Prediction method for horizontal well deliverability, storage medium and processor

Abstract

Embodiments of the present application provide a prediction method for horizontal well productivity, a storage medium and a processor. The method comprises: obtaining production data and physical constraint parameters of a horizontal well in a historical time period, the physical constraint parameters including at least one of reservoir properties, logging curves, horizontal well information, fracturing parameters and well control changes; preprocessing the production data and the physical constraint parameters to obtain a preprocessed data set; determining a functional equation of a prediction model through a physical driving neural network basic method; determining hyperparameters of the prediction model through a Bayesian optimization method to determine a model architecture of the prediction model; constructing the prediction model according to the functional equation and the model; dividing the preprocessed data set into multiple groups of data, and sequentially inputting each group of data into the prediction model; for each group of data, outputting a predicted productivity of the horizontal well in a prediction time period through the prediction model; and determining a true predicted productivity of the horizontal well in the prediction time period according to the multiple predicted productivities.

Description

Technical Field

[0001] This application relates to the field of oil and gas field development technology, specifically to a method for predicting the productivity of horizontal wells, a productivity prediction system, a storage medium, and a processor. Background Technology

[0002] Production forecasting is crucial for the economic exploration and development of unconventional resources, particularly in fracturing design, economic evaluation, and decision-making. However, predicting well dynamics remains challenging due to multiphase fluid flows and complex multi-scale fracture networks in tight oil formations, and the underlying physical laws are not fully understood. With the rapid development of computer science and the explosive accumulation of oilfield data, machine learning (ML) has been widely applied in the petroleum industry due to its powerful nonlinear approximation capabilities. This paper employs random forest, adaptive augmentation, support vector, and neural network algorithms, using well information and fracturing enhancement parameters as inputs, to predict the cumulative production capacity for the first year. Formation thickness, depth, perforation count, and logging curves are used as inputs, and the cumulative production capacity for the first three months is used as the output. However, in the above ML model, the dependence between input parameters and target production capacity often depends on the amount and quality of data. Furthermore, the prediction target is the cumulative production capacity over a fixed period, which cannot reflect the dynamic changes in well production capacity over time.

[0003] To estimate dynamic time-series production capacity, recurrent neural networks and their variants—namely, long short-term memory, gated recurrent units, and bidirectional gated recurrent units—are introduced into ML-based methods. These methods can predict future production capacity based on historical production data. However, in practical applications of long-term production capacity forecasting, these pure ML methods suffer from inaccurate predictions due to the increasing amount of data required as the complexity of the neural network increases. Furthermore, when the ML model is fed inputs exceeding the range of training data, the output predictions become inaccurate, failing to provide accurate predictions of oil well production capacity. Summary of the Invention

[0004] The purpose of this application is to provide a method, system, storage medium, and processor for predicting the productivity of horizontal wells.

[0005] To achieve the above objectives, the first aspect of this application provides a method for predicting the productivity of horizontal wells, comprising: Obtain production data and physical constraint parameters of horizontal wells over a historical period. Physical constraint parameters include at least one of reservoir properties, logging curves, horizontal well information, fracturing parameters, and well control changes. The production data and physical constraint parameters are preprocessed to obtain a preprocessed dataset corresponding to the production data and physical constraint parameters; The functional equations of the prediction model are determined using the basic method of physical-driven neural networks; The hyperparameters of the prediction model are determined by using Bayesian optimization methods, thereby determining the model architecture of the prediction model. Construct a prediction model based on the functional equation and the model; The preprocessed dataset is divided into multiple groups of data, and each group of data is input into the prediction model in turn. For each set of data, the predicted production capacity of the horizontal well within the predicted time period is output by the prediction model; The actual predicted production capacity of horizontal wells within the predicted time period is determined based on multiple predicted production capacities.

[0006] In one embodiment, determining the functional equation of the prediction model using the physical-driven neural network basic method includes: determining the physical loss term and the neural network loss term of the functional equation; and determining the functional equation of the prediction model based on the preset neural network loss term and the physical loss term.

[0007] In one embodiment, determining the physical loss term of the functional equation includes determining it according to formula (1): (1) in, For physical loss items, This represents the predicted production capacity of horizontal wells over a historical period, calculated using the Duong production decline model. y Let T be the actual production capacity of the horizontal well over a historical period, where T is the historical period and a and m are constants.

[0008] In one embodiment, the method further includes determining the Duong production decline model according to formula (2): (2) in, This represents the real-time production capacity of horizontal wells over a historical period. The cumulative productivity of a horizontal well over a historical period is represented by n, where n is the fracture time index and T is the preset time period. This represents the initial production capacity of a horizontal well over a historical period.

[0009] In one embodiment, determining the neural network loss term of the functional equation includes determining it according to formula (3): (3) in, For the neural network loss term, This represents the predicted production capacity of horizontal wells over a historical period, calculated using the Duong production decline model. y This represents the actual production capacity of horizontal wells over a historical period.

[0010] In one embodiment, determining the functional equation of the prediction model based on the preset neural network loss term and physical loss term includes determining it according to formula (4): (4) in, The total loss function of the prediction model, For the neural network loss term, For physical loss terms, λ is the weighting factor. ).

[0011] In one embodiment, hyperparameters include at least one of the following: the size of the history window, the number of hidden layers, the number of neurons per layer, and the activation function.

[0012] A second aspect of this application provides a processor configured to execute the above-described method for predicting the productivity of horizontal wells.

[0013] A third aspect of this application provides a capacity forecasting system, comprising: The data acquisition module is used to acquire production data and physical constraint parameters of horizontal wells over a historical period. The physical constraint parameters include at least one of reservoir properties, logging curves, horizontal well information, fracturing parameters, and well control changes. The preprocessing module is used to preprocess the production data and physical constraint parameters to obtain a preprocessed dataset corresponding to the production data and physical constraint parameters. The model building module is used to determine the functional equation of the prediction model through the basic method of physical-driven neural networks; and to determine the hyperparameters of the prediction model through the Bayesian optimization method, so as to determine the model architecture of the prediction model. The prediction module divides the preprocessed dataset into multiple groups of data and inputs each group of data into the prediction model in sequence. For each group of data, the prediction model outputs the predicted production capacity of the horizontal well within the prediction period. Based on the multiple predicted production capacities, the actual predicted production capacity of the horizontal well within the prediction period is determined.

[0014] A fourth aspect of this application provides a machine-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the aforementioned method for predicting horizontal well productivity.

[0015] The above technical solution enables accurate prediction of horizontal well productivity.

[0016] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings: Figure 1 The schematic diagram illustrates a flow chart of a method for predicting horizontal well productivity according to an embodiment of this application; Figure 2 This illustration schematically shows a normalization preprocessing flow diagram of raw data from a field-fracturing horizontal well according to an embodiment of this application. Figure 3 A schematic diagram of the structure of a fourth folding model according to an embodiment of this application is shown. Figure 4 The diagram illustrates the evolution of a loss function during six cross-validation cycles according to an embodiment of this application. Figure 5 The illustration shows a schematic diagram of the prediction results of different physical constraint parameters on a test set according to an embodiment of this application; Figure 6 The illustration shows a schematic diagram of the prediction performance of the PC-BiGRU model under different physical constraint parameters according to embodiments of this application; Figure 7 A schematic diagram illustrating the structure of a capacity forecasting system according to an embodiment of this application is shown. Figure 8 The diagram illustrates the internal structure of a computer device according to an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0019] Figure 1 A schematic flowchart illustrating a method for predicting horizontal well productivity according to an embodiment of this application is shown. Figure 1 As shown in one embodiment of this application, a method for predicting the productivity of horizontal wells is provided, comprising the following steps: Step 101: Obtain production data and physical constraint parameters of the horizontal well over a historical period. The physical constraint parameters include at least one of the following: reservoir properties, logging curves, horizontal well information, fracturing parameters, and well control changes.

[0020] Step 102: Preprocess the production data and physical constraint parameters to obtain a preprocessed dataset corresponding to the production data and physical constraint parameters.

[0021] Step 103: Determine the functional equation of the prediction model using the basic method of physical-driven neural networks.

[0022] Step 104: Determine the hyperparameters of the prediction model using the Bayesian optimization method to determine the model architecture of the prediction model.

[0023] Step 105: Construct a prediction model based on the function equation and the model.

[0024] Step 106: Divide the preprocessed dataset into multiple groups of data, and input each group of data into the prediction model in sequence.

[0025] Step 107: For each set of data, output the predicted production capacity of the horizontal well within the predicted time period using the prediction model.

[0026] Step 108: Determine the actual predicted production capacity of the horizontal well within the predicted time period based on multiple predicted production capacities.

[0027] Oil and natural gas are buried tens to thousands of meters underground in oil and gas layers. To extract them, an oil and gas channel needs to be established between the surface and the underground oil and gas layers; this channel is called a well. A horizontal well is a special type of well where the maximum inclination angle reaches or approaches 90° (generally not less than 86°) and maintains a certain length of horizontal section within the target layer. Generally, horizontal wells are suitable for thin oil and gas layers or fractured oil and gas reservoirs, aiming to increase the exposed area of ​​the oil and gas layer. Horizontal wells typically employ staged fracturing technology, a production enhancement technique. When fracturing fluid enters the reservoir at high speed through the perforation, it generates perforation friction, which increases with the pumping rate, leading to an increase in bottom hole pressure. Once the bottom hole pressure exceeds the fracture pressure of multiple fracturing sections, fractures are forced open in each section. This requires the fracture pressure of each section to be relatively close, which can be adjusted using perforation friction. In this technical solution, horizontal wells can be used to extract tight oil. Tight oil refers to oil accumulations formed in tight reservoirs sandwiched or adjacent to high-quality source rocks without large-scale, long-distance migration. It is a large-area, continuously distributed oil resource that coexists with or is adjacent to the source rock system. The reservoir lithology mainly includes tight sandstone, tight limestone, and carbonate rocks, with an average overburden matrix permeability of less than 0.1 mD, and no natural industrial production capacity per well. After short-distance migration, tight oil is mainly found in tight layers adjacent to the source rocks.

[0028] Furthermore, raw data from fractured horizontal wells at the oil extraction site is collected. This raw data includes production data and physical constraint parameters of the horizontal wells over a historical time period. The historical time period can be selected based on actual needs; for example, it could be the period from the initial production date to the current date. Physical constraints refer to the use of physical methods to restrict the movement of an object, preventing it from exceeding certain limits. In this technical solution, physical constraint parameters refer to parameters that impose certain restrictions on the horizontal well and oil through physical methods. For example, physical constraint parameters can include at least one of the following: reservoir properties, logging curves, horizontal well information, fracturing parameters, and well control changes. Reservoir properties can refer to the permeability, porosity, and oil saturation of tight oil reservoirs; fracturing parameters can refer to the number of fracturing stages, total sand volume, total fluid volume, and slickwater volume; and well control changes can refer to nozzle size, shut-in time, oil pressure, and production days. Well control changes are time-series data, which can constrain time-series production predictions from a time perspective. Reservoir properties, logging curves, well information, and fracturing parameters reflect the spatial variations of different wells and constrain production variations from a spatial perspective. After acquiring the raw data from the field-fracturing horizontal wells, the processor preprocesses the production data and physical constraint parameters to obtain a preprocessed dataset corresponding to the production data and physical constraint parameters. Preprocessing can refer to performing data normalization on the production data and physical constraint parameters. Normalization can eliminate the variability of different feature quantities, and min-max normalization is used to make all features between 0 and 1.

[0029] Furthermore, the functional equation of the prediction model is determined using the basic method of physical-driven neural networks, and the hyperparameters of the prediction model are determined using Bayesian optimization, thus defining the model architecture of the prediction model. The processor can then construct the prediction model based on the functional equation and the model. Here, the functional equation can be a loss function, which maps the values ​​of a random event or its related random variables to non-negative real numbers to represent the "risk" or "loss" of that random event. The prediction model can refer to a neural network model. Specifically, determining the functional equation of the prediction model using the basic method of physical-driven neural networks means defining the neural network loss function through physical parameter constraints and yield-decreasing function physical terms. Hyperparameters are tuning parameters in machine learning algorithms; they are parameters whose values ​​are set before the learning process begins and need to be manually set. Typically, hyperparameters need to be optimized to select an optimal set of hyperparameters for the learning machine to improve learning performance and effectiveness. In this technical solution, the optimal hyperparameters are obtained by using Bayesian optimization to determine the best model architecture of the prediction model.

[0030] Furthermore, after constructing the prediction model using the basic methods of physical-driven neural networks and the Bayes optimization method, the processor divides the preprocessed dataset into multiple groups of data and sequentially inputs each group into the prediction model. For each group of data, the prediction model outputs the predicted capacity of the horizontal well within the prediction time period, and the actual predicted capacity of the horizontal well within the prediction time period is determined based on multiple predicted capacities. Specifically, the data input into the prediction model can be cross-validated to obtain the final prediction result. Cross-validation makes the evaluated model more accurate and reliable. Cross-validation involves dividing a portion of the training data input into the model into a training set and another portion into a test set, and changing the test set in each test until all test results are obtained, and the average is taken as the final result. In this technical solution, the preprocessed dataset can be divided into six groups of data, and a six-fold cross-validation method is used to test the training data output into the prediction model. Five-fold of the dataset is used as the training set to train the model, and the remaining one-fold dataset is used as the test set to evaluate the model performance. This process is repeated until all folds have been tested once. After all data testing is completed, the average of the six test results can be taken as the final prediction result for output. The processor can then use the results output by the prediction model as the actual predicted production capacity of the horizontal well within the prediction period.

[0031] Furthermore, for example, production data and physical constraint parameters of fractured horizontal wells at oil extraction sites from the first to the tenth day can be collected. After data normalization preprocessing to obtain a preprocessed dataset, it can be output to the prediction model, and six-fold cross-validation can be used to obtain the predicted production capacity of the horizontal well on the eleventh day.

[0032] The above technical solution enables accurate prediction of horizontal well productivity.

[0033] Figure 1 This is a flowchart illustrating a method for predicting horizontal well productivity in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0034] In one embodiment, determining the functional equation of the prediction model using the physical-driven neural network basic method includes: first, determining the physical loss term and the neural network loss term of the functional equation; and second, determining the functional equation of the prediction model based on the preset neural network loss term and physical loss term. Specifically, the physical loss term of the functional equation includes terms determined according to formula (1): (1) in, For physical loss items, y represents the predicted production capacity of the horizontal well over a historical period using the Duong production decline model, y represents the actual production capacity of the horizontal well over a historical period, T represents the historical period, and a and m are constants.

[0035] In actual horizontal well production curve prediction, it is difficult to embed the seepage differential equation into the loss function because parameters such as pressure and saturation are usually unknown. The Duong production decline curve, however, is a broad approximation of production in unconventional oil and gas reservoir fractured wells and can be used to standardize long-term production prediction results. The Duong production decline model is determined according to formula (2): (2) Wherein, is the real-time production capacity of the horizontal well within the historical time period, is the cumulative production capacity of the horizontal well within the historical time period, n is the fracture time index, T is the preset time period, and is the initial production capacity of the horizontal well within the historical time period.

[0036] Furthermore, based on the linear flow assumption of the crack: (2-1) Where a and m are constants, the above formula is an empirical formula, and generally a=0.5 and m=1.23 can be taken.

[0037] Therefore, we can obtain the Duong production decline formula: (2-2) The rate of decline in output, D(T), can be defined as follows: (2-3) Substituting formula (2-2) into (2-3), we get: (2-4) make: (2-5) Then its derivative is: (2-6) Substituting equations (2-6) and (2-5) into equation (2-4), we get: (2-7) Further derivation of equation (2-7) yields: (2-8) Equation (2-3) can be modified as follows: (2-9) Substituting equation (2-8) into equation (2-9), we get: (2-10) Based on the above calculation formula, the physical loss term can be defined as formula (1).

[0038] In one embodiment, the Duong production decline formula is a broad approximation of the long-term production situation of fractured wells. Incorporating this formula into the loss function aims to constrain the long-term production prediction trend. However, the initial production of a specific fractured well does not satisfy production decline, therefore the physical loss term is not directly minimized, but rather embedded as a soft constraint in the total loss function. Specifically, the neural network loss term of the functional equation includes the terms determined according to formula (3): (3) in, For the neural network loss term, y represents the predicted production capacity of the horizontal well over a historical period, calculated using the Duong production decline model, while y represents the actual production capacity of the horizontal well over the historical period.

[0039] Furthermore, based on the physical loss term and the neural network loss term, the total loss function of the neural model is obtained, which is determined according to formula (4): (4) in, The total loss function of the prediction model, For the neural network loss term, For physical loss terms, λ is the weighting factor. ).

[0040] In one embodiment, after determining the functional equation of the prediction model using the basic method of physical-driven neural networks, the processor obtains the optimal hyperparameters by selecting a Bayesian optimization method to determine the best model architecture for the prediction model. Bayesian optimization is a commonly used hyperparameter optimization method in the field of machine learning and can be considered the most advanced optimization framework currently available. In this technical solution, the hyperparameters of the prediction model can refer to the size of the historical window, the number of hidden layers, the number of neurons per layer, and the activation function.

[0041] In one specific embodiment, a method is provided for constructing a novel physical constraint deep learning model based on bidirectional gated recursive units and deep hybrid neural networks, and predicting the production capacity of multi-stage fractured horizontal wells in tight oil. Specifically, by modifying the Python source program, raw data from field fractured horizontal wells are collected, including production data and physical constraint parameters. The physical constraint parameters include reservoir properties, logging curves, well information, fracturing parameters, and well control variations. A data normalization preprocessing procedure is then performed to eliminate the variability of different feature quantities.

[0042] like Figure 2 The diagram shows a normalization preprocessing flow for raw data from a horizontal well undergoing fracturing in the field. Specifically, data normalization is first performed to eliminate the variability of different feature quantities. Min-max normalization is used to make all features between 0 and 1, and the formula is defined as follows: Formula (5) (5) in, and And these represent the original and normalized input features at time step T, respectively. and These represent the minimum and maximum values ​​of the input sequence, respectively.

[0043] To meet the input requirements of a temporal neural network model, production data should be converted into input-output pairs using a sliding history window. The specific format conversion process is as follows: Figure 2 As shown, suppose there is a production sequence with (n+3) data points. Starting from day 1, the production data of the first n days are considered as the input to the first sample, and the production information of day (n+1) is considered as the output of the first sample. Next, the sliding window moves forward one day. The production data from day 2 to day (n+1) are used as the input to the second sample, and the production data of day (n+2) is used as the output of the first sample, continuing this process until all data is covered. In this way, the production sequence with (n+3) data points is transformed into three input-output samples.

[0044] Furthermore, such as Figure 3The diagram illustrates the structure of the fourth-fold model. Specifically, the oil rate input from the time series is converted into a three-dimensional tensor (5052, 8, 1) and fed into an 81-node BiGRU layer to obtain the temporal correlation between production sequences. Based on the data dimension, physical constraints are passed to fully connected layers and convolutional layers to extract useful representations. Two-dimensional constraints (including well information, fracturing parameters, and operational variations) are passed through two dense layers and 72 nodes, while three-dimensional constraints (such as logging sequences) are passed to the convolutional layer. A dense layer is added after the time series network layers. Except for the output layer, where the activation function is linear, the activation functions of all other layers are optimized to sigmoid. In this way, temporal and spatial physical parameters are integrated into the long-term prediction network architecture, which helps to output more accurate and physically consistent predictions.

[0045] Furthermore, Bayesian optimization was employed to optimize the hyperparameters of the PC-BiGRU ensemble neural network model. Due to six cross-validations, a total of six models were required for optimization. The optimized hyperparameters included the length of the history window, the number of hidden layers, and the number of neurons and activation functions within these layers. Based on prediction accuracy and computational cost, we approximated the search range of layers through iterative trials, using a history window of 14 and a neuron count of 64.

[0046] like Figure 4 The diagram illustrates the evolution of the loss function during six cross-validation iterations. Specifically, when the number of layers in all modules exceeds four, computation time increases significantly while accuracy drops sharply. Therefore, the search range for the number of layers is set to 1-4. Bayesian optimization is then used to find the optimal hyperparameters for the six models.

[0047] Furthermore, a validation set was used to evaluate the accuracy of the pre-configured tight oil fracturing horizontal well productivity prediction network model obtained through training, and the quality of the pre-configured reservoir network model based on root mean square error was further verified. The root mean square error formula is defined as follows (6).

[0048] (6) In the formula, This indicates the production capacity using the original collected tight oil. H represents the tight oil production capacity predicted using a pre-configured reservoir network model at the same time point, where H represents the time step.

[0049] Furthermore, such as Figure 5The diagram illustrates the prediction results on the test set with different physical constraint parameters. After obtaining the PC-BiGRU combined neural network model with optimal hyperparameters, network model parameter optimization is performed on the training set. The optimization parameters are set as follows: Adam is used as the optimizer, and the learning rate is set to 0.001. All weight matrices are initialized using Xavier initialization, and the bias matrix is ​​initialized to zero. All models are run for 50 epochs with a batch size of 50. Figure 4 The evolution of the loss function with respect to the period is described, showing that all cross-validated PC-BiGRU models converge after 50 periods without overfitting. These models are coded in Python 3.7 using the Keras deep learning library and run on an Intel® Xeon® 3.40 GHz CPU computer. The impact of different physical constraints on model performance will be explored, which helps determine the contribution of different physical constraints to improving production prediction accuracy. Well information and fracturing parameters are considered as engineering constraints (EC) because well completion and fracturing are related to engineering operations. Well logging curves are classified as geological constraints (GC) because they can indirectly reflect reservoir properties and rock mechanics. Operational changes are represented by OC, and ALL indicates that all constraints are considered. Different types of physical constraints are integrated into the PC-BiGRU models separately.

[0050] Furthermore, such as Figure 6 As shown, a schematic diagram illustrating the prediction performance of the PC-BiGRU model under different physical constraint parameters is provided. Figure 6 It can be seen that OC contributes the most to the model performance, followed by EC and GC. This is likely due to the rich dynamic information related to oil rate fluctuations contained in daily operational changes. Furthermore, completion and fracturing parameters can reflect the scale of production increase, implying the magnitude of the increase, while geological features characterize reservoir heterogeneity and inter-well variability. It can be observed that in multi-step advanced production prediction tasks, GC and EC are responsible for constraining the overall trend of long-term production, while OC focuses on the subtle fluctuations in time-series production.

[0051] The aforementioned technical solution is based on a physically constrained bidirectional recurrent neural network for predicting the production capacity of multi-stage fractured horizontal wells in tight oil. It collects raw data from fractured horizontal wells in the field, performs data normalization preprocessing, and conducts cross-validation on the preprocessed dataset to mitigate the impact of random dataset partitioning and objectively evaluate the neural network's prediction performance. Secondly, it employs Bayesian optimization to obtain optimal hyperparameters, determines the best model architecture, and defines the neural network loss function through physical parameter constraints and production decline function physical constraints. Deep learning model training is then conducted, and the prediction results are evaluated on a test set. This addresses the technical problem of low accuracy in predicting the production capacity of fractured horizontal wells relying solely on pure data-driven methods.

[0052] In one embodiment, such as Figure 7 As shown, a capacity forecasting system 700 is provided, comprising: The data acquisition module 710 is used to acquire production data and physical constraint parameters of the horizontal well over a historical period. The physical constraint parameters include at least one of reservoir properties, logging curves, horizontal well information, fracturing parameters, and well control changes. The preprocessing module 720 is used to preprocess the production data and physical constraint parameters to obtain a preprocessed dataset corresponding to the production data and physical constraint parameters. The model building module 730 is used to determine the functional equation of the prediction model through the basic method of physical-driven neural networks; and to determine the hyperparameters of the prediction model through the Bayesian optimization method, so as to determine the model architecture of the prediction model. The prediction module 740 divides the preprocessed dataset into multiple groups of data and inputs each group of data into the prediction model in sequence; for each group of data, the prediction model outputs the predicted production capacity of the horizontal well within the prediction period; and determines the actual predicted production capacity of the horizontal well within the prediction period based on the multiple predicted production capacities.

[0053] This application provides a storage medium storing a program that, when executed by a processor, implements the above-described method for predicting horizontal well productivity.

[0054] This application provides a processor for running a program, wherein the program executes the above-described method for predicting horizontal well productivity.

[0055] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8 As shown. The computer device includes a processor A01, a network interface A02, a memory (not shown), and a database (not shown) connected via a system bus. The processor A01 provides computing and control capabilities. The memory includes internal memory A03 and a non-volatile storage medium A04. The non-volatile storage medium A04 stores an operating system B01, a computer program B02, and a database (not shown). The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 in the non-volatile storage medium A04. The database stores prediction data for horizontal well productivity. The network interface A02 communicates with external terminals via a network connection. When executed by the processor A01, the computer program B02 implements a method for predicting horizontal well productivity.

[0056] Those skilled in the art will understand that Figure 8The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0057] This application provides an apparatus including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: acquiring production data and physical constraint parameters of a horizontal well over a historical time period, the physical constraint parameters including at least one of reservoir properties, logging curves, horizontal well information, fracturing parameters, and well control changes; preprocessing the production data and physical constraint parameters to obtain a preprocessed dataset corresponding to the production data and physical constraint parameters; determining the functional equation of a prediction model using a physical-driven neural network basic method; determining the hyperparameters of the prediction model using a Bayesian optimization method to determine the model architecture of the prediction model; constructing a prediction model based on the functional equation and the model; dividing the preprocessed dataset into multiple groups of data and sequentially inputting each group of data into the prediction model; for each group of data, outputting the predicted production capacity of the horizontal well within the prediction time period through the prediction model; and determining the actual predicted production capacity of the horizontal well within the prediction time period based on multiple predicted production capacities.

[0058] In one embodiment, determining the functional equation of the prediction model using the physical-driven neural network basic method includes: determining the physical loss term and the neural network loss term of the functional equation; and determining the functional equation of the prediction model based on the preset neural network loss term and the physical loss term.

[0059] In one embodiment, determining the physical loss term of the functional equation includes determining it according to formula (1): (1) in, For physical loss items, This represents the predicted production capacity of horizontal wells over a historical period, calculated using the Duong production decline model. y Let T be the actual production capacity of the horizontal well over a historical period, where T is the historical period and a and m are constants.

[0060] In one embodiment, the method further includes determining the Duong production decline model according to formula (2): (2) in, This represents the real-time production capacity of horizontal wells over a historical period. The cumulative productivity of a horizontal well over a historical period is represented by n, where n is the fracture time index and T is the preset time period. This represents the initial production capacity of a horizontal well over a historical period.

[0061] In one embodiment, determining the neural network loss term of the functional equation includes determining it according to formula (3): (3) in, For the neural network loss term, This represents the predicted production capacity of horizontal wells over a historical period, calculated using the Duong production decline model. y This represents the actual production capacity of horizontal wells over a historical period.

[0062] In one embodiment, determining the functional equation of the prediction model based on the preset neural network loss term and physical loss term includes determining it according to formula (4): (4) in, The total loss function of the prediction model, For the neural network loss term, For physical loss terms, λ is the weighting factor. ).

[0063] In one embodiment, hyperparameters include at least one of the following: the size of the history window, the number of hidden layers, the number of neurons per layer, and the activation function.

[0064] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program comprising the following steps: acquiring production data and physical constraint parameters of a horizontal well over a historical time period, wherein the physical constraint parameters include at least one of reservoir properties, logging curves, horizontal well information, fracturing parameters, and well control changes; preprocessing the production data and physical constraint parameters to obtain a preprocessed dataset corresponding to the production data and physical constraint parameters; determining the functional equation of a prediction model using a physical-driven neural network basic method; determining the hyperparameters of the prediction model using a Bayesian optimization method to determine the model architecture of the prediction model; constructing a prediction model based on the functional equation and the model; dividing the preprocessed dataset into multiple sets of data and sequentially inputting each set of data into the prediction model; for each set of data, outputting the predicted production capacity of the horizontal well over the prediction time period through the prediction model; and determining the actual predicted production capacity of the horizontal well over the prediction time period based on multiple predicted production capacities.

[0065] In one embodiment, determining the functional equation of the prediction model using the physical-driven neural network basic method includes: determining the physical loss term and the neural network loss term of the functional equation; and determining the functional equation of the prediction model based on the preset neural network loss term and the physical loss term.

[0066] In one embodiment, determining the physical loss term of the functional equation includes determining it according to formula (1): (1) in, For physical loss items, This represents the predicted production capacity of horizontal wells over a historical period, calculated using the Duong production decline model. y Let T be the actual production capacity of the horizontal well over a historical period, where T is the historical period and a and m are constants.

[0067] In one embodiment, the method further includes determining the Duong production decline model according to formula (2): (2) in, This represents the real-time production capacity of horizontal wells over a historical period. The cumulative productivity of a horizontal well over a historical period is represented by n, where n is the fracture time index and T is the preset time period. This represents the initial production capacity of a horizontal well over a historical period.

[0068] In one embodiment, determining the neural network loss term of the functional equation includes determining it according to formula (3): (3) in, For the neural network loss term, This represents the predicted production capacity of horizontal wells over a historical period, calculated using the Duong production decline model. y This represents the actual production capacity of horizontal wells over a historical period.

[0069] In one embodiment, determining the functional equation of the prediction model based on the preset neural network loss term and physical loss term includes determining it according to formula (4): (4) in, The total loss function of the prediction model, For the neural network loss term, For physical loss terms, λ is the weighting factor. ).

[0070] In one embodiment, hyperparameters include at least one of the following: the size of the history window, the number of hidden layers, the number of neurons per layer, and the activation function.

[0071] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0073] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0074] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0075] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0076] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0077] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0078] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0079] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for predicting the productivity of horizontal wells, characterized in that, The prediction method includes: Obtain production data and physical constraint parameters of the horizontal well over a historical period. The physical constraint parameters include at least one of reservoir properties, logging curves, horizontal well information, fracturing parameters, and well control changes. The production data and the physical constraint parameters are preprocessed to obtain a preprocessed dataset corresponding to the production data and the physical constraint parameters; The functional equations of the prediction model are determined using the basic method of physical-driven neural networks; The hyperparameters of the prediction model are determined by Bayesian optimization methods, thereby determining the model architecture of the prediction model; The prediction model is constructed based on the functional equation and the model. The preprocessed dataset is divided into multiple groups of data, and each group of data is input into the prediction model in sequence. For each set of data, the predicted production capacity of the horizontal well within the predicted time period is output through the prediction model; The actual predicted production capacity of the horizontal well within the predicted time period is determined based on multiple predicted production capacities. Among them, the functional equations of the prediction model determined by the basic method of physical-driven neural networks include: Determine the physical loss term and the neural network loss term of the functional equation; The functional equation of the prediction model is determined based on the preset neural network loss term and the physical loss term; Determining the physical loss term of the functional equation includes determining it according to formula (1): （1） in, For the physical loss term, The predicted production capacity of the horizontal well during the historical time period is calculated using the Duong production decline model. y The actual production capacity of the horizontal well during the historical time period is T, where T is the historical time period, and a and m are constants. The Duong production decline model is determined according to formula (2): （2） in, The real-time production capacity of the horizontal well during the historical time period. The cumulative production capacity of the horizontal well during the historical time period is given by n, where n is the fracture time index and T is a preset time period. The initial production capacity of the horizontal well during the historical time period.

2. The method for predicting horizontal well productivity according to claim 1, characterized in that, Determining the neural network loss term of the functional equation includes determining it according to formula (3): （3） in, For the loss term of the neural network, The predicted production capacity of the horizontal well during the historical time period is calculated using the Duong production decline model. y The actual production capacity of the horizontal well during the historical time period.

3. The method for predicting horizontal well productivity according to claim 1, characterized in that, The step of determining the functional equation of the prediction model based on the preset neural network loss term and the physical loss term includes determining it according to formula (4): （4） in, The total loss function of the prediction model is... For the loss term of the neural network, Let λ be the physical loss term, and λ be the weighting factor. ).

4. The method for predicting horizontal well productivity according to claim 1, characterized in that, The hyperparameters include at least one of the following: the size of the history window, the number of hidden layers, the number of neurons per layer, and the activation function.

5. A processor, characterized in that, It is configured to perform the method for predicting horizontal well productivity as described in any one of claims 1 to 4.

6. A capacity forecasting system, characterized in that, Configured to perform the method for predicting horizontal well productivity according to any one of claims 1 to 4, comprising: The data acquisition module is used to acquire production data and physical constraint parameters of horizontal wells over a historical period. The physical constraint parameters include at least one of reservoir properties, logging curves, horizontal well information, fracturing parameters, and well control changes. A preprocessing module is used to preprocess the production data and the physical constraint parameters to obtain a preprocessed dataset corresponding to the production data and the physical constraint parameters. The model building module is used to determine the functional equation of the prediction model through the basic method of physical-driven neural networks; and to determine the hyperparameters of the prediction model through the Bayesian optimization method, so as to determine the model architecture of the prediction model. The prediction module divides the preprocessed dataset into multiple groups of data and inputs each group of data into the prediction model in sequence; for each group of data, the prediction model outputs the predicted production capacity of the horizontal well within the prediction time period; and determines the actual predicted production capacity of the horizontal well within the prediction time period based on multiple predicted production capacities. Among them, the functional equations of the prediction model determined by the basic method of physical-driven neural networks include: Determine the physical loss term and the neural network loss term of the functional equation; The functional equation of the prediction model is determined based on the preset neural network loss term and the physical loss term; Determining the physical loss term of the functional equation includes determining it according to formula (1): （1） in, For the physical loss term, The predicted production capacity of the horizontal well during the historical time period is calculated using the Duong production decline model. y The actual production capacity of the horizontal well during the historical time period is T, where T is the historical time period, and a and m are constants. The Duong production decline model is determined according to formula (2): （2） in, The real-time production capacity of the horizontal well during the historical time period. The cumulative production capacity of the horizontal well during the historical time period is given by n, where n is the fracture time index and T is a preset time period. The initial production capacity of the horizontal well during the historical time period.

7. A machine-readable storage medium storing instructions thereon, characterized in that, When executed by a processor, this instruction causes the processor to be configured to perform the method for predicting horizontal well productivity according to any one of claims 1 to 4.

Images