Oil and gas reservoir exploration and development risk prediction method and device

Abstract

The invention relates to the field of exploration and development data processing, and discloses an oil and gas reservoir exploration and development risk prediction method and device, which can construct a normalized adjacency matrix, a node feature matrix and a high-dimensional time sequence matrix according to parameter values of a plurality of production related parameters in an oil and gas well in a target time period. And inputting the normalized adjacency matrix and the node feature matrix into a trained graph neural network for risk prediction to obtain a risk feature matrix and a first risk score corresponding to each production related parameter. And inputting the risk feature matrix and the high-dimensional time sequence matrix into a trained multivariable time sequence processing model for risk prediction to obtain a second risk score corresponding to each production related parameter. And determining the risk grade of the oil and gas well based on the first risk score and the second risk score corresponding to each production related parameter. The accuracy of oil and gas well risk grade prediction can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of exploration and development data processing, and in particular to a method and device for predicting risks in oil and gas reservoir exploration and development. Background Art

[0002] With the deepening of the digitalization process of oil and gas exploration and development, the amount and types of data generated in the exploration and development process are growing exponentially.

[0003] Faced with such a huge amount of data, how to efficiently process and analyze data, promptly identify potential risks, and quickly issue warnings and adjustments in abnormal situations has become a key challenge to ensure the safe operation of oil and gas fields and improve exploration and development efficiency.

[0004] Related technologies generally require technical personnel to set indicator thresholds based on experience, and combine the indicator thresholds and monitoring data to determine whether there is a risk, which has low accuracy. Summary of the Invention

[0005] The present invention provides a method and device for predicting the risks of oil and gas reservoir exploration and development, which is used to solve the defect in related technologies that technicians set indicator thresholds based on experience and combine the indicator thresholds with monitoring data to determine whether there is a risk, resulting in low accuracy, thereby improving the accuracy of predicting the risk level of oil and gas wells.

[0006] In a first aspect, the present invention provides a method for predicting risks in oil and gas reservoir exploration and development, comprising: Based on the parameter values of multiple production-related parameters in oil and gas wells within the target period, a normalized adjacency matrix, a node feature matrix and a high-dimensional time series matrix are constructed; Inputting the normalized adjacency matrix and the node feature matrix into a trained graph neural network for risk prediction, thereby obtaining a risk feature matrix and a first risk score corresponding to each of the production-related parameters; wherein each row of data in the risk feature matrix is risk feature data corresponding to each of the production-related parameters; Inputting the risk feature matrix and the high-dimensional time series matrix into a trained multivariate time series processing model for risk prediction to obtain a second risk score corresponding to each of the production-related parameters; The risk level of the oil and gas well is determined based on the first risk score and the second risk score corresponding to each of the production-related parameters.

[0007] Optionally, a normalized adjacency matrix and a node feature matrix are constructed based on the parameter values of multiple production-related parameters in the oil and gas wells within the target period, including: determining, based on the parameter value of each production-related parameter within the target time period, a correlation between any two of the production-related parameters; Taking each of the production-related parameters as a node, taking each of the correlations as an edge between two corresponding nodes, constructing a corresponding adjacency matrix according to each of the nodes and each of the edges, and normalizing the adjacency matrix to obtain the normalized adjacency matrix; Determining a statistical indicator sequence corresponding to each of the production-related parameters according to the parameter value of each of the production-related parameters within the target time period; The statistical indicator sequence corresponding to each of the production-related parameters is used as the first row of data for constructing a matrix, and the first row of data corresponding to each of the production-related parameters is arranged based on a set parameter order to construct the node feature matrix.

[0008] Optionally, the graph neural network includes a graph convolutional network and a graph attention network; Inputting the normalized adjacency matrix and the node feature matrix into a trained graph neural network for risk prediction to obtain a risk feature matrix and a first risk score corresponding to each of the production-related parameters includes: Inputting the normalized adjacency matrix and the node feature matrix into the graph convolutional network, so that the graph convolution performs neighborhood aggregation on the first row of data corresponding to each of the production-related parameters in the node feature matrix based on the normalized adjacency matrix, to obtain first aggregated row data corresponding to each of the production-related parameters and a first aggregated feature matrix, wherein the first aggregated feature matrix includes the first aggregated row data corresponding to each of the production-related parameters; The graph attention network is used to perform risk prediction on the first aggregated feature matrix to obtain the risk feature matrix and a first risk score corresponding to each of the production-related parameters.

[0009] Optionally, using the graph attention network to perform risk prediction on the first aggregated feature matrix to obtain the risk feature matrix and a first risk score corresponding to each of the production-related parameters includes: The first aggregated feature matrix is input into the graph attention network so that the graph attention network: based on the attention mechanism, performs neighborhood aggregation on the first aggregated row data corresponding to each of the production-related parameters in the first aggregated feature matrix to obtain second aggregated row data and a second aggregated feature matrix corresponding to each of the production-related parameters, the second aggregated feature matrix includes the second aggregated row data corresponding to each of the production-related parameters, each of the second aggregated row data in the second aggregated feature matrix is used as the risk feature data, and the second aggregated feature matrix is used as the risk feature matrix, and feature mapping is performed on each of the risk feature data in the risk feature matrix to obtain the first risk score corresponding to each of the production-related parameters.

[0010] Optionally, the target period includes multiple time steps; a high-dimensional time series matrix is constructed based on parameter values of multiple production-related parameters in the oil and gas wells within the target period, including: For any of the production-related parameters, sorting the parameter values of each time step of the production-related parameter in the target period in chronological order to obtain a parameter value time series of the production-related parameter; Using the parameter value time series of each production-related parameter as a column data for constructing a matrix; Arranging the column data corresponding to each of the production-related parameters based on a set parameter order to construct a target matrix; Key features are extracted from each row of the target matrix to obtain the high-dimensional time series matrix, where each row of the high-dimensional time series matrix is key feature data extracted from each row of the target matrix.

[0011] Optionally, inputting the risk feature matrix and the high-dimensional time series matrix into a trained multivariate time series processing model for risk prediction to obtain a second risk score corresponding to each of the production-related parameters includes: The risk feature matrix and the high-dimensional time series matrix are input into a trained multivariate time series processing model so that the multivariate time series processing model: transposes the high-dimensional time series matrix to obtain a transposed matrix, splices the row data with equal row order in the transposed matrix and the risk feature matrix to obtain feature spliced data corresponding to each of the production-related parameters, performs risk prediction on the feature spliced data corresponding to each of the production-related parameters, and obtains a second risk score corresponding to each of the production-related parameters.

[0012] Optionally, determining the risk level of the oil and gas well based on the first risk score and the second risk score corresponding to each of the production-related parameters includes: Obtaining a set weight corresponding to each of the production-related parameters and a risk score interval corresponding to a plurality of set risk levels; Based on the set weight corresponding to each of the production-related parameters, a weighted sum of the first risk score and the second risk score corresponding to each of the production-related parameters is performed to obtain a global risk score; According to the global risk score, a matching target score interval is determined in the risk score intervals corresponding to the multiple set risk levels, and the set risk level corresponding to the target score interval is determined as the risk level of the oil and gas well.

[0013] Optionally, determining the risk level of the oil and gas well based on the first risk score and the second risk score corresponding to each of the production-related parameters includes: Inputting the first risk score corresponding to each of the production-related parameters into a trained classification model for risk level classification, thereby obtaining a first risk probability distribution output by the classification model; and inputting the second risk score corresponding to each of the production-related parameters into the classification model for risk level classification, thereby obtaining a second risk probability distribution output by the classification model; Obtaining a first set weight corresponding to the graph neural network and a second set weight corresponding to the multivariate time series processing model; Based on the first set weight and the second set weight, performing a weighted summation on the first risk probability distribution and the second risk probability distribution to obtain a final risk probability distribution; The risk level of the oil and gas well is determined according to the final risk probability distribution.

[0014] Optionally, when the first risk probability distribution and the second risk probability distribution include a high risk probability, a medium risk probability, and a low risk probability, the final risk probability distribution includes a high risk probability final value, a medium risk probability final value, and a low risk probability final value; The step of performing weighted summation of the first risk probability distribution and the second risk probability distribution based on the first set weight and the second set weight to obtain a final risk probability distribution includes: Based on the first set weight and the second set weight, performing a weighted summation of the high risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain the high risk probability final value, performing a weighted summation of the medium risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain the medium risk probability final value, and performing a weighted summation of the low risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain the low risk probability final value; Determining the risk level of the oil and gas well according to the final risk probability distribution includes: A maximum probability final value is determined among the high risk probability final value, the medium risk probability final value, and the low risk probability final value, and the risk level corresponding to the maximum probability final value is determined as the risk level of the oil and gas well.

[0015] In a second aspect, the present invention provides a device for predicting risks in oil and gas reservoir exploration and development, comprising: A construction unit, used to construct a normalized adjacency matrix, a node feature matrix and a high-dimensional time series matrix according to parameter values of multiple production-related parameters in the oil and gas wells within a target period; A first prediction unit is configured to input the normalized adjacency matrix and the node feature matrix into a trained graph neural network for risk prediction, thereby obtaining a risk feature matrix and a first risk score corresponding to each of the production-related parameters; wherein each row of data in the risk feature matrix is risk feature data corresponding to each of the production-related parameters; A second prediction unit is configured to input the risk feature matrix and the high-dimensional time series matrix into a trained multivariate time series processing model to perform risk prediction, thereby obtaining a second risk score corresponding to each of the production-related parameters; A determination unit is configured to determine a risk level of the oil and gas well based on a first risk score and a second risk score corresponding to each of the production-related parameters.

[0016] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the method for predicting oil and gas reservoir exploration and development risks according to the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0017] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the oil and gas reservoir exploration and development risk prediction method of the above-mentioned first aspect or any corresponding embodiment thereof.

[0018] The present invention provides a method and device for predicting oil and gas reservoir exploration and development risks. Based on the values of multiple production-related parameters in an oil and gas well within a target time period, a normalized adjacency matrix, a node feature matrix, and a high-dimensional time series matrix are constructed. The normalized adjacency matrix and the node feature matrix are input into a trained graph neural network for risk prediction, resulting in a risk feature matrix and a first risk score corresponding to each production-related parameter. Each row of data in the risk feature matrix represents risk feature data corresponding to each production-related parameter. The risk feature matrix and the high-dimensional time series matrix are input into a trained multivariate time series processing model for risk prediction, resulting in a second risk score corresponding to each production-related parameter. Based on the first and second risk scores corresponding to each production-related parameter, the risk level of the oil and gas well is determined. Using the trained graph neural network and multivariate time series processing model, the present invention mines data associations and time series features between multiple production-related parameters, accurately extracting complex relationships between multiple production-related parameter data and improving the ability to capture dynamic risks. The risk level of the oil and gas well is ultimately determined based on the risk scores output by the graph neural network and multivariate time series processing model, effectively improving the accuracy of risk level prediction for the oil and gas well. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or related technologies, the following is a brief introduction to the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are 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.

[0020] Figure 1 A flow chart of a method for predicting risks in oil and gas reservoir exploration and development provided by an embodiment of the present invention; Figure 2 A flowchart of another method for predicting risks in oil and gas reservoir exploration and development provided by an embodiment of the present invention; Figure 3 A schematic structural diagram of a device for predicting risks in oil and gas reservoir exploration and development provided by an embodiment of the present invention; Figure 4 A schematic structural diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0022] The following combination Figure 1-Figure 2 The present invention describes the method for predicting risks in oil and gas reservoir exploration and development.

[0023] like Figure 1 As shown, this embodiment proposes a first method for predicting risks in oil and gas reservoir exploration and development, which may include the following steps: S101. Construct a normalized adjacency matrix, a node feature matrix, and a high-dimensional time series matrix based on parameter values of multiple production-related parameters in oil and gas wells within a target period.

[0024] Among them, oil and gas wells may be production wells that require risk assessment.

[0025] Specifically, production-related parameters can be physical parameters related to production during the exploration and development of oil and gas wells. These can include key downhole parameters such as bottomhole pressure, temperature, and flow rate, as well as surface production parameters such as output, equipment status, and energy consumption.

[0026] The target period may be a certain production period.

[0027] Specifically, this embodiment can construct a normalized adjacency matrix, a node feature matrix, and a high-dimensional time series matrix according to multiple production-related parameter values in the oil and gas wells within the target period.

[0028] Optionally, a normalized adjacency matrix and a node feature matrix are constructed based on the parameter values of multiple production-related parameters in the oil and gas wells within the target period, including: Determine the correlation between any two production-related parameters based on the parameter value of each production-related parameter within the target period; Each production-related parameter is regarded as a node, and each correlation degree is regarded as an edge between two corresponding nodes. A corresponding adjacency matrix is constructed according to each node and each edge, and the adjacency matrix is normalized to obtain a normalized adjacency matrix. Determine the statistical indicator sequence corresponding to each production-related parameter based on the parameter value of each production-related parameter within the target period; The statistical indicator sequence corresponding to each production-related parameter is used as the first row of data for constructing the matrix, and the first row of data corresponding to each production-related parameter is arranged based on the set parameter order to construct the node feature matrix.

[0029] In this embodiment, a statistical indicator sequence corresponding to the production-related parameters can be constructed based on the parameter values of the production-related parameters at multiple time steps within the target period.

[0030] Specifically, this embodiment can determine statistical values such as the minimum value, maximum value, average value, current step parameter value and standard deviation among the parameter values of the production-related parameters in multiple time steps within the target period, and arrange the determined statistical values in a certain order to obtain a statistical indicator sequence corresponding to the production-related parameters.

[0031] It can be understood that the node feature matrix is obtained by arranging the first row of data corresponding to multiple production-related parameters, and the row data in the node feature matrix is the first row of data corresponding to the production-related parameters.

[0032] Optionally, the target period includes multiple time steps. Based on the parameter values of multiple production-related parameters in the oil and gas wells within the target period, a high-dimensional time series matrix is constructed, including: For any production-related parameter, sort the parameter values of each time step of the production-related parameter in the target period in chronological order to obtain the parameter value time series of the production-related parameter; The parameter value time series of each production-related parameter is used as a column data for constructing the matrix; Arrange the column data corresponding to each production-related parameter based on the set parameter order to construct a target matrix; Key features are extracted from each row of the target matrix to obtain a high-dimensional time series matrix. Each row of the high-dimensional time series matrix is key feature data extracted from each row of the target matrix.

[0033] It can be understood that the target matrix is formed by arranging column data corresponding to multiple production-related parameters, and the column data in the target matrix is the column data corresponding to the production-related parameters.

[0034] Specifically, this embodiment can extract key features from each row of data in the target matrix, and arrange the key feature data extracted from each row of data in the order of the corresponding row data to obtain a high-dimensional time series matrix.

[0035] Specifically, for any row of data in the target matrix, this embodiment can perform seasonal trend decomposition, empirical mode decomposition, and singular spectrum analysis on the row of data, decompose the row of data into trend term, seasonal term, and residual term features, and use the whole as key feature data.

[0036] S102. Input the normalized adjacency matrix and the node feature matrix into the trained graph neural network for risk prediction, and obtain a risk feature matrix and a first risk score corresponding to each production-related parameter; wherein each row of data in the risk feature matrix is the risk feature data corresponding to each production-related parameter.

[0037] It should be noted that, in this embodiment, sample training data can be used to train the graph neural network to be trained until a trained graph neural network is obtained.

[0038] In this embodiment, sample training data can be first constructed based on the parameter values of multiple production-related parameters in oil and gas wells over a historical period. Specifically, a sample normalized adjacency matrix and a sample node feature matrix can be constructed based on the parameter values of multiple production-related parameters in oil and gas wells over a historical period. The sample normalized adjacency matrix and the sample node feature matrix are used as sample training data and input into a pre-trained network for risk prediction to obtain a prediction result. Based on the difference between the prediction result and the corresponding actual result, the parameters of the pre-trained network are updated until a trained graph neural network is obtained.

[0039] Optionally, the graph neural network includes a graph convolutional network and a graph attention network. Step S102 may include: Input the normalized adjacency matrix and the node feature matrix into the graph convolution network, so that the graph convolution performs neighborhood aggregation on the first row of data corresponding to each production-related parameter in the node feature matrix based on the normalized adjacency matrix, and obtains the first aggregated row data corresponding to each production-related parameter and the first aggregated feature matrix, wherein the first aggregated feature matrix includes the first aggregated row data corresponding to each production-related parameter; The graph attention network is used to perform risk prediction on the first aggregated feature matrix to obtain the risk feature matrix and the first risk score corresponding to each production-related parameter.

[0040] Optionally, the graph attention network is used to perform risk prediction on the first aggregated feature matrix to obtain the risk feature matrix and a first risk score corresponding to each production-related parameter, including: The first aggregate feature matrix is input into the graph attention network so that the graph attention network: based on the attention mechanism, the first aggregate row data corresponding to each production-related parameter in the first aggregate feature matrix is neighborhood aggregated to obtain the second aggregate row data and the second aggregate feature matrix corresponding to each production-related parameter, the second aggregate feature matrix includes the second aggregate row data corresponding to each production-related parameter, each second aggregate row data in the second aggregate feature matrix is used as risk feature data, and the second aggregate feature matrix is used as a risk feature matrix, feature mapping is performed on each risk feature data in the risk feature matrix, and the first risk score corresponding to each production-related parameter is obtained.

[0041] It should be noted that graph neural networks can extract potential interactions between multiple production-related parameters and reveal the influence paths between different physical variables.

[0042] S103: Input the risk feature matrix and the high-dimensional time series matrix into the trained multivariate time series processing model to perform risk prediction, and obtain a second risk score corresponding to each production-related parameter.

[0043] It should be noted that, in this embodiment, the pre-training model can be trained using sample training data until a trained multivariate time series processing model is obtained.

[0044] Specifically, the multivariate time series processing model may be an iTransformer model or other models, which is not limited in this embodiment.

[0045] Among them, this embodiment can construct a sample high-dimensional time series matrix based on the parameter values of multiple production-related parameters at multiple time steps within the target time period, and obtain the sample risk feature data of the corresponding time period, and input the sample high-dimensional time series matrix and the sample risk feature data into the pre-trained multivariate time series processing model for risk prediction to obtain the prediction result, and update the parameters of the pre-trained multivariate time series processing model according to the difference between the prediction result and the corresponding actual result until a trained multivariate time series processing model is obtained.

[0046] Specifically, after obtaining the risk feature matrix output by the graph neural network, this embodiment can input the risk feature matrix and the constructed high-dimensional time series matrix into a trained multivariate time series processing model for risk prediction, and obtain a second risk score corresponding to each generation-related parameter output by the model.

[0047] Optionally, step S103 may include: The risk feature matrix and the high-dimensional time series matrix are input into the trained multivariate time series processing model so that the multivariate time series processing model: transposes the high-dimensional time series matrix to obtain a transposed matrix, splices the row data with equal row order in the transposed matrix and the risk feature matrix to obtain feature splicing data corresponding to each production-related parameter, performs risk prediction on the feature splicing data corresponding to each production-related parameter, and obtains a second risk score corresponding to each production-related parameter.

[0048] Specifically, this embodiment transposes the high-dimensional time series matrix, that is, transforms the row data in the high-dimensional time series matrix into column data, and the column data in the transposed matrix is the row data in the high-dimensional time series matrix.

[0049] It should be noted that the input data of the multivariate time series processing model includes A high-dimensional time series matrix, where T represents the time step, DRepresents the dimension of the observed variable (i.e., the number of multivariables), and each time step corresponds to a D-dimensional variable vector. The evolution trend of each variable at all time steps is encoded as a token vector, forming a token sequence of length D that is input into the encoder of the multivariate time series processing model. During the encoding process, the multivariate time series processing model can automatically model the interdependence between variables through the self-attention mechanism and dynamically assign attention weights according to the importance of the variables. After processing by the encoding layer, the model output is a set of representation vector sequences that integrate contextual dependencies and structural relationships between variables, in the form of ,in d Represents the embedding dimension of each variable. This output can be further used in downstream tasks such as risk prediction, anomaly detection, or classification as high-level input features for decision models.

[0050] It should be noted that the multivariate time series data processing model can extract implicit correlations among multiple generation-related parameters.

[0051] S104: Determine the risk level of the oil and gas well based on the first risk score and the second risk score corresponding to each production-related parameter.

[0052] Among them, the risk level of oil and gas wells can be high risk, medium risk or low risk.

[0053] Specifically, this embodiment can evaluate the risk level of the oil and gas well based on the first risk score corresponding to each production-related parameter and the second risk score corresponding to each production-related parameter.

[0054] Optionally, step S104 may include: Obtain the set weight corresponding to each production-related parameter, as well as the risk score ranges corresponding to multiple set risk levels; Based on the set weight corresponding to each production-related parameter, the first risk score and the second risk score corresponding to each production-related parameter are weightedly summed to obtain a global risk score; According to the global risk score, a matching target score interval is determined in the risk score intervals corresponding to the multiple set risk levels, and the set risk level corresponding to the target score interval is determined as the risk level of the oil and gas well.

[0055] Among them, the setting weights corresponding to different production-related parameters can be different or the same.

[0056] Specifically, the risk level may be set to high risk, medium risk, and low risk. In this embodiment, the technical personnel may set the risk score intervals corresponding to high risk, medium risk, and low risk according to actual conditions.

[0057] Specifically, this embodiment may perform a weighted summation of the risk scores corresponding to all generation-related parameters based on the set weights corresponding to all production-related parameters to obtain a global risk score.

[0058] The oil and gas reservoir exploration and development risk prediction method proposed in this embodiment constructs a normalized adjacency matrix, a node feature matrix, and a high-dimensional time series matrix based on the parameter values of multiple production-related parameters in an oil and gas well within a target time period. The normalized adjacency matrix and the node feature matrix are input into a trained graph neural network for risk prediction, resulting in a risk feature matrix and a first risk score corresponding to each production-related parameter. Each row of data in the risk feature matrix represents risk feature data corresponding to each production-related parameter. The risk feature matrix and the high-dimensional time series matrix are input into a trained multivariate time series processing model for risk prediction, resulting in a second risk score corresponding to each production-related parameter. Based on the first and second risk scores corresponding to each production-related parameter, the risk level of the oil and gas well is determined. This embodiment utilizes the trained graph neural network and multivariate time series processing model to mine data associations and time series features between multiple production-related parameters, accurately extracting complex relationships between multiple production-related parameter data and improving the ability to capture dynamic risks. The risk level of the oil and gas well is ultimately determined based on the risk scores output by the graph neural network and multivariate time series processing model, effectively improving the accuracy of risk level prediction for the oil and gas well.

[0059] based on Figure 1 This embodiment proposes a second method for predicting risks in oil and gas reservoir exploration and development. Step S104 may include: Inputting the first risk score corresponding to each production-related parameter into the trained classification model for risk level classification, thereby obtaining a first risk probability distribution output by the classification model; and inputting the second risk score corresponding to each production-related parameter into the classification model for risk level classification, thereby obtaining a second risk probability distribution output by the classification model; Obtaining a first set weight corresponding to the graph neural network and a second set weight corresponding to the multivariate time series processing model; Based on the first set weight and the second set weight, performing a weighted summation on the first risk probability distribution and the second risk probability distribution to obtain a final risk probability distribution; Determine the risk level of the oil and gas well based on the final risk probability distribution.

[0060] It should be noted that this embodiment can construct sample training data and use the sample training data to train a pre-trained classification model until a trained classification model is obtained. Specifically, this embodiment can obtain a sample risk score corresponding to each production-related parameter, input the sample risk score as sample training data into the pre-trained classification model for training, obtain a classification result output by the pre-trained classification model, and update the parameters in the pre-trained classification model based on the difference between the classification result and the corresponding true result until a trained classification model is obtained.

[0061] Optionally, when the first risk probability distribution and the second risk probability distribution include high risk probability, medium risk probability and low risk probability, the final risk probability distribution includes a high risk probability final value, a medium risk probability final value and a low risk probability final value.

[0062] At this time, based on the first set weight and the second set weight, the first risk probability distribution and the second risk probability distribution are weighted and summed to obtain the final risk probability distribution, including: Based on the first set weight and the second set weight, performing a weighted summation of the high risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain a high risk probability final value, performing a weighted summation of the medium risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain a medium risk probability final value, and performing a weighted summation of the low risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain a low risk probability final value; The above-mentioned risk level of the oil and gas well is determined based on the final risk probability distribution, including: The maximum probability final value is determined among the high risk probability final value, the medium risk probability final value and the low risk probability final value, and the risk level corresponding to the maximum probability final value is determined as the risk level of the oil and gas well.

[0063] It is understood that when the high risk probability final value is the maximum probability final value, this embodiment can determine the risk level of the oil and gas well as high risk. When the medium risk probability final value is the maximum probability final value, this embodiment can determine the risk level of the oil and gas well as medium risk. When the low risk probability final value is the maximum probability final value, this embodiment can determine the risk level of the oil and gas well as low risk.

[0064] The oil and gas reservoir exploration and development risk prediction method proposed in this embodiment can determine the risk level of oil and gas wells based on the risk scores corresponding to different generation-related parameters, as well as the weights corresponding to the graph neural network and the multivariate time series processing model, thereby effectively realizing the prediction of the risk level of oil and gas wells.

[0065] With the deepening digitalization of oil and gas exploration and development, the volume and variety of data generated across drilling, logging, production, and other processes are growing exponentially. This data not only encompasses multiple measurement dimensions but also exhibits complex time series characteristics, including long-term dependencies, nonstationarity, and nonlinear evolution. Faced with such a massive data volume and complex data characteristics, efficient data processing and analysis, timely identification of potential risks, and rapid early warning and adjustment in abnormal situations have become key challenges in ensuring the safe operation of oil and gas fields and improving exploration and development efficiency. Static monitoring methods and simple statistical analysis approaches used in related technologies are unable to meet the current demands of processing high-dimensional, multi-source, and time-varying data. Therefore, developing advanced intelligent data analysis methods to more accurately understand the correlations between multivariate data, capture dynamic evolution trends, and effectively warn of potential abnormal events has become a key research direction in the oil and gas exploration and development industry.

[0066] Related technologies for analyzing time series data in oil and gas exploration and development primarily rely on the following approaches, but each has limitations. First, fixed statistical indicator prediction methods primarily rely on statistical features such as the mean and standard deviation. However, these methods are only applicable when the data distribution is relatively stable and cannot fully characterize the dynamic changes in the data, making it difficult to capture non-stationary characteristics and complex temporal evolution patterns. Second, while machine learning methods in related technologies (such as decision trees and support vector machines) can extract certain patterns from the data, they lack the ability to effectively model the long-range dependencies and cyclical characteristics of time series data. This is particularly true when dealing with multidimensional, nonlinear data, and they are prone to misjudgment or omission. Furthermore, single time series analysis methods (such as autoregressive moving average models and exponential smoothing) perform well when processing stationary data. However, their modeling capabilities are significantly limited when dealing with complex environments with multiple variables, heterogeneous data, and sudden changes, making them difficult to effectively adapt to dynamically changing time series characteristics.

[0067] like Figure 2 As shown, to address these issues, this embodiment proposes a method that combines the multivariate time series processing model iTransformer with a graph neural network (GNN) to overcome the shortcomings of related technologies. This embodiment not only fully utilizes the advantages of iTransformer in time series data modeling, but also leverages the powerful graph structure modeling capabilities of GNN to more accurately extract complex relationships between multivariate data and improve the ability to capture dynamic risks, thereby providing a more efficient and accurate intelligent monitoring and early warning solution in the field of oil and gas exploration and development.

[0068] Among the other oil and gas reservoir exploration and development risk prediction methods proposed in this embodiment, this embodiment leverages sensor networks, IoT technology, and intelligent monitoring systems at the real-time data acquisition layer to achieve real-time collection of key downhole parameters (such as bottomhole pressure, temperature, and flow rate) and surface production data (such as output, equipment status, and energy consumption). This data is then initially processed using edge computing technology to reduce latency and improve data transmission and processing efficiency. Furthermore, the system integrates a geological monitoring module that acquires formation pressure, geostress distribution, and microseismic monitoring data, providing high-quality input data for reservoir evaluation and production optimization.

[0069] At the historical data integration layer, this embodiment extracts historical drilling, production, and geological data from distributed databases, cloud storage systems, and various data warehouses, and performs preprocessing such as data format standardization, deduplication, and cleansing. To ensure data time alignment, the system uses time series alignment algorithms (such as dynamic time warping and multi-scale time alignment) to synchronize data from different data sources, ensuring good temporal consistency and cross-source matching capabilities. Furthermore, to support offline training and online inference of large-scale data, this embodiment has designed a tiered storage strategy that intelligently divides cold and hot data based on data usage frequency, thereby optimizing storage resources and improving data retrieval and call efficiency.

[0070] During oil and gas exploration and development, data quality directly impacts the predictive accuracy and risk identification capabilities of subsequent models. Due to diverse data sources, the measurement process is often accompanied by missing values, outliers, noise, and redundant information. If unaddressed, these issues can lead to model misjudgments or unstable predictions. This embodiment utilizes intelligent time series data preprocessing and feature extraction modules to improve data quality, providing high-quality input data for subsequent modeling.

[0071] During the data cleaning and completion phase, this embodiment employs a variety of data repair strategies, such as interpolation algorithms (linear and spline interpolation), mean filling, and K-nearest neighbor filling, to automatically identify and repair missing data. The system also incorporates an anomaly detection mechanism to automatically screen for data mutations and dynamically adjusts data completion strategies based on historical data trends. Furthermore, to address the repetitive nature of time series data, the system utilizes hash matching and deduplication algorithms to remove redundant data, ensuring data uniqueness and timeliness.

[0072] In the time series decomposition and feature engineering stage, this embodiment can use seasonal trend decomposition, empirical mode decomposition, singular spectrum analysis and other methods to decompose complex time series data into trend terms, seasonal terms and residual terms, thereby helping to understand the long-term evolution pattern and periodic change characteristics of the data. At the same time, the system extracts multiple key features, including the rate of change of data, fluctuation amplitude, periodic components, mutation points, time lag correlation, etc., thereby enhancing the model's ability to analyze nonlinear and non-stationary data. In addition, this embodiment supports automatic feature extraction based on self-supervised learning, and performs deep feature learning on time series through convolutional neural networks and long short-term memory networks, improving data expression capabilities and providing richer input information for subsequent modeling.

[0073] In response to the complex, multidimensional, and heterogeneous time series data in the oil and gas exploration and development process, this embodiment proposes a predictive modeling method that combines iTransformer and graph neural networks to overcome the shortcomings of related technologies in time series modeling methods in terms of nonlinearity, long-term dependence, and complex correlation modeling.

[0074] In the iTransformer design, this embodiment adopts a modeling approach that uses variables as tokens, enabling the iTransformer architecture to better handle high-dimensional time series data and enhancing the model's ability to capture the relationships between multiple variables. By introducing an attention mechanism, iTransformer automatically assigns weights to different features, enabling the model to focus on key variables and ignore irrelevant information. Furthermore, this embodiment optimizes the time complexity of the iTransformer architecture, making it more efficient when processing very long time series data and reducing computational costs.

[0075] In graph neural network design, this embodiment can combine graph convolutional networks with graph attention networks to construct graph-structured relationships between data sources, device monitoring points, or sensors. This approach can automatically learn the coupling relationships between variables and leverage neighborhood information to enhance the perception of system status, thereby improving the accuracy of risk prediction.

[0076] To ensure the stability and reliability of the prediction, this embodiment can adopt a fusion strategy, combining weighted fusion and voting mechanisms, to dynamically fuse the prediction results of iTransformer and GNN to improve the robustness of the overall prediction and make it more adaptable to the mutations and non-stationarity in the oil and gas production process.

[0077] To improve the safety and production efficiency of oil and gas exploration and development, this embodiment can set up an intelligent real-time decision support and feedback adaptation module. This module can monitor key production parameters in real time and dynamically adjust operation plans and risk response strategies based on the results of the prediction model.

[0078] In terms of real-time early warning mechanisms, this embodiment employs a multi-level alarm strategy. When key parameters exceed set thresholds, the system automatically triggers warning signals and provides detailed risk analysis reports. The early warning system integrates expert knowledge and historical data trends to support personalized risk threshold settings, improving the accuracy of early warnings.

[0079] In terms of feedback loops and adaptive adjustments, this embodiment incorporates a reinforcement learning mechanism, enabling the system to continuously optimize model parameters and risk thresholds over the long term. By combining historical prediction errors with new data feedback, the system can dynamically adjust threshold settings, improving anomaly detection accuracy, reducing false positives and missed negatives, and thus enhancing overall risk management capabilities.

[0080] This embodiment builds a complete system integration and visualization module that supports efficient collaboration in data processing, risk prediction, and result presentation. The system utilizes a collaborative offline and online processing architecture. The offline component focuses on batch processing of large-scale data and model training, while the online component focuses on real-time monitoring and predictive reasoning, enabling efficient data-driven decision support.

[0081] This embodiment proposes an efficient and intelligent data acquisition and integration module to meet the needs of accurate management and efficient processing of massive, multi-source, and complex time-series data during oil and gas exploration and development. With the accelerated digitalization of oil and gas fields, the types and volumes of data collected by downhole and surface sensors are constantly increasing, including drilling parameters, logging data, production indicators, equipment operating status, and geological monitoring information. This data not only has the characteristics of long time series, non-stationary and nonlinearity, but may also involve multiple data sources and different sampling frequencies. Therefore, achieving accurate data acquisition, storage, and integration is the key to improving the performance of prediction models.

[0082] In other oil and gas reservoir exploration and development risk prediction methods proposed in this embodiment, the following steps may be included: Step 1: Data collection and preprocessing.

[0083] 1) Data source: The data preprocessing module processes the multi-dimensional, heterogeneous data collected by the downhole sensor network and the surface monitoring system in real time to ensure the accuracy and stability of subsequent modeling and analysis.

[0084] 2) Missing value processing: Since these data come from a wide range of sources, have different collection frequencies, and may contain outliers, missing values, or duplicate data, in order to improve data quality, this embodiment first uses outlier detection to eliminate invalid or noisy data, and then fills in missing values through linear interpolation or other interpolation algorithms to ensure data continuity.

[0085] 3) Duplicate data processing: At the same time, duplicate data removal is adopted to eliminate data duplication caused by redundant acquisition of multiple sensors or system errors, thereby reducing data redundancy and improving computing efficiency.

[0086] This embodiment can be widely applied across all aspects of oil and gas exploration and development. It aims to construct an intelligent risk prediction model based on time-series data analysis by comprehensively analyzing drilling parameters, production data, and geological monitoring data. This model fully utilizes the multi-source, heterogeneous data collected during the oil and gas drilling process and, combined with advanced machine learning and deep learning techniques, deeply explores and models the changing trends of various key parameters, thereby improving the reliability and scientific nature of the prediction results.

[0087] In terms of technical architecture, this embodiment can adopt a combination of iTransformer and graph neural network to give full play to the advantages of both in time series data processing and complex relationship modeling. Among them, iTransformer accurately captures the dynamic changes of drilling parameters, production indicators and geological information in the time dimension through self-attention mechanism and multivariate time series modeling technology, thereby identifying the nonlinear coupling relationship between key variables and enhancing the model's perception of long-term trends and short-term abnormal fluctuations. Graph neural network, by constructing a graph structure between variables, uses graph convolutional networks and graph attention mechanisms to further explore the potential dependencies between different data sources, learn the complex interaction patterns between physical quantities such as drilling parameters, downhole temperature, formation pressure, fluid flow rate, and form a more accurate risk assessment framework.

[0088] The fusion technology proposed in this embodiment can effectively improve the accuracy and stability of risk prediction, giving the system stronger real-time monitoring and early warning capabilities. When certain parameters fluctuate abnormally during drilling operations, the intelligent system can quickly analyze data trends, predict potential risks, and provide feasible response strategies to help operators take timely adjustment measures, thereby reducing the risk of equipment damage and improving the safety of downhole operations. In addition, based on an adaptive learning mechanism, the system can continuously optimize the prediction model as data accumulates, making it more consistent with the characteristics of different reservoir types, drilling environments, and production conditions, providing more intelligent and precise technical support for oil and gas exploration and development.

[0089] This embodiment uses a data preprocessing module to process the multi-dimensional, heterogeneous data collected by the downhole sensor network and the ground monitoring system in real time to ensure the accuracy and stability of subsequent modeling and analysis. The system integrates various types of sensor equipment distributed underground and on the ground, including drilling parameter monitors, formation pressure sensors, fluid flow rate measuring devices, temperature sensors, and other environmental monitoring equipment, thereby achieving a full range of perception of the drilling process and oil and gas production status. Since these data come from a wide range of sources, have different collection frequencies, and may contain outliers, missing values, or duplicate data, in order to improve data quality, this embodiment first uses an outlier detection method to eliminate invalid or noisy data, and fills in missing values through linear interpolation or other interpolation algorithms to ensure data continuity. At the same time, the method of removing duplicate data is used to eliminate data duplication caused by redundant collection of multiple sensors or system errors, thereby reducing data redundancy and improving computing efficiency.

[0090] After completing data cleaning, this embodiment further standardizes the data, mapping data of different dimensions and units to a unified numerical range to ensure that the comparison and calculation of different physical parameters are not affected by the dimension. For example, the numerical ranges and changing trends of physical quantities such as drilling torque, formation pressure, and wellhead temperature vary. If these are directly input into the model for calculation, the impact of certain variables on the prediction results may be amplified or weakened. Therefore, through standardization, the data distribution is made more consistent, which facilitates the training and reasoning of subsequent deep learning models and improves the stability and generalization ability of the model.

[0091] High-quality data processed by the data preprocessing module is fed into the iTransformer and graph neural networks, respectively, for intelligent risk prediction. iTransformer analyzes the temporal trends of various drilling and production parameters based on time series modeling. Its multi-head self-attention mechanism mines long-term dependencies between variables to identify potential fault signals or abnormal patterns. Meanwhile, the graph neural network constructs a graph structure between different monitoring variables, learns the physical and statistical relationships between them, and extracts high-order correlation features through graph convolution operations, further enhancing the accuracy of risk prediction and its ability to model complex nonlinear relationships.

[0092] Step 2: Time Series Feature Extraction and Modeling This step enables the system to extract and model time series data features, facilitating subsequent risk prediction.

[0093] 1) Graph Neural Network (GNN): Based on the physical and statistical relationships between drilling, production, and geological monitoring data variables, a graph structure is constructed between data sources. When drilling deep into highly permeable formations, changes in formation pressure can have a nonlinear impact on production rates, while adjustments to mud flow rates can further affect permeability. Using a graph-based message passing mechanism, GNNs can automatically extract potential interactions between these variables, revealing the influence paths between different physical variables.

[0094] 2) iTransformer, a multivariate time series processing model: Data variables processed by the GNN are input into iTransformer. iTransformer treats multi-dimensional data as independent tokens. Through its self-attention mechanism, iTransformer is able to capture the temporal interactions and trend changes between these variables. iTransformer not only captures drilling pressure fluctuations and changes in wellhead temperature sensor data, but also monitors changes in drill string torque, abnormal downhole fluid flow rates, formation permeability fluctuations, changes in wellhead vibration frequency, trends in drill bit temperature increases, changes in mud density and viscosity, and abnormal characteristics of downhole acoustic signals, extracting implicit correlations between different variables.

[0095] The risk prediction model uses a graph neural network to construct a graph structure between data sources based on the physical and statistical relationships between drilling, production, and geological monitoring data variables. When drilling deep into highly permeable formations, changes in formation pressure can have a nonlinear impact on production rates, while adjustments to mud flow rates further influence changes in permeability. In this case, linear statistical models struggle to capture these complex interactions. However, through a graph-based message passing mechanism, graph neural networks can automatically extract the potential interactions between these variables and reveal the influence paths between different physical variables.

[0096] After processing the graph neural network, the data variables are fed into iTransformer. This module treats multidimensional data as independent tokens and, through iTransformer's self-attention mechanism, is able to capture temporal interactions and trend changes between these variables. This module not only captures drilling pressure fluctuations and changes in wellhead temperature sensor data, but also monitors changes in drill string torque, abnormal downhole fluid flow rates, formation permeability fluctuations, changes in wellhead vibration frequency, trends in drill bit temperature increases, changes in mud density and viscosity, and abnormal characteristics of downhole acoustic signals, extracting implicit correlations between these different variables.

[0097] Each token composed of variables will independently learn the deep representations in time series data and extract long-term dependencies and trend information. In the oil and gas production process, these variables are closely related to the risk of equipment failure. iTransformer can issue intelligent warnings in advance before these key signals accumulate to dangerous thresholds, optimize drilling parameters, and reduce the probability of unplanned downtime.

[0098] Step 3: Risk prediction and decision support.

[0099] This step enables the system to perform prediction and decision support, as well as verification.

[0100] 1) Prediction Module: The risk prediction module combines the outputs of GNN and iTransformer, synthesizing multiple prediction results through a voting mechanism to improve prediction accuracy and stability. During drilling operations, the system integrates data from different sources and intelligently adjusts the weights of each module based on historical trends and real-time monitoring information. This scoring not only integrates the short-term variation characteristics of different variables, but also incorporates long-term trends and nonlinear coupling relationships between variables, enabling more accurate identification of potential equipment failures or complex downhole conditions.

[0101] 2) Real-time Warning Module: This embodiment also provides real-time warning functionality to ensure immediate and efficient safety management. When the system detects that a well's risk score exceeds a preset threshold, it automatically triggers a real-time warning mechanism. Through intelligent analysis, it determines the likely cause of the anomaly and notifies relevant personnel to take appropriate measures to prevent further incidents. Warning information can be simultaneously delivered via a remote monitoring platform, mobile application, or on-site control system, ensuring that operators receive alerts immediately and can make rapid response decisions based on the anomaly analysis report provided by the system.

[0102] The risk prediction module combines the output of the graph neural network and iTransformer, and integrates multiple prediction results through a voting mechanism to improve the accuracy and stability of the prediction. During the drilling operation, the system will fuse data from different sources and intelligently adjust the weights of each module based on historical trends and real-time monitoring information. When the model identifies abnormal temperature increases, drastic pressure fluctuations, or abnormal deviations in fluid parameters in the monitoring data of a certain well, the system will automatically calculate and output a risk score. The higher the score, the greater the operational risk the equipment may face. This score not only integrates the short-term change characteristics of different variables, but also combines long-term trends with the nonlinear coupling relationship between variables, which can more accurately reveal potential equipment failures or complex downhole conditions.

[0103] At the same time, this embodiment provides a real-time early warning function to ensure the immediacy and efficiency of safety management. When the system detects that the risk score of a well exceeds the preset threshold, the system will automatically trigger the real-time early warning mechanism, determine the possible cause of the abnormality through intelligent analysis, and notify relevant personnel to take corresponding measures to prevent further development of the accident. The early warning information can be pushed synchronously through the remote monitoring platform, mobile application or on-site control system to ensure that the operator can receive the alarm information in the first time and make a quick response decision based on the abnormal analysis report provided by the system. If the system detects that the wellhead pressure fluctuates violently and is accompanied by an increase in the bottomhole temperature, it may mean that the downhole fluid circulation is blocked or the equipment is overloaded. The system will recommend that the operator immediately adjust the drilling parameters or check the status of related equipment to reduce potential risks. In addition, the early warning system also has adaptive learning capabilities. It can continuously optimize the warning threshold and risk score calculation method based on historical data, improve the robustness and accuracy of the model, thereby achieving more efficient intelligent safety management and ensuring the stable operation of drilling and production operations.

[0104] Step 4: Adaptive feedback and model optimization.

[0105] This embodiment enables the system to perform adaptive feedback and model optimization based on new data during actual operation, and provide evaluation.

[0106] 1) Model Optimization: By comparing the model's predictions with actual monitoring results, the system leverages an adaptive feedback mechanism to dynamically adjust model parameters and continuously optimize its predictive capabilities. If the system detects a significant discrepancy between the model's prediction of a device's failure and the actual situation, it automatically triggers a reinforcement learning strategy to adjust key parameters, improving both prediction accuracy and generalization.

[0107] 2) Furthermore, the model's adaptive feedback mechanism allows for personalized optimization for different operating conditions, enabling it to adapt to the changing operating conditions of different oil and gas wells. As data accumulates, the model will continue to iterate and evolve, increasing its sensitivity to potential failures and enabling more efficient and intelligent drilling and equipment health management.

[0108] Through this adaptive feedback mechanism, the model can be continuously optimized during long-term operation and adapt to changes in the data environment.

[0109] By comparing the model's predictions with actual monitoring results, the system leverages an adaptive feedback mechanism to dynamically adjust model parameters and continuously optimize its predictive capabilities. During drilling and equipment monitoring, the model continuously learns the deviations between predicted and observed values, self-correcting based on the patterns of change in different variables. If the system detects a significant deviation between the model's prediction of a particular device's failure and the actual situation, it automatically triggers a reinforcement learning strategy to adjust key parameters, improving both prediction accuracy and generalization.

[0110] This optimization process relies on the reward mechanism in reinforcement learning. When the model's prediction results are closer to the actual monitoring data, the system will provide positive feedback to enhance the current prediction strategy. When the prediction error is large, the system will make penalty adjustments to encourage the model to explore better parameter settings. If the system underestimates the failure risk of a certain device in a certain prediction, resulting in a failure to issue a timely warning, and the actual monitoring data shows that the device has indeed failed, the model will increase the weight of the relevant features in the subsequent training process to enhance the recognition ability of this type of failure. At the same time, the model will also use historical data retrospective analysis to find potential causes of errors, such as whether a certain variable weight is insufficient, the time window selection is unreasonable, or there is a deviation in the data distribution, and automatically adjust based on this information.

[0111] Furthermore, the model's adaptive feedback mechanism allows for personalized optimization for different operating conditions, enabling it to adapt to the changing operating conditions of different oil and gas wells. In complex and changing downhole environments, fixed-parameter prediction models often struggle to adapt to long-term operational needs. This system, however, continuously learns from new data and optimizes decision paths, ensuring stable and accurate fault prediction capabilities under diverse geological conditions and equipment states. As data accumulates, the model will continue to iterate and evolve, increasing its sensitivity to potential faults and enabling more efficient and intelligent drilling and equipment health management.

[0112] Through this adaptive feedback mechanism, the model can be continuously optimized during long-term operation and adapt to changes in the data environment.

[0113] Step 5: System integration and application.

[0114] 1) Offline and online integration. Offline: Data preprocessing, batch model training, and rule management. Online: Real-time data monitoring, triggering anomaly detection and alerting.

[0115] 2) Visualization interface and reporting system: Provides intuitive real-time data forecasts, alarm levels, and model optimization status.

[0116] Generate regular risk assessment reports and support multi-terminal access and push.

[0117] The oil and gas reservoir exploration and development risk prediction method proposed in this embodiment has the following technical effects: 1) Fusion of iTransformer and GNN: iTransformer accurately depicts the correlation between multi-dimensional time series through the idea of dimensionality inversion, while graph neural networks extract potential relationships from the global graph structure.

[0118] 2) Adaptive feedback and online optimization: Dynamically adjust the model through reinforcement learning to ensure highly consistent prediction results.

[0119] 3) Multi-dimensional data closed-loop management: Achieve a full-process closed-loop of data collection, cleaning, feature extraction, risk prediction and feedback.

[0120] 4) Real-time visualization and alarm mechanism: Through dynamic visualization, key parameters and risk trends are displayed to quickly identify risk points and issue timely warnings.

[0121] This embodiment successfully solves the problem of capturing complex nonlinear relationships, multivariable interactions, and long-range dependencies of multidimensional time series data in the process of oil and gas exploration and development by combining iTransformer with graph neural network GNN. iTransformer uses an inversion design with variables as tokens to more accurately identify time series interactions and mutation signals between variables, while graph neural network can construct effective association graphs between multi-source data and explore potential nonlinear coupling relationships and risk propagation paths. Through this fusion, this embodiment can improve the accuracy and stability of risk prediction, especially in an environment with dynamic changes in multidimensional data and long time series. In addition, the adaptive feedback mechanism combined with reinforcement learning enables the model to adjust parameters and risk thresholds in real time, effectively adapt to environmental changes, ensure the high reliability of prediction results, and further enhance the system's real-time warning and online update capabilities.

[0122] like Figure 3 As shown, this embodiment provides a device for predicting risks in oil and gas reservoir exploration and development, which may include: A construction unit 301 is configured to construct a normalized adjacency matrix, a node feature matrix, and a high-dimensional time series matrix based on parameter values of multiple production-related parameters in the oil and gas wells within a target period; The first prediction unit 302 is configured to input the normalized adjacency matrix and the node feature matrix into the trained graph neural network for risk prediction, thereby obtaining a risk feature matrix and a first risk score corresponding to each production-related parameter; wherein each row of data in the risk feature matrix is risk feature data corresponding to each production-related parameter; The second prediction unit 303 is used to input the risk feature matrix and the high-dimensional time series matrix into the trained multivariate time series processing model to perform risk prediction and obtain a second risk score corresponding to each production-related parameter; The determination unit 304 is configured to determine a risk level of the oil and gas well based on the first risk score and the second risk score corresponding to each production-related parameter.

[0123] It should be noted that the processing of the construction unit 301, the first prediction unit 302, the second prediction unit 303 and the determination unit 304 and the beneficial effects thereof can be referred to in the respective Figure 1 Steps S101 to S104 in the above are not described in detail.

[0124] Optionally, the construction unit 301 is further configured to: Determine the correlation between any two production-related parameters based on the parameter value of each production-related parameter within the target period; Each production-related parameter is regarded as a node, and each correlation degree is regarded as an edge between two corresponding nodes. A corresponding adjacency matrix is constructed according to each node and each edge, and the adjacency matrix is normalized to obtain a normalized adjacency matrix. Determine the statistical indicator sequence corresponding to each production-related parameter based on the parameter value of each production-related parameter within the target period; The statistical indicator sequence corresponding to each production-related parameter is used as the first row of data for constructing the matrix, and the first row of data corresponding to each production-related parameter is arranged based on the set parameter order to construct the node feature matrix.

[0125] Optionally, the graph neural network includes a graph convolutional network and a graph attention network; The first prediction unit 302 is further configured to: Input the normalized adjacency matrix and the node feature matrix into the graph convolution network, so that the graph convolution performs neighborhood aggregation on the first row of data corresponding to each production-related parameter in the node feature matrix based on the normalized adjacency matrix, and obtains the first aggregated row data corresponding to each production-related parameter and the first aggregated feature matrix, wherein the first aggregated feature matrix includes the first aggregated row data corresponding to each production-related parameter; The graph attention network is used to perform risk prediction on the first aggregated feature matrix to obtain the risk feature matrix and the first risk score corresponding to each production-related parameter.

[0126] Optionally, the first prediction unit 302 is further configured to: The first aggregate feature matrix is input into the graph attention network so that the graph attention network: based on the attention mechanism, the first aggregate row data corresponding to each production-related parameter in the first aggregate feature matrix is neighborhood aggregated to obtain the second aggregate row data and the second aggregate feature matrix corresponding to each production-related parameter, the second aggregate feature matrix includes the second aggregate row data corresponding to each production-related parameter, each second aggregate row data in the second aggregate feature matrix is used as risk feature data, and the second aggregate feature matrix is used as a risk feature matrix, feature mapping is performed on each risk feature data in the risk feature matrix, and the first risk score corresponding to each production-related parameter is obtained.

[0127] Optionally, the target period includes multiple time steps; The construction unit 301 is further configured to: For any production-related parameter, sort the parameter values of each time step of the production-related parameter in the target period in chronological order to obtain the parameter value time series of the production-related parameter; The parameter value time series of each production-related parameter is used as a column data for constructing the matrix; Arrange the column data corresponding to each production-related parameter based on the set parameter order to construct a target matrix; Key features are extracted from each row of the target matrix to obtain a high-dimensional time series matrix. Each row of the high-dimensional time series matrix is key feature data extracted from each row of the target matrix.

[0128] Optionally, the second prediction unit 303 is further configured to: The risk feature matrix and the high-dimensional time series matrix are input into the trained multivariate time series processing model so that the multivariate time series processing model: transposes the high-dimensional time series matrix to obtain a transposed matrix, splices the row data with equal row order in the transposed matrix and the risk feature matrix to obtain feature splicing data corresponding to each production-related parameter, performs risk prediction on the feature splicing data corresponding to each production-related parameter, and obtains a second risk score corresponding to each production-related parameter.

[0129] Optionally, the determining unit 304 is further configured to: Obtain the set weight corresponding to each production-related parameter, as well as the risk score ranges corresponding to multiple set risk levels; Based on the set weight corresponding to each production-related parameter, the first risk score and the second risk score corresponding to each production-related parameter are weightedly summed to obtain a global risk score; According to the global risk score, a matching target score interval is determined in the risk score intervals corresponding to the multiple set risk levels, and the set risk level corresponding to the target score interval is determined as the risk level of the oil and gas well.

[0130] Optionally, the determining unit 304 is further configured to: Inputting the first risk score corresponding to each production-related parameter into the trained classification model for risk level classification, thereby obtaining a first risk probability distribution output by the classification model; and inputting the second risk score corresponding to each production-related parameter into the classification model for risk level classification, thereby obtaining a second risk probability distribution output by the classification model; Obtaining a first set weight corresponding to the graph neural network and a second set weight corresponding to the multivariate time series processing model; Based on the first set weight and the second set weight, performing a weighted summation on the first risk probability distribution and the second risk probability distribution to obtain a final risk probability distribution; Determine the risk level of the oil and gas well based on the final risk probability distribution.

[0131] Optionally, when the first risk probability distribution and the second risk probability distribution include a high risk probability, a medium risk probability, and a low risk probability, the final risk probability distribution includes a high risk probability final value, a medium risk probability final value, and a low risk probability final value; The determining unit 304 is further configured to: Based on the first set weight and the second set weight, performing a weighted summation of the high risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain a high risk probability final value, performing a weighted summation of the medium risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain a medium risk probability final value, and performing a weighted summation of the low risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain a low risk probability final value; The determining unit 304 is further configured to: The maximum probability final value is determined among the high risk probability final value, the medium risk probability final value and the low risk probability final value, and the risk level corresponding to the maximum probability final value is determined as the risk level of the oil and gas well.

[0132] The oil and gas reservoir exploration and development risk prediction device proposed in this embodiment can use the trained graph neural network and multivariate time series processing model to mine the data associations and time series characteristics between multiple production-related parameters, accurately extract the complex associations between multiple production-related parameter data, improve the ability to capture dynamic risks, and ultimately determine the risk level of the oil and gas well based on the risk score output by the graph neural network and multivariate time series processing model, thereby effectively improving the accuracy of the risk level prediction of the oil and gas well.

[0133] The oil and gas reservoir exploration and development risk prediction device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0134] The embodiment of the present invention also provides a computer device having the above Figure 3 The device shown is for predicting risks in oil and gas reservoir exploration and development.

[0135] See also Figure 4, a structural diagram of a computer device provided by an optional embodiment of the present invention, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are connected to each other using different buses for communication, and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories. Similarly, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 4 A processor 10 is taken as an example.

[0136] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0137] The memory 20 stores instructions that can be executed by at least one processor 10, so as to enable at least one processor 10 to execute the method shown in the above embodiment.

[0138] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function. The data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0139] The memory 20 may include volatile memory, such as random access memory. The memory may also include non-volatile memory, such as flash memory, a hard disk, or a solid-state drive. The memory 20 may also include a combination of the above types of memory.

[0140] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0141] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for predicting risks in oil and gas reservoir exploration and development, characterized in that: include: Based on the parameter values of multiple production-related parameters in oil and gas wells within the target period, a normalized adjacency matrix, a node feature matrix and a high-dimensional time series matrix are constructed; Inputting the normalized adjacency matrix and the node feature matrix into a trained graph neural network for risk prediction, thereby obtaining a risk feature matrix and a first risk score corresponding to each of the production-related parameters; wherein each row of data in the risk feature matrix is risk feature data corresponding to each of the production-related parameters; Inputting the risk feature matrix and the high-dimensional time series matrix into a trained multivariate time series processing model for risk prediction to obtain a second risk score corresponding to each of the production-related parameters; The risk level of the oil and gas well is determined based on the first risk score and the second risk score corresponding to each of the production-related parameters.

2. The method according to claim 1, characterized in that Based on the parameter values of multiple production-related parameters in oil and gas wells within the target period, a normalized adjacency matrix and a node feature matrix are constructed, including: determining, based on the parameter value of each production-related parameter within the target time period, a correlation between any two of the production-related parameters; Taking each of the production-related parameters as a node, taking each of the correlations as an edge between two corresponding nodes, constructing a corresponding adjacency matrix according to each of the nodes and each of the edges, and normalizing the adjacency matrix to obtain the normalized adjacency matrix; Determining a statistical indicator sequence corresponding to each of the production-related parameters according to the parameter value of each of the production-related parameters within the target time period; The statistical indicator sequence corresponding to each of the production-related parameters is used as the first row of data for constructing a matrix, and the first row of data corresponding to each of the production-related parameters is arranged based on a set parameter order to construct the node feature matrix.

3. The method according to claim 2, characterized in that The graph neural network includes a graph convolutional network and a graph attention network; Inputting the normalized adjacency matrix and the node feature matrix into a trained graph neural network for risk prediction to obtain a risk feature matrix and a first risk score corresponding to each of the production-related parameters includes: Inputting the normalized adjacency matrix and the node feature matrix into the graph convolutional network, so that the graph convolution performs neighborhood aggregation on the first row of data corresponding to each of the production-related parameters in the node feature matrix based on the normalized adjacency matrix, to obtain first aggregated row data corresponding to each of the production-related parameters and a first aggregated feature matrix, wherein the first aggregated feature matrix includes the first aggregated row data corresponding to each of the production-related parameters; The graph attention network is used to perform risk prediction on the first aggregated feature matrix to obtain the risk feature matrix and a first risk score corresponding to each of the production-related parameters.

4. The method according to claim 3, characterized in that The using the graph attention network to perform risk prediction on the first aggregated feature matrix to obtain the risk feature matrix and a first risk score corresponding to each of the production-related parameters includes: The first aggregated feature matrix is input into the graph attention network so that the graph attention network: based on the attention mechanism, performs neighborhood aggregation on the first aggregated row data corresponding to each of the production-related parameters in the first aggregated feature matrix to obtain second aggregated row data and a second aggregated feature matrix corresponding to each of the production-related parameters, the second aggregated feature matrix includes the second aggregated row data corresponding to each of the production-related parameters, each of the second aggregated row data in the second aggregated feature matrix is used as the risk feature data, and the second aggregated feature matrix is used as the risk feature matrix, and feature mapping is performed on each of the risk feature data in the risk feature matrix to obtain the first risk score corresponding to each of the production-related parameters.

5. The method according to claim 1, characterized in that The target period includes multiple time steps; a high-dimensional time series matrix is constructed based on the parameter values of multiple production-related parameters in the oil and gas wells within the target period, including: For any of the production-related parameters, sorting the parameter values of each time step of the production-related parameter in the target period in chronological order to obtain a parameter value time series of the production-related parameter; Using the parameter value time series of each production-related parameter as a column data for constructing a matrix; Arranging the column data corresponding to each of the production-related parameters based on a set parameter order to construct a target matrix; Key features are extracted from each row of the target matrix to obtain the high-dimensional time series matrix, where each row of the high-dimensional time series matrix is key feature data extracted from each row of the target matrix.

6. The method according to claim 5, characterized in that Inputting the risk feature matrix and the high-dimensional time series matrix into a trained multivariate time series processing model for risk prediction to obtain a second risk score corresponding to each production-related parameter includes: The risk feature matrix and the high-dimensional time series matrix are input into a trained multivariate time series processing model so that the multivariate time series processing model: transposes the high-dimensional time series matrix to obtain a transposed matrix, splices the row data with equal row order in the transposed matrix and the risk feature matrix to obtain feature spliced data corresponding to each of the production-related parameters, performs risk prediction on the feature spliced data corresponding to each of the production-related parameters, and obtains a second risk score corresponding to each of the production-related parameters.

7. The method according to claim 1, characterized in that Determining the risk level of the oil and gas well based on the first risk score and the second risk score corresponding to each of the production-related parameters includes: Obtaining a set weight corresponding to each of the production-related parameters and a risk score interval corresponding to a plurality of set risk levels; Based on the set weight corresponding to each of the production-related parameters, a weighted sum of the first risk score and the second risk score corresponding to each of the production-related parameters is performed to obtain a global risk score; According to the global risk score, a matching target score interval is determined in the risk score intervals corresponding to the multiple set risk levels, and the set risk level corresponding to the target score interval is determined as the risk level of the oil and gas well.

8. The method according to claim 1, characterized in that Determining the risk level of the oil and gas well based on the first risk score and the second risk score corresponding to each of the production-related parameters includes: Inputting the first risk score corresponding to each of the production-related parameters into a trained classification model for risk level classification, thereby obtaining a first risk probability distribution output by the classification model; and inputting the second risk score corresponding to each of the production-related parameters into the classification model for risk level classification, thereby obtaining a second risk probability distribution output by the classification model; Obtaining a first set weight corresponding to the graph neural network and a second set weight corresponding to the multivariate time series processing model; Based on the first set weight and the second set weight, performing a weighted summation on the first risk probability distribution and the second risk probability distribution to obtain a final risk probability distribution; The risk level of the oil and gas well is determined according to the final risk probability distribution.

9. The method according to claim 8, characterized in that When the first risk probability distribution and the second risk probability distribution include a high risk probability, a medium risk probability, and a low risk probability, the final risk probability distribution includes a high risk probability final value, a medium risk probability final value, and a low risk probability final value; The step of performing weighted summation of the first risk probability distribution and the second risk probability distribution based on the first set weight and the second set weight to obtain a final risk probability distribution includes: Based on the first set weight and the second set weight, performing a weighted summation of the high risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain the high risk probability final value, performing a weighted summation of the medium risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain the medium risk probability final value, and performing a weighted summation of the low risk probabilities in the first risk probability distribution and the second risk probability distribution to obtain the low risk probability final value; Determining the risk level of the oil and gas well according to the final risk probability distribution includes: A maximum probability final value is determined among the high risk probability final value, the medium risk probability final value, and the low risk probability final value, and the risk level corresponding to the maximum probability final value is determined as the risk level of the oil and gas well.

10. A device for predicting risks in oil and gas reservoir exploration and development, characterized in that: include: A construction unit, used to construct a normalized adjacency matrix, a node feature matrix and a high-dimensional time series matrix according to parameter values of multiple production-related parameters in the oil and gas wells within a target period; A first prediction unit is configured to input the normalized adjacency matrix and the node feature matrix into a trained graph neural network for risk prediction, thereby obtaining a risk feature matrix and a first risk score corresponding to each of the production-related parameters; wherein each row of data in the risk feature matrix is risk feature data corresponding to each of the production-related parameters; A second prediction unit is configured to input the risk feature matrix and the high-dimensional time series matrix into a trained multivariate time series processing model to perform risk prediction, thereby obtaining a second risk score corresponding to each of the production-related parameters; A determination unit is configured to determine a risk level of the oil and gas well based on a first risk score and a second risk score corresponding to each of the production-related parameters.

Images