Method and system for intelligently predicting yield of offshore oil and gas well network

Abstract

The embodiment of the invention discloses an intelligent prediction method and system for the yield of an offshore oil and gas well network. The intelligent prediction method comprises the steps that time sequence data of the gas production rate and time sequence data of production key parameters corresponding to each single well in the well network are obtained, and the static inter-well relation representing the relative distance between the single wells in the well network is obtained; feature extraction is carried out based on the time sequence data of the gas production rate corresponding to each single well and the time sequence data of the production key parameters, and a feature matrix corresponding to each single well is obtained; the feature matrix corresponding to each single well is input into a trained prediction model to obtain the predicted gas production rate of the well pattern, and the trained prediction model comprises an adjacent matrix representing the inter-well interference relation. The problem of prediction misalignment of an existing data model for a complex well network system with the inter-well dynamic interference phenomenon is solved, and accurate prediction of the productivity of the well network system is achieved.

Description

Technical Field

[0001] Multiple embodiments of this specification relate to the technical field of offshore oil and gas field productivity prediction, and specifically to accuracy optimization of oil and gas well network productivity prediction. Background Art

[0002] The inherent complexity and dynamic nature of offshore oil and gas reservoirs present multiple challenges in achieving efficient development. First, offshore reservoirs generally exhibit spatial heterogeneity and dynamic interference between wells, resulting in complex spatiotemporal dynamics in the evolution of well productivity. Second, the variability and uncertainty of the marine environment, such as tides, currents, and temperature-salinity gradients, further complicate reservoir dynamics.

[0003] In recent years, with the rapid development of computer technology and artificial intelligence, data-driven intelligent forecasting methods have garnered widespread attention. These methods leverage historical oil and gas field production data and, through machine learning algorithms, establish a mapping relationship between input and output to directly predict future production capacity dynamics. Artificial intelligence technologies, particularly deep learning, demonstrate significant potential in the field of oil and gas production capacity forecasting, thanks to their powerful feature extraction and nonlinear mapping capabilities.

[0004] However, existing methods still have significant limitations. Existing models are primarily limited to modeling at the scale of a single well and lack an effective representation of the spatial dependencies between wells. This makes them difficult to apply to complex well patterns with significant inter-well interference, making it difficult to accurately predict the production capacity of these patterns. Summary of the Invention

[0005] The embodiments of this specification provide a method and system for intelligent prediction of offshore oil and gas well network production, which solves the problem of inaccurate prediction of existing data models for complex well network systems with dynamic interference between wells, and realizes accurate prediction of the production capacity of the well network system.

[0006] The technical solution is as follows:

[0007] In a first aspect, the embodiments of this specification provide a method for intelligently predicting the production of an offshore oil and gas well network, comprising the following steps:

[0008] Obtain the time series data of gas production and key production parameters corresponding to each individual well in the well network, as well as the static well-to-well relationship that characterizes the relative distance between each individual well in the well network;

[0009] Based on the time series data of gas production and key production parameters corresponding to each single well, feature extraction is performed to obtain the feature matrix corresponding to each single well;

[0010] Inputting the characteristic matrix corresponding to each single well into a trained prediction model to obtain the predicted gas production of the well pattern, wherein the trained prediction model includes an adjacency matrix representing the interference relationship between wells;

[0011] The prediction model to be trained includes a parameter matrix with initial values ​​set based on the static well-to-well relationship, and the adjacency matrix is ​​a post-learning parameter matrix corresponding to the trained prediction model.

[0012] As a preferred solution, the method of obtaining the time series data of the gas production and the time series data of the key production parameters corresponding to each single well in the well pattern includes:

[0013] Obtain the time series data of gas production and various production parameters corresponding to each single well in the well pattern;

[0014] Based on the similarity between the time series data of the gas production corresponding to each single well in the well network and the time series data of multiple production parameters, key production parameters are selected from the multiple production parameters.

[0015] As a preferred solution, the key production parameters are selected from the multiple production parameters based on the similarity between the time series data of the gas production corresponding to each single well in the well pattern and the time series data of multiple production parameters, including:

[0016] Based on the time series data of gas production and multiple production parameters corresponding to each single well in the well pattern, the dynamic time warping method is used to obtain the similarity score ranking information of the multiple production parameters corresponding to each single well;

[0017] Based on the similarity score ranking information of the multiple production parameters corresponding to all the single wells, key production parameters are selected from the multiple production parameters.

[0018] As a preferred solution, the parameter matrix includes a pair of a first parameter matrix and a second parameter matrix characterizing asymmetric interference between wells, and a third parameter matrix characterizing the characteristics of each single well in the well network.

[0019] As a preferred solution, the training method of the prediction model includes the following steps:

[0020] Setting the initial values ​​of the parameter matrix based on the static well-to-well relationship;

[0021] The current parameter matrix and the feature matrix corresponding to each single well are input into the prediction model to be trained. The current parameter matrix is ​​updated based on the loss function of the prediction model to be trained, and this step is repeated with the updated parameter matrix as the new current parameter matrix until the loss function of the prediction model to be trained meets the convergence condition to obtain the trained prediction model.

[0022] As a preferred solution, the updating of the current parameter matrix based on the loss function of the prediction model to be trained includes:

[0023] Perform a preliminary update on the current parameter matrix based on the loss function of the prediction model to be trained;

[0024] Obtain the weight matrices of the two parameter matrices corresponding to the update and after through the multi-head attention mechanism;

[0025] A secondary update is performed based on the two parameter matrices and their weight matrices corresponding to each other before and after the update to obtain an updated parameter matrix.

[0026] As a preferred solution, the feature extraction is performed based on the time series data of the gas production volume and the time series data of the key production parameters corresponding to each single well to obtain the feature matrix corresponding to each single well, including:

[0027] For any single well, local fine features and global trend features are extracted based on the corresponding time series data of gas production and key production parameters;

[0028] Based on the local fine features and global trend features, feature fusion is performed to obtain the feature matrix corresponding to the single well;

[0029] Repeat the above steps until the characteristic matrix corresponding to each single well is obtained.

[0030] As a preferred solution, for any single well, local fine features and global trend features are extracted based on the corresponding time series data of gas production and time series data of key production parameters, including:

[0031] For any single well, based on the corresponding time series data of gas production and key production parameters, a sliding window method is used to extract local fine features, and a global average pooling method is used to extract global trend features.

[0032] As a preferred solution, the prediction method further includes:

[0033] Output a visualization of the well-to-well relationship diagram based on the parameter matrix.

[0034] In a second aspect, the embodiments of this specification provide an intelligent prediction system for offshore oil and gas well network production, including a data acquisition module, a feature extraction module, and a production capacity prediction module;

[0035] The data acquisition module acquires the time series data of gas production and key production parameters corresponding to each single well in the well pattern, as well as the static well-to-well relationship representing the relative distance between each single well in the well pattern;

[0036] The feature extraction module extracts features based on the time series data of gas production and key production parameters corresponding to each single well, and obtains a feature matrix corresponding to each single well;

[0037] The production capacity prediction module inputs the characteristic matrix corresponding to each single well into a trained prediction model to obtain the predicted gas production of the well network. The trained prediction model includes an adjacency matrix that characterizes the interference relationship between wells. The prediction model to be trained includes a parameter matrix with initial values ​​set based on static well-to-well relationships. The adjacency matrix is ​​the learned parameter matrix corresponding to the trained prediction model.

[0038] In a third aspect, an embodiment of this specification provides an electronic device comprising a processor and a memory; the processor is connected to the memory; the memory is used to store executable program code; the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to execute the steps described in the first aspect of the above embodiment.

[0039] In a fourth aspect, an embodiment of this specification provides a computer storage medium, wherein the computer storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executing the steps described in the first aspect of the above embodiment.

[0040] The beneficial effects of the technical solutions provided by some embodiments of this specification include at least:

[0041] 1. Existing models are mainly limited to modeling at the scale of a single well and lack an effective expression of the spatial dependencies between wells. This makes them difficult to apply to complex well network systems with significant inter-well interference effects, and therefore makes it difficult to accurately predict the production capacity of well network systems. This method proposes an innovative framework for adaptive spatiotemporal graph neural networks. On the one hand, by introducing a graph network structure, the topological relationship of the well network is explicitly modeled, and a graph attention mechanism is designed to adaptively learn the dynamic correlation strength between wells. On the other hand, based on the graph network, spatiotemporal convolution and attention modules are incorporated to extract trend and periodic patterns in time series data from a multi-scale perspective. This method fully considers the spatiotemporal dynamic characteristics of the well network system and the inter-well interference effects during the development of offshore gas reservoirs, solves the problem of inaccurate prediction of existing data models for complex well network systems with dynamic inter-well interference, and achieves accurate prediction of the production capacity of the well network system.

[0042] 2. Modeling and training based on the relative distances between individual wells in the well network and the time series data of each individual well can effectively guide the model in the right direction at the beginning of training and improve training speed.

[0043] 3. Interpretable feature analysis based on dynamic time warping improves the ability to identify key features.

[0044] 4. The high- and low-frequency attention fusion mechanism is introduced to capture the multi-scale spatiotemporal dynamic patterns in time series data, significantly enhancing the ability to depict complex production dynamics.

[0045] 5. The present invention performs a visual analysis of the interference between individual wells in the well pattern system. The visual analysis results corresponding to different production stages can intuitively reveal the evolution law of the interference between wells in different development stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 This is a flow chart of an intelligent prediction method for offshore oil and gas well network production provided by an embodiment of this specification;

[0048] Figure 2 This is a comparison chart of the prediction effects of the adaptive spatiotemporal graph neural network model and the existing data model on the test set in the embodiment of this specification;

[0049] Figure 3 It is a visualization result diagram of the evolution of dynamic relationships between wells;

[0050] Figure 4 This is a schematic diagram of the structure of an offshore oil and gas well network production intelligent prediction system provided by an embodiment of this specification;

[0051] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of this specification will be described clearly and completely below in conjunction with the drawings in the embodiments of this specification.

[0053] The terms "first," "second," "third," and so on, in the description and claims of this specification and in the accompanying drawings are used to distinguish between different items, not to describe a specific order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements, but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.

[0054] The following description provides examples and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made to the functions and arrangements of the elements described without departing from the scope of this specification. Various examples may appropriately omit, replace, or add various processes or components. For example, the described methods may be performed in an order different from the order described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined in other examples.

[0055] This invention aims to address the technical challenges of multi-well production capacity prediction in the development of complex offshore oil and gas reservoirs. Traditional methods use single-well time-series prediction models to predict the production values ​​of individual wells and aggregate these values ​​to form the total production forecast for the well network. However, this approach struggles to effectively model the dynamic evolution of the well network system, resulting in limited prediction accuracy and adaptability. Therefore, an intelligent method is needed that can fully exploit the spatiotemporal dynamic characteristics inherent in well network production data and accurately predict the dynamic production capacity of multiple wells. This approach will provide crucial support for scientific decision-making and efficient development of offshore gas fields.

[0056] Reference Figure 1 As shown, Figure 1 A flowchart of an intelligent prediction method for offshore oil and gas well network production provided in one embodiment of this specification may include at least the following steps:

[0057] Step 202: Obtain the time series data of the gas production and key production parameters corresponding to each well in the well pattern, as well as the static well-to-well relationship representing the relative distance between the wells in the well pattern.

[0058] For illustration, the gas production and key production parameters of each single well in the offshore oil and gas well network are continuously collected, and through pre-processing such as data cleaning and normalization, the time series data of the gas production and the time series data of the key production parameters corresponding to each single well are obtained.

[0059] For illustrative purposes, the relative distances between individual wells in the well network are determined based on the static inter-well relationships of the well network. For example, the distance between any two wells is calculated from the coordinates of the individual wells. After numbering the individual wells, a distance matrix X for the well network is generated. The elements in distance matrix X are the inverses of the distances between any two wells. The closer the distance, the stronger the physical connection, and the correspondingly higher the interference. This matrix serves as the initial value for the parameter matrix. Other forms of parameter matrix initial values ​​based on the relative distances between individual wells are also within the scope of this embodiment.

[0060] It's important to note that due to the complexity of reality, such as the presence of heterogeneous geological structures like fractures and faults in underground reservoirs, two wells physically close together may be isolated by a fault, while wells farther apart may be connected by fractures. Therefore, the inter-well interference relationships in the well network must be learned rather than directly determined by physical distance. The distance matrix X is used as the initial state of the inter-well interference relationships, which is then adjusted during model training to correct geological understanding (e.g., for uncalibrated faults). Using the distance matrix X as the initial state of the inter-well interference relationships provides a scientific "starting point" for the model, rather than forcing it to learn from scratch. Instead, a relatively accurate initial sketch of relationships is first drawn based on distance relationships. The primary function of this initial sketch is to guide the model in the right direction during the initial training phase. Once model training begins, the model continuously learns and refines this sketch using real-world data, ultimately forming a more accurate dynamic network of relationships used for prediction, namely the adjacency matrix.

[0061] In one embodiment of the present specification, obtaining time series data of gas production and key production parameters corresponding to each well in the well pattern includes:

[0062] Step 2022: Obtain the time series data of the gas production and the time series data of various production parameters corresponding to each single well in the well pattern;

[0063] Step 2024: Based on the similarity between the time series data of the gas production corresponding to each single well in the well pattern and the time series data of the multiple production parameters, a key production parameter is selected from the multiple production parameters.

[0064] Explanatory, by evaluating the correlation between each production parameter and the time series data of gas production, the parameters with the highest similarity are selected as key production parameters, thereby excluding parameters that are irrelevant to gas production or have low correlation, avoiding interference with the well network gas production prediction.

[0065] Illustratively, the key production parameters may include multiple production parameters, and a preset number of production parameters are selected as the key production parameters in descending order of correlation.

[0066] For example, the selected key production parameters are wellhead temperature, wellhead pressure and nozzle opening.

[0067] In one embodiment of the present specification, based on the similarity between the time series data of the gas production corresponding to each single well in the well pattern and the time series data of multiple production parameters, key production parameters are selected from the multiple production parameters, including:

[0068] Based on the time series data of gas production and multiple production parameters corresponding to each single well in the well pattern, the dynamic time warping method is used to obtain the similarity score ranking information of the multiple production parameters corresponding to each single well;

[0069] Based on the similarity score ranking information of the multiple production parameters corresponding to all the single wells, key production parameters are selected from the multiple production parameters.

[0070] For illustration, Dynamic Time Warping (DTW) is a method that measures the similarity between two time series, finding the most natural alignment between the series, even if they have different lengths, frequencies, phase offsets, and local deformations. Dynamic Time Warping is used to calculate the correlation between each production parameter and its gas production time series data for each well. The calculation formula is:

[0071]

[0072] in, represents the time series data of the i-th production parameter, and y represents the time series data of gas production. The correlation score for the corresponding production parameter is calculated. A higher score indicates a stronger dynamic similarity between the production parameter and gas production. The production parameters are sorted from high to low based on their correlation scores to obtain similarity score ranking information. The top few production parameters in the similarity score ranking information corresponding to each well are selected, and the preset number of production parameters with the highest number of occurrences are counted as key production parameters.

[0073] Step 204 : Perform feature extraction based on the time series data of the gas production and the time series data of the key production parameters corresponding to each single well to obtain a feature matrix corresponding to each single well.

[0074] Illustratively, this embodiment proposes an innovative framework for an adaptive spatiotemporal graph neural network. This adaptive spatiotemporal graph neural network primarily comprises a dynamic well-to-well relationship graph learning module and a spatiotemporal convolutional representation module. The dynamic well-to-well relationship graph learning module explicitly models the well network topology by introducing a graph network structure and employs a graph attention mechanism to adaptively learn the dynamic strength of inter-well correlations. Building on the graph network, the spatiotemporal convolutional representation module incorporates spatiotemporal convolution and attention modules to extract trends and cyclical patterns in time series data from a multi-scale perspective.

[0075] Explanatory, the spatiotemporal convolutional representation module extracts time series features for each single well and obtains the corresponding feature matrix, which serves as the feature input for subsequent prediction of well network production.

[0076] Specifically, in one embodiment of the present specification, step 204 performs feature extraction based on the time series data of the gas production volume and the time series data of the key production parameters corresponding to each single well, and obtains a feature matrix corresponding to each single well, including:

[0077] For any single well, local fine features and global trend features are extracted based on the corresponding time series data of gas production and key production parameters;

[0078] Based on the local fine features and global trend features, feature fusion is performed to obtain the feature matrix corresponding to the single well;

[0079] Repeat the above steps until the characteristic matrix corresponding to each single well is obtained.

[0080] Explanatory, spatiotemporal convolutional representation module designs multi-scale convolution kernels and pooling operations to achieve high- and low-frequency fusion design. By adopting parallel high- and low-frequency self-attention structures, local fine features in the time series are modeled separately. (short span) and global trend characteristics (long span) to characterize the feature evolution law under different time spans, and then realize the adaptive fusion of high-frequency and low-frequency features through the fusion mechanism based on the gate control unit.

[0081] Specifically, the fusion process is as follows:

[0082]

[0083]

[0084] in, is the sigmoid activation function, represents element-wise multiplication, and is a learnable parameter, G is the weight matrix of local fine features, 1-G is the weight matrix of global trend features, is the fused feature matrix.

[0085] Furthermore, in one embodiment of the present specification, for any single well, local fine features and global trend features are extracted based on the corresponding time series data of gas production and time series data of key production parameters, including:

[0086] For any single well, based on the corresponding time series data of gas production and key production parameters, a sliding window method is used to extract local fine features, and a global average pooling method is used to extract global trend features.

[0087] Illustratively, the extraction of local fine features uses a sliding time window to process the high-frequency components in the time series data and capture hourly or daily fluctuations; the extraction of global trend features uses global pooling to process the low-frequency components in the time series data and capture monthly or yearly trends.

[0088] Specifically:

[0089] (1) For high-frequency components, a self-attention mechanism is applied in a sliding time window manner to capture short-term fluctuations in the time series. The high-frequency attention is calculated using the following formula:

[0090]

[0091]

[0092] Among them, w is the window length, t is the starting position of the window, is a time window in the input time series, are all learnable weight matrices, used to generate queries, keys, and values, respectively. Linear(·) is a linear transformation function that maps the input to a new space through the weight matrix, representing the linear layer (fully connected layer) of the neural network. They are the query, key, and value matrices of high-frequency attention, respectively, and are passed to the subsequent attention scoring step. is the high-frequency attention weight, It is a vector dimension, and the Softmax function converts the vector into a probability distribution.

[0093] (2) For the low-frequency component, global average pooling is used to generate the low-frequency representation of the time series, and the low-frequency attention is calculated by the following formula:

[0094]

[0095]

[0096] in, are all learnable weight matrices, used to generate queries, keys, and values, respectively. Average Pooling compresses the time dimension of the input X, preserving the overall distribution information. Linear(·) is a linear transformation function that maps the input to a new space through a weight matrix. The query, key, and value matrices of low-frequency attention are passed to the subsequent attention scoring step. is the low-frequency attention weight.

[0097] Step 206: Input the characteristic matrix corresponding to each single well into the trained prediction model to obtain the predicted gas production of the well pattern. The trained prediction model includes an adjacency matrix that characterizes the interference relationship between wells.

[0098] The prediction model to be trained includes a parameter matrix with initial values ​​set based on the static well-to-well relationship, and the adjacency matrix is ​​a learned parameter matrix corresponding to the trained prediction model.

[0099] Explanatory note: The parameter matrix in the prediction model uses the distance matrix X, set based on the static well-to-well relationships in the above example, as its initial value to represent the estimated well-to-well interference relationships at the current development stage. This matrix, combined with the characteristic matrix corresponding to each individual well, predicts the well pattern's predicted gas production. The parameter matrix corresponding to the trained prediction model is the adjacency matrix that best represents the actual well-to-well interference relationships at the current development stage.

[0100] It's important to note that the interactions between wells in a well pattern change during different development stages, necessitating retraining of the prediction model based on the time series data from that development stage. Each trained prediction model corresponds to a fixed adjacency matrix. The differences in the adjacency matrices corresponding to different development stages reveal the dynamic evolution of inter-well interference.

[0101] In one embodiment of the present specification, the parameter matrix includes a pair of a first parameter matrix and a second parameter matrix characterizing asymmetric interference between wells, and a third parameter matrix characterizing the characteristics of each well in the well pattern.

[0102] Illustratively, the first parameter matrix W1 and the second parameter matrix W2 are used to generate an antisymmetric matrix The third parameter matrix represents the characteristics of each individual well, adding a bias correction term to each well to correct for its own characteristics.

[0103] The explanatory, dynamic well-to-well relationship graph learning module utilizes an antisymmetric matrix and a correction design tailored to the characteristics of individual wells, enabling the parameter matrix to continuously and adaptively learn as the model trains. Ultimately, an adjacency matrix is ​​generated that captures the complex, asymmetric relationships between wells. A loss function is calculated to measure the deviation between the predicted and true values. This loss function is then used to perform a gradient update on the current parameter matrix (initially, the distance matrix X), thereby updating the first, second, and third parameter matrices within the parameter matrix.

[0104] Specifically, the expression of the parameter matrix is:

[0105]

[0106] Among them, ReLU is the activation function. W1 and W2 are learnable parameter matrices. is the product of the transpose of W1 and W2, is the product of the transpose of W1 and W2, β is the learnable diagonal element parameter vector, and diag(β) is the diagonal matrix of the diagonal element parameter vector β.

[0107] In one embodiment of the present specification, a method for training a prediction model includes the following steps:

[0108] Setting the initial values ​​of the parameter matrix based on the static well-to-well relationship;

[0109] The current parameter matrix and the feature matrix corresponding to each single well are input into the prediction model to be trained. The current parameter matrix is ​​updated based on the loss function of the prediction model to be trained, and this step is repeated with the updated parameter matrix as the new current parameter matrix until the loss function of the prediction model to be trained meets the convergence condition to obtain the trained prediction model.

[0110] Illustratively, the deviation between the predicted value and the true value is measured by calculating the loss function, and the current parameter matrix is ​​gradient updated by back propagation and gradient descent using the loss function, thereby training the model until the loss function of the prediction model to be trained meets the convergence condition, and the training of the prediction model is completed.

[0111] Specifically, the parameters of the prediction model are differentiated, and the gradients of the loss function with respect to the parameters are also derived for each parameter matrix in the expression of the parameter matrix. An optimizer (such as Adam) is used to simultaneously update the parameters of the prediction model and each parameter matrix of the parameter matrix.

[0112] In one embodiment of the present specification, updating the current parameter matrix based on the loss function of the prediction model to be trained includes:

[0113] Perform a preliminary update on the current parameter matrix based on the loss function of the prediction model to be trained;

[0114] Obtain the weight matrices of the two parameter matrices corresponding to the update and after through the multi-head attention mechanism;

[0115] A secondary update is performed based on the two parameter matrices and their weight matrices corresponding to each other before and after the update to obtain an updated parameter matrix.

[0116] Illustratively, the dynamic graph well relationship learning module adaptively learns the dynamic correlation strength between wells through the graph attention mechanism to capture the mutual interference effect between wells. The key technical points in this module include the aggregation strategy of multi-head attention in addition to the dynamic update of the parameter matrix. The gradient update of the parameter matrix through backpropagation and gradient descent is used as the initial update. The updated parameter matrix is , the parameter matrix before updating is ,Furthermore, a multi-head attention mechanism is used to balance the learned new parameter matrix with the old parameter matrix, including the following calculation steps:

[0117] Compute the query matrix Q, key matrix K, and value matrix V:

[0118]

[0119] in, are all learnable parameter matrices, It is the transposed matrix of the two parameter matrices before and after the update, and then multiplied by the three learnable parameter matrices to obtain the query matrix Q, key matrix K and value matrix V;

[0120] Calculate single-head attention:

[0121]

[0122] Single-head attention is the standard attention mechanism, refer to the above explanation of the attention mechanism. Multiple independent "attention heads" are executed in parallel, and each head performs independent attention calculations. Since each head uses random initialization parameters, the corresponding Different, so the multi-head attention mechanism can extract information from different angles;

[0123] Calculation of the weight matrix of the two parameter matrices corresponding to before and after the update:

[0124]

[0125] In this embodiment, the multi-head attention mechanism includes two independent "attention heads", and the attention calculation results correspond to 、 , after concatenating the two by Concat(·), the weight matrix corresponding to the updated parameter matrix is ​​generated by the MLP(·) multi-layer perceptron , the weight matrix corresponding to the parameter matrix before updating is ;

[0126] Second update of parameter matrix:

[0127]

[0128] in, is the updated parameter matrix, Represents element-wise multiplication, integrating the results of the two "attention heads" to capture richer feature information.

[0129] By integrating the dynamic well-to-well relationship graph learning module and the high-low frequency fusion spatiotemporal convolution prediction module described in the above embodiments, an adaptive spatiotemporal graph neural network model is constructed. Figure 2As shown in the figure, the root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the prediction performance of the model, and the spatiotemporal graph neural network model (AST-GNN) in the present invention is compared with multiple traditional time series prediction models (CNN, GUR, LSTM, Transformer). The experimental results show that the method proposed in the present invention is significantly superior to traditional machine learning and time series prediction models in terms of gas production prediction accuracy and multi-step prediction stability, providing a new technical approach for the efficient development of offshore gas fields.

[0130] It can be understood that the gas production is predicted by the adaptive spatiotemporal graph neural network model, and the gas production predicted in the previous step is re-input as the input for the next prediction to achieve recursive calling and self-update of the model at different time scales.

[0131] In one embodiment of this specification, intelligent production prediction further includes:

[0132] Output a visualization of the well-to-well relationship diagram based on the parameter matrix and distance matrix.

[0133] Explanatory analysis reveals that the interactions between wells in a well pattern change at different development stages, necessitating a relearning of the adjacency matrix corresponding to the current development stage. The adjacency matrices corresponding to different development stages reveal the inter-well interference relationships at each stage, and the differences in the adjacency matrices corresponding to different development stages reveal the dynamic evolution of inter-well interference.

[0134] For example, Figure 3 As shown, visualization analysis can be used to display the adjacency matrices corresponding to different development stages, thereby intuitively revealing the dynamic evolution of interference between gas wells at different development stages and fully exploring the spatiotemporal dynamic characteristics inherent in well pattern production data. Compared to other methods that focus on numerical prediction, this visualization analysis provides oil and gas field engineers with more intuitive and understandable decision-making support information.

[0135] Working principle:

[0136] Step 1: Data preprocessing.

[0137] Step 2: Feature selection: The dynamic time warping (DTW) method is used to calculate the dynamic similarity between various production parameters and gas well production, and to evaluate the dynamic correlation between different features and production.

[0138] Step 3: Build a dynamic well-to-well relationship graph learning module. Update the parameter matrix to obtain the adjacency matrix.

[0139] Step 4: Construct a high- and low-frequency fusion spatiotemporal convolution prediction module. Parallel design of high- and low-frequency self-attention structures captures local fine features and global trend features in the time series, respectively, and achieves adaptive feature fusion through gating units.

[0140] Step 5: Build an adaptive spatiotemporal graph neural network model. Feed the output of the dynamic inter-well relationship graph learning module into the high- and low-frequency fusion spatiotemporal convolution prediction module to simultaneously model the dynamic evolution of inter-well relationships and the multi-scale patterns of production time series.

[0141] Step 6: Capacity prediction: Use the trained model to predict production on the test set data.

[0142] Step 7: Visualization Analysis: Visualize the dynamic well-to-well relationships learned by the model to reveal the dynamic evolution of well-to-well interactions at different development stages.

[0143] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0144] See next Figure 4 , Figure 4 A schematic diagram of the structure of an offshore oil and gas well network production intelligent prediction system provided in an embodiment of this specification is shown.

[0145] The intelligent prediction system 400 includes a data acquisition module 401, a feature extraction module 402 and a production capacity prediction module 403;

[0146] The data acquisition module 401 acquires the time series data of the gas production and key production parameters corresponding to each well in the well pattern, as well as the static well-to-well relationship representing the relative distance between each well in the well pattern;

[0147] A feature extraction module 402 performs feature extraction based on the time series data of gas production and key production parameters corresponding to each well to obtain a feature matrix corresponding to each well;

[0148] The production capacity prediction module 403 inputs the characteristic matrix corresponding to each single well into the trained prediction model to obtain the predicted gas production of the well network. The trained prediction model includes an adjacency matrix that characterizes the interference relationship between wells. Among them, the prediction model to be trained includes a parameter matrix with initial values ​​set based on static well-to-well relationships, and the adjacency matrix is ​​the learned parameter matrix corresponding to the trained prediction model.

[0149] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referenced to each other. Each embodiment focuses on the differences from other embodiments. In particular, the intelligent yield forecasting system embodiment is generally similar to the intelligent yield forecasting method embodiment, so the description is relatively simple. For relevant parts, refer to the description of the intelligent yield forecasting method embodiment.

[0150] See also Figure 5 A schematic structural diagram of an electronic device provided in an embodiment of this specification is shown.

[0151] like Figure 5 As shown, the electronic device 500 may include: at least one processor 501 , at least one network interface 504 , a user interface 503 , a memory 505 and at least one communication bus 502 .

[0152] The communication bus 502 may be used to implement connection and communication among the above components.

[0153] The user interface 503 may include buttons, and the optional user interface may also include a standard wired interface or a wireless interface.

[0154] The network interface 504 may include, but is not limited to, a Bluetooth module, an NFC module, a Wi-Fi module, and the like.

[0155] Among them, the processor 501 may include one or more processing cores. The processor 501 uses various interfaces and lines to connect the various parts of the entire electronic device 500, and executes various functions of the electronic device 500 and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 505, and calling data stored in the memory 505. Optionally, the processor 501 can be implemented in at least one hardware form of DSP, FPGA, and PLC. The processor 501 can integrate one or a combination of CPU, GPU, modem, etc. Among them, the CPU mainly processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; and the modem is used to handle wireless communications. It is understandable that the above-mentioned modem may not be integrated into the processor 501, but may be implemented separately through a chip.

[0156] Memory 505 may include either RAM or ROM. Optionally, memory 505 may include non-transitory computer-readable media. Memory 505 may be used to store instructions, programs, codes, code sets, or instruction sets. Memory 505 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch control, sound playback, image playback, etc.), and instructions for implementing the aforementioned method embodiments. The data storage area may store data related to the aforementioned method embodiments. Memory 505 may also optionally be at least one storage device located remotely from the processor 501. Memory 505, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an intelligent yield prediction application. Processor 501 may be configured to invoke the intelligent yield prediction application stored in memory 505 and execute the steps of the intelligent yield prediction method described in the aforementioned embodiments.

[0157] The embodiments of this specification also provide a computer-readable storage medium containing instructions that, when executed on a computer or processor, cause the computer or processor to perform one or more steps of the aforementioned intelligent yield forecasting method embodiment. If the components of the aforementioned electronic device are implemented as software functional units and sold or used as independent products, they may be stored in a computer-readable storage medium.

[0158] In the above embodiments, all or part of the embodiments may be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer program product. A computer program product comprises one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to the embodiments of this specification are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted via a computer-readable storage medium. The computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. Available media may be magnetic media (eg, floppy disks, hard disks, tapes), optical media (eg, digital versatile discs (DVDs)), or semiconductor media (eg, solid state drives (SSDs)).

[0159] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When executed, the program can include the processes of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. The technical features of this embodiment and the implementation scheme can be combined in any manner unless they conflict.

[0160] The above embodiments are merely preferred embodiments of this specification and are not intended to limit the scope of this specification. Without departing from the design spirit of this specification, various modifications and improvements made to the technical solutions of this specification by ordinary technicians in this field should fall within the scope of protection determined by the claims of this specification.

Claims

1. An intelligent prediction method for offshore oil and gas well network production, characterized in that: The following steps are involved: Obtain the time series data of gas production and key production parameters corresponding to each individual well in the well network, as well as the static well-to-well relationship that characterizes the relative distance between each individual well in the well network; Based on the time series data of gas production and key production parameters corresponding to each single well, feature extraction is performed to obtain the feature matrix corresponding to each single well; Inputting the characteristic matrix corresponding to each single well into a trained prediction model to obtain the predicted gas production of the well pattern, wherein the trained prediction model includes an adjacency matrix representing the interference relationship between wells; The prediction model to be trained includes a parameter matrix with initial values ​​set based on the static well-to-well relationship, and the adjacency matrix is ​​a post-learning parameter matrix corresponding to the trained prediction model.

2. The method for intelligent prediction of offshore oil and gas well network production according to claim 1, characterized in that: The method of obtaining the time series data of the gas production and the time series data of the key production parameters corresponding to each single well in the well pattern includes: Obtain the time series data of gas production and various production parameters corresponding to each single well in the well pattern; Based on the similarity between the time series data of the gas production corresponding to each single well in the well network and the time series data of multiple production parameters, key production parameters are selected from the multiple production parameters.

3. The method for intelligent prediction of offshore oil and gas well network production according to claim 2, characterized in that: The key production parameters are selected from the multiple production parameters based on the similarity between the time series data of the gas production corresponding to each single well in the well pattern and the time series data of the multiple production parameters, including: Based on the time series data of gas production and multiple production parameters corresponding to each single well in the well pattern, the dynamic time warping method is used to obtain the similarity score ranking information of the multiple production parameters corresponding to each single well; Based on the similarity score ranking information of the multiple production parameters corresponding to all the single wells, key production parameters are selected from the multiple production parameters.

4. The method for intelligent prediction of offshore oil and gas well pattern production according to claim 1, characterized in that: The parameter matrix includes a pair of first parameter matrix and second parameter matrix characterizing asymmetric interference between wells, and a third parameter matrix characterizing the characteristics of each single well in the well network.

5. The method for intelligent prediction of offshore oil and gas well network production according to claim 1, characterized in that: The training method of the prediction model comprises the following steps: Setting the initial values ​​of the parameter matrix based on the static well-to-well relationship; The current parameter matrix and the feature matrix corresponding to each single well are input into the prediction model to be trained. The current parameter matrix is ​​updated based on the loss function of the prediction model to be trained, and this step is repeated with the updated parameter matrix as the new current parameter matrix until the loss function of the prediction model to be trained meets the convergence condition to obtain the trained prediction model.

6. The method for intelligent prediction of offshore oil and gas well pattern production according to claim 5, characterized in that: The updating of the current parameter matrix based on the loss function of the prediction model to be trained includes: Perform a preliminary update on the current parameter matrix based on the loss function of the prediction model to be trained; Obtain the weight matrices of the two parameter matrices corresponding to the update and after through the multi-head attention mechanism; A secondary update is performed based on the two parameter matrices and their weight matrices corresponding to each other before and after the update to obtain an updated parameter matrix.

7. The method for intelligent prediction of offshore oil and gas well pattern production according to claim 1, characterized in that: The feature extraction is performed based on the time series data of the gas production volume and the time series data of the key production parameters corresponding to each single well to obtain the feature matrix corresponding to each single well, including: For any single well, local fine features and global trend features are extracted based on the corresponding time series data of gas production and key production parameters; Based on the local fine features and global trend features, feature fusion is performed to obtain the feature matrix corresponding to the single well; Repeat the above steps until the characteristic matrix corresponding to each single well is obtained.

8. The method for intelligent prediction of offshore oil and gas well pattern production according to claim 7, characterized in that: For any single well, local fine features and global trend features are extracted based on the corresponding time series data of gas production and time series data of key production parameters, including: For any single well, based on the corresponding time series data of gas production and key production parameters, a sliding window method is used to extract local fine features, and a global average pooling method is used to extract global trend features.

9. The method for intelligent prediction of offshore oil and gas well pattern production according to claim 1, characterized in that: Also includes: Output a visualization of the well-to-well relationship diagram based on the parameter matrix.

10. An intelligent prediction system for offshore oil and gas well network production, characterized in that: It includes data acquisition module, feature extraction module and capacity prediction module; The data acquisition module acquires the time series data of gas production and key production parameters corresponding to each single well in the well pattern, as well as the static well-to-well relationship representing the relative distance between each single well in the well pattern; The feature extraction module extracts features based on the time series data of gas production and key production parameters corresponding to each single well, and obtains a feature matrix corresponding to each single well; The production capacity prediction module inputs the characteristic matrix corresponding to each single well into a trained prediction model to obtain the predicted gas production of the well network. The trained prediction model includes an adjacency matrix that characterizes the interference relationship between wells. The prediction model to be trained includes a parameter matrix with initial values ​​set based on static well-to-well relationships. The adjacency matrix is ​​the learned parameter matrix corresponding to the trained prediction model.

Images