulated production dynamic data of historical well groups as the target output of the initial production dynamic prediction model, training and optimizing the initial production dynamic prediction model to obtain a trained production dynamic prediction model.

[0022] Optionally, performing reservoir numerical simulation based on the reservoir geological parameters and multi-stage fracture information of the historical well groups to obtain the simulated production dynamic data of the historical well groups, specifically including:

[0023] For any historical well group, designing a reservoir geological parameter plan according to the reservoir geological parameters of the historical well group, and designing a fracture design plan according to the multi-stage fracture information of the historical well group.

[0024] Establishing a regular grid-based numerical simulation model based on the multi-stage fracture information of the historical well group.

[0025] According to the reservoir geological parameter plan and fracture design plan of the historical well group, calling the numerical simulation model to perform reservoir simulation to obtain the simulated production dynamic data of the historical well group.

[0026] Optionally, it further includes: keeping the reservoir geological parameters, fracture statistical characteristics, fracture spatial characteristics, and simulated production dynamic data of several historical well groups corresponding, and dividing them into a training set and a test set; using the data in the training set to train and optimize the initial production dynamic prediction model, and using the data in the test set to test the performance of the trained initial production dynamic prediction model.

[0027] Optionally, the production dynamic prediction model includes two layers of bidirectional LSTM. The number of hidden units in the first layer of bidirectional LSTM is 208, the number of hidden units in the second layer of bidirectional LSTM is 176, the learning rate is 0.0004, and the batch_size is 12.

[0028] In a second aspect, the present application provides a multi-stage fractured horizontal well production dynamic prediction device, including:

[0029] A target well parameter acquisition module for acquiring the reservoir geological parameter information and multi-stage fracture information of the target well.

[0030] A fracture parameter feature extraction module for performing feature extraction and normalization processing on the multi-stage fracture information of the target well, extracting fracture statistical characteristics and fracture spatial characteristics; fully extracting the spatial characteristics of the multi-stage fracture information through a convolutional neural network combined with an attention mechanism.

[0031] The static feature vector splicing module is used to splice the reservoir geological parameter information, the fracture statistical features, and the fracture spatial features of the target well to obtain a static feature vector.

[0032] The production dynamic prediction module is used to input the static feature vector into the trained production dynamic prediction model to obtain the production dynamic prediction result of the target well; the production dynamic prediction model is a model constructed based on the BiLSTM time series prediction model; the production dynamic prediction result is the production data at several time points during the entire life cycle of the target well from the start of production to the final shutdown.

[0033] According to the specific embodiments provided by the present application, the present application discloses the following technical effects:

[0034] The present application provides a method and device for predicting the production dynamics of a multi-stage fractured horizontal well. In this method, feature extraction and normalization processing are performed on the multi-stage fracture information of the target well, and fracture statistical features and fracture spatial features are extracted. Among them, the spatial features of the multi-stage fracture information are extracted through a convolutional neural network combined with an attention mechanism, and the fracture distribution and geometric characteristics are represented by the spatial features and statistical features of the multi-stage fracture information, so as to solve the problem that the data lengths of the multi-stage fracture information of different horizontal wells are inconsistent due to different numbers of fractures; and the reservoir geological parameter information, fracture statistical features, and fracture spatial features of the target well are spliced to obtain a static feature vector and input it into the trained production dynamic prediction model to obtain the production dynamic prediction result of the target well; the production dynamic prediction model used is a model constructed based on the BiLSTM time series prediction model, which can quickly and accurately obtain the production data at several time points during the entire life cycle of the target well from the start of production to the final shutdown, and can well overcome the problems of large consumption of computing resources and strong dependence on historical production data in the traditional numerical simulation method. It is of great significance for realizing the rapid and accurate prediction of the production dynamics during the entire life cycle of a multi-stage fractured horizontal well in unconventional oil and gas reservoirs from the start of production to the final shutdown. Description of the Drawings

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0036] Figure 1 It is a flowchart of a method for predicting the production dynamics of a multi-stage fractured horizontal well provided by an embodiment of the present application.

[0037] Figure 2Schematic diagram of multi-stage fracture design for a typical horizontal well in a multi-stage fractured horizontal well production dynamic prediction method provided by an embodiment of the present application.

[0038] Figure 3 Flowchart of step A2 in a multi-stage fractured horizontal well production dynamic prediction method provided by an embodiment of the present application.

[0039] Figure 4 Schematic diagram of the structure of traditional LSTM.

[0040] Figure 5 Schematic diagram of the structure of BiLSTM adopted by the present application.

[0041] Figure 6 Schematic diagram of the visualization of the production dynamics of multiple horizontal wells in a multi-stage fractured horizontal well production dynamic prediction method provided by an embodiment of the present application.

[0042] Figure 7 Technical roadmap of a multi-stage fractured horizontal well production dynamic prediction method provided by an embodiment of the present application.

[0043] Figure 8 Schematic diagram of the functional modules of a multi-stage fractured horizontal well production dynamic prediction device provided by an embodiment of the present application.

[0044] Figure 9 Schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0045] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0046] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the drawings and specific implementation manners.

[0047] A multi-stage fractured horizontal well production dynamic prediction method provided by an embodiment of the present application, in an exemplary embodiment, as Figure 1 shown, includes the following steps:

[0048] A1. Obtain the reservoir geological parameter information and multi - stage fracture information of the target well. Specifically, in this embodiment, the reservoir geological parameter information includes key parameters such as porosity, permeability, thickness, Young's modulus, relative permeability, reservoir pressure, ratio of desorption pressure to reservoir pressure, Langmuir volume, Langmuir pressure, fracture permeability, well - controlled area, etc. Subsequently, according to the reservoir geological parameter information of different types of unconventional oil and gas reservoirs (assuming predicting the production dynamics of a coalbed methane reservoir, only the reservoir geological parameters related to the coalbed methane reservoir can be used, and they cannot be confused with the reservoir geological parameters of other unconventional oil and gas reservoirs), and combined with the number of wells participating in the training, a reasonable reservoir geological parameter scheme is formulated to ensure the adaptability and accuracy of the prediction model.

[0049] The multi - stage fracture information includes the ratio of the well - controlled area to the well - controlled reservoir (the well - controlled area is also called the effective area), the number of fractures, fracture spacing, and fracture length. The specific design range is adjusted according to the reservoir type and geological conditions to optimize the fracture distribution and conductivity. For example, the number of fractures is set according to the reservoir permeability and stress field characteristics, the fracture spacing is flexibly adjusted in combination with the reservoir stimulation requirements, and the fracture length is restricted by the width of the well - controlled area (the effective area represents its length along the horizontal direction and its width along the vertical direction; the fracture stimulation area calculates the fracture spacing along the horizontal direction and the fracture length along the vertical direction) to ensure the effectiveness of the fracturing stimulation. Specifically, the Latin hypercube sampling method can be used to design the positions of multi - stage fractures in horizontal wells, forming a fracture design scheme applicable to different types of unconventional oil and gas reservoirs. A schematic diagram of the multi - stage fracture design of a typical horizontal well is as Figure 2 shown.

[0050] A2. Extract features and perform normalization processing on the multi - stage fracture information of the target well, and obtain fracture statistical features and fracture spatial features; extract the spatial features of the multi - stage fracture information through a convolutional neural network combined with an attention mechanism. In this embodiment, as Figure 3 shown, step A2 specifically includes the following steps:

[0051] A21. Analyze each row of data of the multi - stage fracture information of the target well to obtain the initial fracture statistical features.

[0052] A22. Convert the multi - stage fracture information of the target well into a picture form, and extract the initial fracture spatial features through a convolutional neural network combined with an attention mechanism.

[0053] A23. Perform normalization processing on the initial fracture statistical features and the initial fracture spatial features respectively to obtain the fracture statistical features and the fracture spatial features.

[0054] A3. Concatenate the reservoir geological parameter information, fracture statistical features, and fracture spatial features of the target well to obtain a static feature vector.

[0055] A4. Input the static feature vector into the trained production dynamic prediction model to obtain the production dynamic prediction result of the target well; the production dynamic prediction model is a model constructed based on the BiLSTM time series prediction model; the production dynamic prediction result is the production data at several time points during the entire life cycle of the target well from the start of production to the final shutdown.

[0056] In another exemplary embodiment of the present application, the production dynamic prediction method for multi-stage fractured horizontal wells further includes: B1. Construct an initial production dynamic prediction model, and train it according to the reservoir geological parameters and multi-stage fracture information of several well groups, and optimize the initial production dynamic prediction model to obtain the trained production dynamic prediction model; specifically including:

[0057] B11. Based on the BiLSTM time series prediction model, construct an initial production dynamic prediction model. Since the predicted production dynamics have time series characteristics, BiLSTM is selected to predict the production dynamics of unconventional oil and gas. At the same time, compared with general time series models such as LSTM (the structural schematic diagram of LSTM is as Figure 4 shown), BiLSTM introduces a bidirectional structure in the network, has stronger time series feature extraction ability, and is more suitable for long-term prediction. The structural schematic diagram of BiLSTM adopted in the present application is as Figure 5 shown.

[0058] B12. Obtain the reservoir geological parameters and multi-stage fracture information of several historical well groups. In this embodiment, 1000 groups of reservoir geological parameters and multi-stage fracture information are collected to form a reservoir geological parameter plan and a fracture design plan respectively.

[0059] B13. For any historical well group, perform reservoir numerical simulation according to the reservoir geological parameters and multi-stage fracture information of the historical well group to obtain the simulated production dynamic data of the historical well group. After obtaining the reservoir geological parameter plan and the fracture design plan, use numerical simulation software MATLAB and CMG to simulate the production process of the multi-stage fractured horizontal well for 20 years, and finally extract the production dynamic data including the daily gas production and the cumulative gas production. For each horizontal well, with a step of 1 month, a total of 240 time steps of corresponding production data are obtained. In this embodiment, step B13 specifically includes:

[0060] B131. For any historical well group, design a reservoir geological parameter plan according to the reservoir geological parameters of the historical well group, and design a fracture design plan according to the multi-stage fracture information of the historical well group.

[0061] B132. Establish a numerical simulation model based on regular grids using the multi - segment fracture information of historical well groups. In this embodiment, a numerical simulation model based on regular grids is established according to the multi - segment fracture information of historical well groups, aiming to improve the calculation accuracy of the numerical simulation process and reasonably characterize the fracture distribution. The model is divided into several grid units both in the horizontal and vertical directions, and the size of each grid unit is set according to the actual geological conditions to balance the calculation efficiency and simulation accuracy. This grid division scheme ensures the accurate description of multi - segment fractures and provides a reliable data basis for subsequent production performance prediction.

[0062] B133. According to the reservoir geological parameter scheme and fracture design scheme of historical well groups, call the numerical simulation model to conduct reservoir simulation and obtain the simulated production performance data of historical well groups. Package the reservoir geological parameter scheme and fracture design scheme into files (such as CSV or MAT files) respectively, and use the numerical calculation tool MATLAB and the reservoir numerical simulation software CMG for reservoir simulation. For example: First, write a script file for performing reservoir simulation using MATLAB. Read the reservoir geological parameter scheme and fracture design scheme in this script file, and then implement the numerical simulation model based on regular grids in this script file. Subsequently, call the GEM tool in CMG in the script file to batch implement the reservoir simulation process. Finally, execute the script file, extract the required production performance data (such as gas production, oil production, etc.) from the generated simulation results, and save them in an Excel file.

[0063] B14. Extract features and perform normalization processing on the multi - segment fracture information of historical well groups to obtain the fracture statistical features and fracture spatial features of historical well groups.

[0064] After step B13 processing, the shape of the reservoir geological parameters is (1000, 15), indicating 1000 samples, and each sample includes 15 geological parameters. The shape of the statistical features extracted from multi - segment fractures is (1000, 6), and each sample includes 6 types of information: the proportion of the effective fracture area, the number of fractures, the mean and variance of fracture lengths, and the mean and variance of fracture spacings. Then convert the multi - segment fracture information into a picture with a shape of 128 * 256, input it into the CNN - AM model for spatial feature extraction, and the shape of the spatial features after screening is (1000, 20). Finally, splice the reservoir geological parameters, the statistical features and spatial features of multi - segment fracture information together to form a static feature vector with a shape of (1000, 41).

[0065] For reservoir geological parameters, whether to perform data normalization should be determined in combination with the distribution of the actually collected reservoir geological parameters. According to observations, the values of all the collected data are between 0 and 1. Therefore, data normalization is not required. At the same time, through inspection, it is found that there are no missing values or outliers, and the lengths of the reservoir geological parameters of each horizontal well are the same. For the multi-segment fracture information, since the number of fractures in each horizontal well is not exactly the same, the data lengths of their fracture information are inconsistent.

[0066] Specifically, for the multi-segment fracture information with inconsistent input data lengths, by parsing each row of the input data, statistical features such as the proportion of the well-controlled area, the number of fractures, the length, and the spacing are extracted. First, calculate the ratio of the well-controlled area to the well-controlled reservoir according to the boundary values of the given area; then, parse the fracture center position and half-length information, calculate the length of each fracture, the spacing between adjacent fractures, and further calculate their mean and variance; finally, summarize the calculation results and perform min-max normalization on each column of eigenvalue to standardize the data for subsequent analysis.

[0067] The fracture statistical features include the proportion of the well-controlled area, the number of fractures, the mean and variance of the fracture length, and the mean and variance of the fracture spacing. The proportion of the well-controlled area is the ratio of the well-controlled area to the well-controlled reservoir; the number of fractures will be processed by min-max normalization, as shown in formula (1):

[0068] (1).

[0069] Where, x is the number of fractures in a certain horizontal well, min is the minimum value of the number of fractures among all these horizontal wells, max is the maximum value of the number of fractures, x' is the number of fractures after normalization, and the result is between [0, 1].

[0070] The calculation of the mean and variance of the fracture length and spacing is shown in formulas (2) to (5):

[0071] (2).

[0072] (3).

[0073] (4).

[0074] (5).

[0075] Where, μ L is the mean of the fracture length in a certain horizontal well, is the variance of the fracture length in this well,N L is the number of fractures in the well, x L,i is the length of one of the fractures in the well, μ D is the mean of the spacings in a horizontal well, is the variance of the spacings in the well, N D is the number of spacings in the well, x D,i is the spacing between two adjacent fractures in the well.

[0076] After that, the multi - segment fracture information is converted into a picture format, and CNN and AM are used to extract its spatial features. CNN and AM are mainly used to extract the spatial features of multi - segment fracture information. Using spatial and channel attention mechanisms can enhance the attention to fractures. Through the convolutional operation of CNN, the local and global spatial features of fractures are gradually extracted. At the same time, combined with channel attention and spatial attention mechanisms, they focus on important channels and significant spatial regions in the feature map respectively, highlighting key features and suppressing redundant information, so as to achieve more efficient and accurate feature extraction of fracture distribution, morphology and characteristics. After the spatial features of multi - segment fractures are extracted, it is necessary to screen the blank column data caused by the sparse distribution of fractures, so as to reduce the data dimension and improve the accuracy of subsequent prediction. Finally, the maximum - minimum normalization processing is also carried out on the feature values of each column of spatial features.

[0077] Specifically in this embodiment, the constructed CNN - AM model architecture includes an input layer, a convolutional and pooling layer, a global average pooling layer, a fully - connected layer and an output layer. Among them, the input layer receives two - dimensional input data with a shape of (128, 256, 1). The convolutional and pooling layer contains 3 layers: the first - layer convolution extracts low - level features, uses 32 3x3 convolutional kernels, and the activation function is ReLU. Subsequently, the feature map resolution is reduced by 2x2 max - pooling; the second - layer convolution uses 64 3x3 convolutional kernels and introduces a spatial attention module to enhance the model's focus on key regions. Subsequently, the dimension is reduced again by 2x2 max - pooling. The third - layer convolution uses 128 3x3 convolutional kernels and introduces a channel attention module to enhance the feature selection ability. The global average pooling layer: converts the feature map into a fixed - size feature vector, reducing the number of parameters. The fully - connected layer: generates the final feature vector with a dimension of num_features = 32, and the activation function is ReLU. The output layer outputs a 32 - dimensional feature vector.

[0078] For the simulated production dynamic data, through such as Figure 6It is found from the visualization diagram of the production performance of the multi-lateral horizontal wells shown that there is a large numerical difference in the cumulative production (cumulative gas production or cumulative oil production) between different horizontal wells. Therefore, during the training of the model, the min-max normalization process is also performed on it to enhance the stability of the model and improve the model training efficiency.

[0079] As an alternative implementation, to ensure a better production performance prediction model for training, the method further includes: keeping the reservoir geological parameters, fracture statistical characteristics, fracture spatial characteristics, and simulated production performance data of several historical well groups corresponding, and dividing them into a training set and a test set; the training set is used for the training of the model, and the test set is used to verify the prediction performance of the model, and the repeatability of data division is ensured by fixing the random seed.

[0080] B15. Use the reservoir geological parameters, fracture statistical characteristics, and fracture spatial characteristics of the historical well group as the input of the initial production performance prediction model, and use the simulated production performance data of the historical well group as the target output of the initial production performance prediction model, and perform training and optimization on the initial production performance prediction model to obtain a trained production performance prediction model.

[0081] The above model training method of this application is implemented based on TensorFlow in Python. The static feature vectors and production performance data input into the BiLSTM model need to be reshaped into three-dimensional tensors with a shape of (number of samples, time steps, feature dimension) to adapt to the input requirements of the LSTM before the model can be trained. The BiLSTM model architecture constructed in this embodiment mainly consists of three parts: First, the first layer of the model is a bidirectional LSTM (Bidirectional LSTM), which consists of a forward and a backward LSTM network. Each LSTM cell contains 208 hidden units and uses the ReLU activation function to capture the non-linear features in the input data. By setting return_sequences=True for this layer, the hidden state at each time step is output, providing complete time series information for the subsequent LSTM layer. The second layer is another bidirectional LSTM, which contains 176 hidden units but does not return the sequence (return_sequences=False). It extracts global time features and further compresses the time dimension information. Finally, the model outputs the time step prediction values of the target sequence through a fully connected layer (Dense), and the output dimension is the same as that of the target sequence.

[0082] By capturing the forward and backward temporal dependencies in time series data simultaneously, the BiLSTM model can comprehensively extract context information. It combines the gate mechanism of LSTM to effectively solve the problems of gradient vanishing and gradient explosion in ordinary recurrent neural networks. At the same time, it extracts historical and future information through the forward and backward LSTM layers respectively, and then generates a more accurate feature representation after fusion.

[0083] The calculation of the forward LSTM is the same as that of the standard LSTM, starting from the first time step of the sequence. The following calculation formulas (6) to (11) are the forget gate, input gate, candidate memory unit, memory unit update, output gate, and hidden state from top to bottom in turn.

[0084] (6).

[0085] (7).

[0086] (8).

[0087] (9).

[0088] (10).

[0089] (11).

[0090] Among them, the superscript ( f ) represents the calculation of the forward LSTM, x t is the cell input vector, , , respectively represent the output of the forget gate, input gate, and output gate of the forward LSTM at time t ; , , respectively represent the memory cell state, candidate memory cell, and hidden state of the forward LSTM at time t; , , , respectively represent the weight matrices of the forget gate, input gate, memory cell, and output gate of the forward LSTM; , , , respectively represent the bias terms of the forget gate, input gate, memory cell, and output gate of the forward LSTM; is the Sigmoid activation function, with a range between 0 and 1; tanh is the hyperbolic tangent activation function, with a range between -1 and 1.

[0091] The calculation of the reverse LSTM starts from the last time step of the sequence and performs backpropagation. The calculation formulas are as shown in Equations (12) to (18):

[0092] (12).

[0093] (13).

[0094] (14).

[0095] (15).

[0096] (16).

[0097] (17).

[0098] where the superscript ( b ) represents the calculation of the reverse LSTM, and the meanings of the other symbols are the same as those of the forward LSTM.

[0099] The final output of the BiLSTM is the concatenation (or weighted average) of the hidden states of the forward and reverse LSTMs. Assuming the outputs of the forward and reverse LSTMs are and , the final output is:[[]]

[0100] (18).

[0101] The model is trained using the Adam optimizer (setting the learning rate) and the mean squared error (MSE) as the loss function to compile the model. In the training phase, the model is trained with a batch size of 8, iterated at most 300 times, and 10% of the training data is used as the validation set. To prevent overfitting, an early stopping mechanism is set, and the training is terminated early when the validation loss does not improve for 20 consecutive epochs.

[0102] The model optimization process uses Keras Tuner for hyperparameter tuning. First, a hyperparameter model building function is defined, which includes the number of BiLSTM layers, the number of units in each BiLSTM layer, and the hyperparameter range of the learning rate. Hyperparameter search is performed using the random search method, with the maximum number of trials set to 20, and each trial is executed twice to find the best hyperparameter combination. During the search process, an early stopping strategy is used to prevent overfitting, and the model is evaluated on the validation set. Finally, the final model is built and trained using the best hyperparameters, and its performance is evaluated on the test set. In this embodiment, the production dynamic prediction model includes two layers of bidirectional LSTM. The number of hidden units in the first layer of bidirectional LSTM is 208, the number of hidden units in the second layer of bidirectional LSTM is 176, the learning rate is 0.0004, and the batch_size is 12.

[0103] The prediction and evaluation process of the model first makes predictions on the training set and the test set through the final model. Then, a custom evaluation function is used to calculate and return the model evaluation metrics for the training set and the test set, including the correlation coefficient ( R 2 ), the mean squared error ( MSE ), the root mean squared error ( RMSE ), and the mean absolute error ( MAE ), as shown in formulas (19) - (22) respectively. Finally, the evaluation results of the training set and the test set are output, and some predicted images are displayed to help further analyze the model performance.

[0104] (19).

[0105] (20).

[0106] (21).

[0107] (22).

[0108] Among them, y i is the true value in the production data, is the predicted value in the production data, n is the number of time steps of the production dynamics in the multi-stage fractured horizontal well. Among these evaluation metrics, R 2 the closer the value of MSE , RMSE and MAE is to 1, and the smaller the values of Figure 7As shown, it includes five steps: data acquisition, data preprocessing, data preparation and partitioning, model construction and training, and model optimization and prediction. After the above training and optimization process, the prediction effects of the trained model on the training set and the test set are shown in Table 1.

[0109] Table 1 Prediction Result Evaluation

[0110]

[0111] The above experimental cases prove that the method proposed in the above embodiments of the present application performs well in predicting the production dynamics of multi-stage fractured horizontal wells in coalbed methane reservoirs by using static reservoir geological parameters and multi-stage fracture information. It not only has high accuracy, but also has a faster prediction speed compared with empirical formulas and numerical simulation methods.

[0112] Based on the same inventive concept, the embodiments of the present application also provide a device for implementing the above-mentioned method for predicting the production dynamics of multi-stage fractured horizontal wells. The solution provided by this device to solve the problem is similar to the solution described in the above method. In an exemplary embodiment, as Figure 8 shown, a device for predicting the production dynamics of multi-stage fractured horizontal wells is provided, including:

[0113] A target well parameter acquisition module, configured to acquire reservoir geological parameter information and multi-stage fracture information of a target well.

[0114] A fracture parameter feature extraction module, configured to perform feature extraction and normalization processing on the multi-stage fracture information of the target well, and extract fracture statistical features and fracture spatial features; the spatial features of the multi-stage fracture information are fully extracted through a convolutional neural network combined with an attention mechanism.

[0115] A static feature vector splicing module, configured to splice the reservoir geological parameter information, the fracture statistical features, and the fracture spatial features of the target well to obtain a static feature vector.

[0116] A production dynamics prediction module, configured to input the static feature vector into a trained production dynamics prediction model to obtain a production dynamics prediction result of the target well; the production dynamics prediction model is a model constructed based on a BiLSTM time series prediction model; the production dynamics prediction result is the production data at several time points during the entire life cycle of the target well from the start of production to the final shutdown.

[0117] Of course, Figure 8 the architecture shown is only exemplary. When implementing different functions, one or at least two components in the device shown may be omitted according to actual needs. Figure 8

[0118] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as shown in Figure 9 . The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the multi-stage fracturing horizontal well production dynamic prediction method provided in the above embodiment can be implemented.

[0119] Those skilled in the art can understand that Figure 9 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0120] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0121] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0122] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0123] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0124] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0125] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0126] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0127] In this article, specific examples are used to elaborate on the principles and implementation manners of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for predicting the production dynamics of a multi-stage fractured horizontal well, characterized in that, Including: Obtain the reservoir geological parameter information and multi - stage fracture information of the target well; Extract features and perform normalization processing on the multi - stage fracture information of the target well, and obtain fracture statistical features and fracture spatial features; Extract the spatial features of the multi - stage fracture information through a convolutional neural network combined with an attention mechanism; Perform feature splicing on the reservoir geological parameter information, the fracture statistical features, and the fracture spatial features of the target well to obtain a static feature vector; Input the static feature vector into the trained production dynamic prediction model to obtain the production dynamic prediction result of the target well; the production dynamic prediction model is a model constructed based on the BiLSTM time - series prediction model; The production dynamic prediction result is the production data at several time points during the entire life cycle of the target well from the start of production to the final shutdown; It also includes: constructing an initial production dynamic prediction model, and training it according to the reservoir geological parameters and multi - stage fracture information of several well groups, and optimizing the initial production dynamic prediction model to obtain the trained production dynamic prediction model; specifically including: Based on the BiLSTM time - series prediction model, construct an initial production dynamic prediction model; Obtain the reservoir geological parameters and multi - stage fracture information of several historical well groups; For any historical well group, perform reservoir numerical simulation according to the reservoir geological parameters and multi - stage fracture information of the historical well group to obtain the simulated production dynamic data of the historical well group; Extract features and perform normalization processing on the multi - stage fracture information of the historical well group, and obtain the fracture statistical features and fracture spatial features of the historical well group; Use the reservoir geological parameters, fracture statistical features, and fracture spatial features of the historical well group as the input of the initial production dynamic prediction model, and use the simulated production dynamic data of the historical well group as the target output of the initial production dynamic prediction model to train and optimize the initial production dynamic prediction model to obtain the trained production dynamic prediction model; the production dynamic prediction model contains two - layer bidirectional LSTM, the number of hidden units in the first - layer bidirectional LSTM is 208, the number of hidden units in the second - layer bidirectional LSTM is 176, the learning rate is 0.0004, and the batch_size is 12.

2. The method for predicting the production performance of a multi-stage fractured horizontal well according to claim 1, wherein Extract features and perform normalization processing on the multi - stage fracture information of the target well, and obtain fracture statistical features and fracture spatial features, specifically including: Parse each row of data of the multi - stage fracture information of the target well to obtain initial fracture statistical features; Convert the multi - stage fracture information of the target well into a picture form, and extract initial fracture spatial features through a convolutional neural network combined with an attention mechanism; Perform normalization processing on the initial fracture statistical features and the initial fracture spatial features respectively to obtain fracture statistical features and fracture spatial features.

3. The multi-stage fracturing horizontal well production dynamic prediction method according to claim 1, wherein Perform reservoir numerical simulation according to the reservoir geological parameters and multi - stage fracture information of the historical well group to obtain the simulated production dynamic data of the historical well group, specifically including: For any historical well group, a reservoir geological parameter scheme is designed based on the reservoir geological parameters of the historical well group, and a fracture design scheme is designed based on the multi - stage fracture information of the historical well group; Based on the multi - stage fracture information of the historical well group, a numerical simulation model based on a regular grid is established; According to the reservoir geological parameter scheme and the fracture design scheme of the historical well group, the numerical simulation model is called for reservoir simulation to obtain the simulated production dynamic data of the historical well group.

4. The multi-stage fracturing horizontal well production performance prediction method according to claim 1, wherein It further includes: Keeping the reservoir geological parameters, fracture statistical characteristics, fracture spatial characteristics, and simulated production dynamic data of several historical well groups corresponding to each other, and dividing them into a training set and a test set; using the data in the training set to train and optimize the initial production dynamic prediction model, and using the data in the test set to test the performance of the trained initial production dynamic prediction model.

5. A device for dynamically predicting the production of a multi-stage fractured horizontal well, characterized in that, For implementing the multi - stage fracturing horizontal well production dynamic prediction method according to any one of claims 1 - 4, the multi - stage fracturing horizontal well production dynamic prediction device includes: A target well parameter acquisition module, configured to acquire the reservoir geological parameter information and multi - stage fracture information of the target well; A fracture parameter feature extraction module, configured to perform feature extraction and normalization processing on the multi - stage fracture information of the target well, and extract fracture statistical characteristics and fracture spatial characteristics; fully extract the spatial characteristics of the multi - stage fracture information through a convolutional neural network combined with an attention mechanism; A static feature vector splicing module, configured to splice the reservoir geological parameter information, the fracture statistical characteristics, and the fracture spatial characteristics of the target well to obtain a static feature vector; A production dynamic prediction module, configured to input the static feature vector into the trained production dynamic prediction model to obtain the production dynamic prediction result of the target well; the production dynamic prediction model is a model constructed based on a BiLSTM time - series prediction model; the production dynamic prediction result is the production data at several time points during the entire life cycle of the target well from the start of production to the final shutdown; the production dynamic prediction model includes two - layer bidirectional LSTM, the number of hidden units in the first - layer bidirectional LSTM is 208, the number of hidden units in the second - layer bidirectional LSTM is 176, the learning rate is 0.0004, and the batch_size is 12.

Images