Oil production prediction model construction method and device, electronic equipment and storage medium

Abstract

The application discloses a method and device for constructing an oil production prediction model, electronic equipment and a storage medium. The method comprises the following steps: constructing an original high-dimensional time series data matrix according to multi-source sensing monitoring parameters collected in the oilfield production process; extracting N potential factors from the original high-dimensional time series data matrix; taking the N potential factors as candidate cause variables and taking an oil production sequence as a target variable, screening M causal factors from the N potential factors, and determining a causal effect coefficient corresponding to each causal factor; training a time series of the N potential factors to obtain an intermediate prediction model, performing attribution analysis, and outputting an attribution contribution proportion of each potential factor; performing two-dimensional quadrant diagnosis based on the causal effect coefficient and the attribution contribution proportion, and determining a combination weight of each causal factor according to a diagnosis result; and taking a time series of the M causal factors as model input characteristics, training with the combination weight, and obtaining an oil production prediction model.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more particularly to the field of oilfield production data analysis and intelligent prediction technology, specifically to a method, apparatus, electronic device and storage medium for constructing an oil production prediction model. Background Technology

[0002] Oil production forecasting is a core component of oilfield production management, and its accuracy directly impacts development plan optimization, production scheduling decisions, and economic benefit assessment. However, oilfield production involves the coupled effects of multiple physical mechanisms, including geological conditions, reservoir properties, wellbore flow, surface gathering and transportation, and artificial control. The monitoring parameters generated at each stage are numerous, multidimensional, and intertwined, resulting in a highly nonlinear and strongly coupled complex relationship between production and its influencing factors. Consequently, existing forecasting methods face challenges in practical applications, such as limited accuracy and insufficient reliability of forecast results, making it difficult to reliably support long-term production decision-making. Therefore, improving oil production forecasting methods is a crucial issue in the industry. Summary of the Invention

[0003] This application provides a method, apparatus, electronic device, and storage medium for constructing an oil production prediction model to improve the accuracy of oil production prediction.

[0004] In a first aspect, embodiments of this application also provide a method for constructing an oil production prediction model, including: Based on the multi-source sensor monitoring parameters collected during the oilfield production process, an original high-dimensional time series data matrix is ​​constructed. Factor extraction is performed on the original high-dimensional time series data matrix to obtain the number of latent factors N, and N latent factors are extracted. Using the N potential factors as candidate causal variables and the oil production sequence as the target variable, a time-series causal discovery algorithm is executed to screen out M causal factors that have a direct causal impact on oil production from the N potential factors, and to determine the causal effect coefficient corresponding to each causal factor. An intermediate prediction model is obtained by training the model using the time series of N latent factors as input features. Attribution analysis is then performed on the intermediate prediction model to output the attribution contribution ratio of each latent factor. Two-dimensional quadrant diagnosis is performed based on the causal effect coefficient and attribution contribution ratio of each potential factor, and the combined weight of each causal factor is determined based on the diagnosis results. Using the time series of M causal factors as the model input features, and training with the combined weights of the causal factors, a trained oil production prediction model is obtained.

[0005] Secondly, embodiments of this application also provide an apparatus for constructing an oil production prediction model, comprising: The multi-source parameter module is used to construct the original high-dimensional time-series data matrix based on the multi-source sensor monitoring parameters collected during the oilfield production process. The factor extraction module is used to perform factor extraction on the original high-dimensional time series data matrix to obtain the number of potential factors N, and extract N potential factors; The causal discovery module is used to use the N potential factors as candidate causal variables and the oil production sequence as the target variable, execute a time-series causal discovery algorithm, screen out M causal factors that have a direct causal impact on oil production from the N potential factors, and determine the causal effect coefficient corresponding to each causal factor. The attribution analysis module is used to train an intermediate prediction model using the time series of N latent factors as model input features, and to perform attribution analysis on the intermediate prediction model using attribution analysis methods to output the attribution contribution ratio of each latent factor. The two-dimensional quadrant diagnosis module is used to perform two-dimensional quadrant diagnosis based on the causal effect coefficient and attribution contribution ratio of each potential factor, and to determine the combined weight of each causal factor based on the diagnosis results. The model training module is used to train the oil production prediction model by using the time series of M causal factors as model input features and the combined weights of the causal factors.

[0006] Thirdly, embodiments of this application also provide an electronic device, which includes: One or more processors; Storage device for storing one or more programs; When one or more programs are executed by one or more processors, the one or more processors implement any of the oil production prediction model construction methods provided in the embodiments of this application.

[0007] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the oil production prediction model construction methods provided in embodiments of this application.

[0008] Fifthly, embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements any of the oil production prediction model construction methods provided in embodiments of this application.

[0009] The technical solution of this application, through multi-source data matrix construction, factor extraction and dimensionality reduction, time-series causal discovery, and model attribution analysis, gradually transforms multi-source heterogeneous monitoring data into predictive features with physical causal constraints, achieving an organic fusion of data-driven and causal priors. Two-dimensional quadrant diagnosis unifies the discrimination of causal effects and attribution contributions, effectively distinguishing between true causal factors and trend pseudo-correspondences, overcoming the defect of traditional attribution that misclassifies proxy variables as key factors. Based on the differentiated weighting strategy triggered by the diagnostic results, the weights integrate statistical attribution and causal priors, numerically correcting the systematic bias of SHAP's inability to identify physical causality. After this weighting mechanism is embedded in the model loss function, it guides the model to prioritize fitting causally validated true factors, ensuring physical consistency and interpretability while maintaining prediction accuracy, significantly improving the prediction reliability and generalization ability during the oilfield decline period. Attached Figure Description

[0010] Figure 1 This is a path diagram of a method for constructing an oil production prediction model according to Embodiment 1 of this application; Figure 2 This is a path diagram of another method for constructing an oil production prediction model according to Embodiment 2 of this application; Figure 3 This is a schematic diagram of a device for constructing an oil production prediction model according to Embodiment 3 of this application; Figure 4 This is a schematic diagram of the structure of an electronic device that implements the oil production prediction model construction method of the embodiments of this application. Detailed Implementation

[0011] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0012] It should be noted that the terms "first" and "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0013] Example 1 Figure 1 This is a path diagram of a method for constructing an oil production prediction model according to Embodiment 1 of this application. This embodiment is applicable to oilfield production scenarios for predicting oil production. It can be executed by an oil production prediction model construction device, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method includes: S101. Based on the multi-source sensor monitoring parameters collected during the oilfield production process, a raw high-dimensional time series data matrix is ​​constructed. Specifically, monitoring parameters collected by various types of sensors during oilfield production are acquired, including reservoir production parameters, well operating parameters, surface equipment parameters, produced fluid parameters, and operational parameters. Reservoir production parameters may include downhole pressure characterizing formation energy and downhole temperature characterizing thermodynamic state. Well operating parameters may include tubing differential pressure characterizing flow resistance and casing pressure characterizing annular state. Surface equipment parameters may include nozzle opening characterizing production control, wellhead pressure characterizing surface back pressure, wellhead temperature characterizing fluid thermodynamics, and nozzle differential pressure characterizing throttling effect. Produced fluid parameters include daily gas production characterizing dissolved gas driving force and daily water production characterizing water intrusion / water drive. Operational parameters include production duration characterizing operating indicators.

[0014] The daily frequency monitoring data collected by each sensor are aligned according to timestamps to form an original high-dimensional time-series data matrix. The rows of the original high-dimensional time-series data matrix correspond to the sampling time, and the columns correspond to each sensor parameter. This matrix preserves the original time-series information of each physical variable during oilfield production, providing data input for subsequent factor extraction steps. By constructing the original high-dimensional time-series data matrix, heterogeneous monitoring data from different physical dimensions and acquisition channels can be unified into a standardized matrix form, facilitating subsequent processing.

[0015] S102. Perform factor extraction on the original high-dimensional time series data matrix to obtain the number of potential factors N, and extract N potential factors; Factor extraction is performed on the original high-dimensional time-series data matrix. By analyzing the variance and covariance structure within the data, the original high-dimensional parameter space is compressed into a low-dimensional latent factor space. Specifically, variance contribution analysis is performed based on the data covariance structure. The optimal number of factors is determined according to model selection criteria and information retention thresholds. A quantitative relationship matrix between factors and original parameters is constructed. Orthogonal rotation is performed on the relationship matrix to simplify its structure. The original data is then linearly mapped using the rotated relationship matrix to extract factor sequences. Factor extraction can eliminate coupling interference between sensors while retaining most of the information in the original data, providing a low-dimensional, redundancy-free latent feature representation for subsequent analysis.

[0016] For example, potential factors extracted from reservoir production parameters, well operating parameters, surface equipment parameters, produced fluid parameters, and operational parameters include: wellbore pressure factor, reflecting the coupling between downhole pressure and tubing pressure differential, characterizing reservoir energy and wellbore flow resistance; thermodynamic production factor, reflecting the coupling between wellhead temperature and associated gas production, governed by dissolved gas evolution and pressure volume temperature (PVT) properties; wellhead control factor, reflecting the control coupling between casing pressure and nozzle opening, governed by throttling effects and manual adjustment; production operation factor, reflecting the production duration and nozzle-set operating mode, governed by operating procedures; surface flow factor, reflecting the surface-to-pipe flow coupling between nozzle opening and wellhead pressure, governed by surface hydrodynamics; and fluid production factor, reflecting the fluid production linkage between water production and gas production, governed by water drive or dissolved gas drive mechanisms. By compressing the original high-dimensional sensor parameter space into a low-dimensional potential factor space, most of the original information is retained while eliminating coupling interference between sensors.

[0017] S103. Using the N potential factors as candidate causal variables and the oil production sequence as the target variable, execute a time-series causal discovery algorithm to screen out M causal factors that have a direct causal impact on oil production from the N potential factors, and determine the causal effect coefficient corresponding to each causal factor; where 1 ≤ M <N; Using N potential factors as candidate causal variables and the oil production series as the target variable, a time-series causal discovery algorithm is executed. This algorithm uses multi-stage conditional independence verification to filter out statistically significant direct causal edges from all candidate time-delay causal links. The core of this algorithm is as follows: first, the input series is preprocessed to achieve stationarity to meet the algorithm's stationarity requirements; then, based on a preset maximum time delay range, a set of candidate directed edges pointing to the target variable from all potential factors at different lag orders is constructed; next, a recursive conditional independence verification stage quickly eliminates a large number of irrelevant candidate links to narrow the search range; finally, a multi-conditional independence verification stage performs a final verification on the remaining candidate links under a more complete set of conditions, identifying potential factors with a direct causal relationship to the target variable as causal factors, and extracting the correlation strength corresponding to each causal edge as the causal effect coefficient of the corresponding causal factor. For example, the three selected causal factors are wellhead control factor, fluid production factor, and thermodynamic production factor, with corresponding causal effect coefficients of -0.188, 0.062, and -0.070, respectively.

[0018] This time-series causal discovery process can quickly eliminate a large number of irrelevant candidate links through recursive conditional independence verification, significantly reducing the search space; through multi-conditional independence verification, it can perform final verification under a more complete set of conditions, effectively controlling the false positive rate of causal discovery and ensuring that the selected causal factors have a reliable direct causal relationship in a statistical sense rather than a spurious association; the association strength, as the causal effect coefficient, not only reflects the direction of the causal relationship, but also provides a continuous measure of the causal strength, providing a numerical basis for subsequent weight calculation.

[0019] S104. Train the intermediate prediction model using the time series of N potential factors as the model input features, and perform attribution analysis on the intermediate prediction model using the attribution analysis method to output the attribution contribution ratio of each potential factor. The intermediate prediction model is trained using N (e.g., N=6) latent factors as input and daily oil production as the target variable. Considering the significant decreasing trend in oilfield production data (oil production monotonically decreases over time), directly modeling the original sequence would lead to distribution drift between the training and test sets after time-series partitioning. Therefore, first-order differencing is performed on both the factor sequences and the target variable to predict the change in daily oil production rather than its absolute value, making the data approximately stationary. The SHAP (SHapley Additive exPlanations) method is used to perform attribution analysis on the intermediate prediction model, outputting the attribution contribution percentage of each latent factor. Specifically, the mean absolute value of the SHAP for each latent factor across all test samples is taken, and then the percentage of each factor's total contribution is calculated as the attribution contribution percentage of each latent factor.

[0020] S105. Perform two-dimensional quadrant diagnosis based on the causal effect coefficient and attribution contribution ratio of each potential factor, and determine the combined weight of each causal factor based on the diagnosis results. The two-dimensional quadrant diagnostics integrates the strength of causal effects and the model's attribution contribution into a single diagnostic space. It uses a dual-threshold classification system—a preset causal significance threshold and a uniform contribution threshold—to categorize each causal factor as a true core factor, an underfitting factor, an overfitting factor, or a noise factor. This allows for a systematic determination of whether the model's dependence is causally supported. The two-dimensional quadrant diagnostics not only identifies whether the model correctly utilizes causal factors but also precisely locates two different types of attribution bias: underfitting and overfitting.

[0021] Based on the quadrant to which each causal factor belongs, the factor type of that causal factor is determined. Then, based on different factor types, different strategies are employed to determine the combined weights of the causal factors according to their attribution contribution ratio and causal effect coefficient. This differentiated processing strategy integrates statistical attribution information and causal prior constraints in determining the combined weights. This two-dimensional quadrant diagnostic-triggered differentiated weighting strategy integrates statistical attribution information and causal prior constraints in determining the combined weights. From a numerical correction perspective, it addresses the systematic deficiency of standard attribution methods in distinguishing between physical causal relationships and statistical spurious associations when trend proxy variables exist. This ensures that the attribution results reflect model behavior while conforming to physical causal relationships, significantly improving the credibility and physical consistency of the predictive model's attribution explanation.

[0022] S106. Using the time series of M causal factors as the model input features, and training the model with the combined weights of the causal factors, a trained oil production prediction model is obtained.

[0023] Specifically, the time series of M causal factors are input into the oil production prediction model to obtain the predicted oil production output of the model. The oil production prediction model is a tree ensemble model, and the output of the model is a weighted combination of the M causal factors, with the combination weight corresponding to each causal factor being the combination weight of the corresponding causal factor. A weighted loss function is constructed based on the actual oil production, predicted oil production, and combination weights corresponding to the time series of the M causal factors to weight the prediction error of the training samples. The model parameters are optimized by minimizing the weighted loss function to obtain the trained oil production prediction model.

[0024] By embedding the combined weights of each causal factor into the loss function of the tree ensemble model, the model prioritizes fitting the causal factor with higher combined weights during parameter optimization, guiding the model to learn a mapping that conforms to physical causal relationships. The tree ensemble model can capture the nonlinear relationship between each causal factor and oil production. The weighted loss function allows the combined weights of each causal factor to directly participate in the optimization process of model parameters, realizing the fusion of causal priors and data-driven approaches.

[0025] The technical solution in this embodiment, through multi-source data matrix construction, factor extraction and dimensionality reduction, time-series causal discovery, and model attribution analysis, gradually transforms multi-source heterogeneous monitoring data into predictive features with physical causal constraints, achieving an organic fusion of data-driven and causal priors. Two-dimensional quadrant diagnosis unifies the discrimination of causal effects and attribution contributions, effectively distinguishing between true causal factors and trend pseudo-correspondences, overcoming the defect of traditional attribution that misclassifies proxy variables as key factors. A differentiated weighting strategy triggered by the diagnostic results integrates statistical attribution and causal priors, numerically correcting the systematic bias of SHAP's inability to identify physical causality. After this weighting mechanism is embedded in the model loss function, it guides the model to prioritize fitting causally validated true factors, ensuring physical consistency and interpretability while maintaining prediction accuracy, significantly improving the prediction reliability and generalization ability during the oilfield decline period.

[0026] In one optional implementation, the step of using the time series of M causal factors as model input features and training the model with the combined weights of the causal factors to obtain a trained oil production prediction model includes: inputting the time series of M causal factors into the oil production prediction model to obtain the predicted oil production output by the model, wherein the oil production prediction model is a tree ensemble model; constructing a weighted loss function based on the actual oil production corresponding to the time series of the M causal factors, the predicted oil production, and the combined weights of the causal factors; and minimizing the weighted loss function to train the model parameters to obtain the trained oil production prediction model.

[0027] Specifically, the time series data of M causal factors are used as input features to the oil production prediction model, and the predicted oil production output is obtained. The oil production prediction model is a tree ensemble model; the tree ensemble model uses the time series data of each causal factor as input features, and the combined weights of each causal factor are embedded in the loss function of the tree ensemble model for training. Based on the actual oil production, predicted oil production, and combined weights of each causal factor corresponding to its time series data, a weighted loss function is constructed. This weighted loss function uses the combined weights of each causal factor as sample weights, weighting the prediction error of each training sample. Causal factors with larger combined weights occupy a larger proportion in the loss function. The model parameters are trained by minimizing this weighted loss function, resulting in the trained oil production prediction model.

[0028] In one optional implementation, the step of performing factor extraction on the original high-dimensional time-series data matrix to obtain the number of latent factors N, and extracting N latent factors, includes: calculating the sample covariance matrix of the original high-dimensional time-series data matrix; performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors arranged in descending order; determining the number of latent factors N using the Bayesian information criterion combined with a cumulative explained variance threshold, wherein the number of latent factors is the minimum number of factors when the cumulative explained variance reaches a preset threshold; constructing a factor loading matrix; performing orthogonal rotation on the factor loading matrix using the maximum variance method to polarize the factor loadings after rotation; and performing linear mapping on the original data through the rotated loading matrix to extract N latent factors.

[0029] Specifically, a sample covariance matrix is ​​calculated from the original high-dimensional time-series data matrix. The elements of the sample covariance matrix reflect the cooperative variation relationships among the various sensing parameters. Eigenvalue decomposition is then performed on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors arranged in descending order. The magnitude of the eigenvalues ​​reflects the variance contribution in each principal component direction, while the eigenvectors define the projection direction from the original parameter space to the latent factor space. Eigenvalue decomposition decomposes the variance structure of the original high-dimensional parameter space into mutually orthogonal principal component directions. Each eigenvalue quantifies the information content in its corresponding direction, providing a mathematical basis for the objective selection of the number of factors and avoiding the arbitrariness of subjectively setting the number of factors.

[0030] Based on the eigenvalues ​​arranged in descending order, the Bayesian Information Criterion (BIC) is used to evaluate the balance between goodness of fit and complexity of the model under different numbers of factors. Simultaneously, the cumulative explained variance threshold is used to determine the number of latent factors N. Starting with the number of factors k=1, each factor is incremented sequentially. When the cumulative explained variance reaches a preset threshold (e.g., 95%), the number of factors k1 is recorded. Then, according to the Bayesian Information Criterion, the number of factors k2 with the lowest BIC value is selected from the candidate factor range. If k1 and k2 are inconsistent, the smaller value is taken as the final number of latent factors N, achieving an optimal balance between information retention and model simplicity. The BIC criterion effectively penalizes model complexity, avoiding overfitting caused by an excessive number of factors; the cumulative explained variance threshold ensures that the selected number of factors retains most of the information in the original data. The combination of these two criteria ensures that the determination of the number of factors is both statistically rigorous and informationally complete.

[0031] The absolute value of the causal effect coefficient corresponding to each causal factor is used as a measure of causal strength to calculate the combined weight of each causal factor. The causal effect coefficients corresponding to each of the M causal factors are obtained, and the absolute value of each causal effect coefficient is used as a measure of the causal strength of the corresponding causal factor. Based on the causal strength measures of each causal factor, the combined weight of each causal factor is calculated: the combined weight is equal to the proportion of the causal strength measure value of each causal factor to the sum of the causal strength measures values ​​of all causal factors. By converting the causal effect coefficient from absolute values ​​to a relative proportion, the causal strength of each causal factor is made comparable under a unified measurement scale, while ensuring that the sum of all weights is 1, satisfying the probability constraint on the weight distribution in subsequent weighted modeling. Using the absolute value of the causal effect coefficient as a measure avoids the influence of positive or negative causal direction on the intensity quantification, ensuring that regardless of whether the causal direction is positive or negative, factors with greater intensity receive higher weights in subsequent modeling.

[0032] A linear mapping is performed on the standardized original high-dimensional time-series data matrix using the load matrix orthogonally rotated using the maximum variance method: the original parameter vector at each sampling time is multiplied by the transpose of the rotated load matrix to obtain the corresponding N-dimensional latent factor vector. This mapping is performed on all sampling times to obtain the complete time series of N latent factors. By compressing the original high-dimensional parameter space into a low-dimensional latent factor space, multicollinearity coupling interference between the original sensors is eliminated. Simultaneously, since the load matrix has undergone orthogonal rotation, the extracted N latent factors remain approximately orthogonal, avoiding redundant information between factors and providing a clean, independent, and physically semantically clear set of candidate causal variables for subsequent time-series causal discovery.

[0033] Example 2 Figure 2 This is a path diagram for another oil production prediction model construction method provided in Embodiment 2 of this application. The technical solution of this embodiment is further refined based on the above technical solution. See also Figure 2 The method for constructing an oil production prediction model, as shown, includes: S201. Based on the multi-source sensor monitoring parameters collected during the oilfield production process, a raw high-dimensional time series data matrix is ​​constructed. S202. Perform factor extraction on the original high-dimensional time series data matrix to obtain the number of potential factors N, and extract N potential factors; S203. Using the N potential factors as candidate causal variables and the oil production sequence as the target variable, execute the time-series causal discovery algorithm to screen out M causal factors that have a direct causal impact on oil production from the N potential factors, and determine the causal effect coefficient corresponding to each causal factor. S204. Train an intermediate prediction model using the time series of N latent factors as model input features. Perform attribution analysis on the intermediate prediction model using attribution analysis method and output the attribution contribution ratio of each latent factor. S205. Map each of the causal factors to a two-dimensional diagnostic space with the causal effect coefficient as the horizontal axis and the attribution contribution ratio as the vertical axis; the two-dimensional diagnostic space includes four quadrants: the first quadrant corresponds to the causal effect coefficient > And attribution contribution > The second quadrant corresponds to causal effects > And attribution contribution ≤ The third quadrant corresponds to causal effects ≤ And attribution contribution > The fourth quadrant corresponds to causal effects ≤ And attribution contribution ≤ The and These are the preset causal significance threshold and the preset uniform contribution threshold, respectively; Specifically, each causal factor is mapped onto a two-dimensional diagnostic space with the causal effect coefficient on the horizontal axis and the attribution contribution ratio on the vertical axis. A causal significance threshold is set. and uniform contribution threshold Among them, the causal significance threshold The maximum inter-class variance method can be used for adaptive determination: For the absolute value sequence of causal effects of all N latent factors, the inter-class variance is maximized. The sequence is then optimally divided into a low-effect group and a high-effect group, and the midpoint of the means of the two groups is taken as the mean. (e.g., 0.1071); Uniform contribution threshold Taking 100% / N (e.g., 100% / 6 = 16.67%) represents the expected share of each factor when the contributions of each factor are perfectly uniform. Based on two thresholds, causal factors are divided into four quadrants: the first quadrant is where causal effects > And attribution contribution > The first quadrant corresponds to the true core factors; the second quadrant represents causal effects. And attribution contribution ≤ The third quadrant corresponds to the underfitting factor; the causal effect ≤ And attribution contribution > The first quadrant corresponds to the overfitting factor; the fourth quadrant represents causal effects ≤ And attribution contribution ≤ , corresponding to the noise factor.

[0034] By integrating the two independent dimensions of causal effect strength and model attribution contribution into a single diagnostic space through four-quadrant diagnosis, a systematic determination of whether model dependence has causal support can be achieved. Through dual-threshold division, it can not only identify whether the model has correctly utilized causal factors, but also accurately locate two different types of attribution bias: underfitting (causality is important but the model ignores it) and overfitting (the model depends on it but there is no causal support).

[0035] S206. Determine the combined weight of each causal factor based on the quadrant to which each causal factor belongs, the attribution contribution ratio of the causal factor, and the causal effect coefficient. Based on the quadrant to which each causal factor belongs, it is determined whether the causal factor is a true core factor, an underfitting factor, an overfitting factor, or a noise factor. If it is a true core factor, its attribution contribution ratio is used as the combined weight of the causal factor; if it is an underfitting or overfitting factor, its causal effect coefficient is used to correct the attribution contribution ratio, and the corrected result is used as the combined weight of the causal factor; if it is a noise factor, it is removed from the model input features. This differentiated processing strategy integrates statistical attribution information and causal prior constraints in determining the combined weight.

[0036] In one optional implementation, determining the combined weight of each causal factor based on the quadrant to which each causal factor belongs, the attribution contribution ratio of the causal factor, and the causal effect coefficient includes: determining the combined weight of each causal factor according to the quadrant to which each causal factor belongs in the following manner: if the causal factor belongs to the first quadrant, then the attribution contribution ratio of the causal factor is used as the combined weight of the causal factor; if the causal factor belongs to the second or third quadrant, then the product of the attribution contribution ratio of the causal factor and the absolute value of the causal effect coefficient is used as the combined weight of the causal factor.

[0037] Specifically, if a causal factor belongs to the first quadrant, its attribution contribution ratio is used as the combined weight of that causal factor; if a causal factor belongs to the second or third quadrant, the product of its attribution contribution ratio and the absolute value of its causal effect coefficient is used as the combined weight of that causal factor; if a latent factor belongs to the fourth quadrant, it is removed from the model input features. This differentiated weighting strategy integrates statistical attribution information and causal prior constraints in determining the combined weights: it keeps the attribution contribution ratio of true core factors unchanged to avoid overcorrection; it increases the attribution share of underfitting factors by weighting them with causal effect coefficients to compensate for insufficient model utilization; it reduces the attribution share of overfitting factors by weighting them with causal effect coefficients—those with zero causal effect coefficients are reduced to zero, and those with smaller coefficients are significantly reduced, thereby eliminating the interference of statistical spurious associations without causal support on the attribution results; and it directly removes noisy factors from the input features to reduce model complexity. This mechanism addresses the systematic deficiency of standard attribution methods in distinguishing between physical causality and statistical spurious associations when trend proxy variables are present, from the perspective of numerical correction. It ensures that attribution results reflect model behavior while conforming to physical causality, significantly improving the credibility and physical consistency of the attribution explanation of the prediction model.

[0038] S207. Using the time series of M causal factors as the model input features, and training the model with the combined weights of the causal factors, a trained oil production prediction model is obtained.

[0039] The technical solution of this embodiment, based on factor extraction and temporal causal discovery, maps each causal factor to a two-dimensional diagnostic space with the causal effect coefficient as the horizontal axis and the attribution contribution ratio as the vertical axis. Through a dual-threshold four-quadrant division, it achieves a systematic determination of whether the model dependency has real causal support, accurately locating two types of attribution bias: underfitting and overfitting. A differentiated weighting strategy is triggered based on quadrant assignment, allowing the combined weights to simultaneously integrate statistical attribution information and causal prior constraints. This weighting mechanism is embedded in the model loss function, guiding the model to prioritize fitting causally validated real factors. This closed-loop solution effectively overcomes the shortcomings of standard attribution methods in distinguishing between physical causal relationships and statistical pseudo-associations, significantly improving the physical consistency, interpretability, and generalization robustness of oilfield decline period predictions while maintaining prediction accuracy.

[0040] In one optional implementation, S203 includes: stabilizing the N potential factors and the oil production sequence to obtain a stabilized candidate causal variable sequence and a target variable sequence; setting a maximum lag period and constructing a candidate causal link set; in the conditional independence verification stage, performing a recursive conditional independence verification on each candidate link in the candidate causal link set, using partial correlation verification as the conditional independence verification method, and eliminating insignificant candidate links at a preset significance level to obtain a reduced candidate link set; in the multi-conditional independence verification stage, performing a multi-conditional independence verification on each candidate link in the reduced candidate link set... A candidate link is selected, and the estimated parent node set of the target variable and the estimated parent node set of the candidate causal variable are added to the condition set simultaneously. Conditional independence is checked, and significant causal links that pass the check are selected using a corresponding preset significance level threshold to form a directed acyclic causal graph. All directed edges that directly point to the target variable are extracted from the directed acyclic causal graph. The latent factors connected by the directed edges are identified as causal factors, and the latent factors that fail the multi-conditional independence check are identified as non-causal factors and removed. The conditional partial correlation coefficients corresponding to each directed edge that passes the multi-conditional independence check are used as the causal effect coefficients of the corresponding causal factors.

[0041] Using N potential factors as candidate causal variables and the oil production series as the target variable, the PCMCI (Peter-Clark Momentary Conditional Independence) time-series causal discovery algorithm is executed to screen out M causal factors that have a direct causal impact on oil production from the N potential factors, and to determine the causal effect coefficient corresponding to each causal factor.

[0042] The core of this algorithm lies in filtering true causal edges from candidate links through a two-stage conditional independence check. Specifically, the N potential factors and the oil production sequence are respectively stabilized (e.g., by first-order differencing) to obtain the stabilized candidate causal variable sequence and target variable sequence, ensuring the stability requirement of the algorithm input; a maximum lag period is set. (For example, taking 8 days to cover the main response period of pressure transmission), construct a candidate causal link set C= This set contains all candidate directed edges pointing to the target variable from each latent factor at different lag orders. In the recursive conditional independence verification phase (PC phase), a recursive conditional independence verification is performed on each candidate link in the candidate causal link set. Partial correlation verification is used as the conditional independence verification method, and insignificant candidate links are eliminated at a preset significance level, resulting in a reduced candidate link set. In the multi-conditional independence verification phase (MCI phase), for each candidate link in the reduced candidate link set, the estimated parent node set of the target variable and the estimated parent node set of the candidate causal variable are added to the condition set simultaneously. Conditional independence verification is then performed, and significant causal links that pass the verification are selected using a corresponding preset significance level threshold, forming a directed acyclic causal graph. All directed edges that directly point to the target variable are extracted from the directed acyclic causal graph. The potential factors connected by these directed edges are identified as causal factors that have a direct causal impact on oil production. The conditional partial correlation coefficients corresponding to each directed edge that passes the multi-conditional independence test are used as the causal effect coefficients of the corresponding causal factors. Potential factors that fail the multi-conditional independence test are identified as non-causal factors and are removed.

[0043] This time-series causal discovery process can quickly eliminate a large number of irrelevant candidate links through recursive conditional independence verification in the PC stage, significantly reducing the search space. By adding the estimated parent node set of both the target variable and the candidate causal variable to the condition set in the MCI stage, the condition set is made more complete, effectively controlling the false positive rate of causal discovery and ensuring that the selected causal factors have a reliable direct causal relationship in a statistical sense rather than a spurious association. The conditional partial correlation coefficient, as the causal effect coefficient, not only reflects the direction of the causal relationship but also provides a continuous measure of the causal strength, providing a numerical basis for subsequent weight calculations.

[0044] Example 3 Figure 3 This is a schematic diagram of a petroleum production prediction model construction device according to Embodiment 3 of this application. This embodiment is applicable to oilfield production scenarios for predicting petroleum production. The petroleum production prediction model construction device can be implemented in hardware and / or software, and can be configured in an electronic device. (Reference) Figure 3 The specific structure of the oil production prediction model construction device 300 is as follows: The multi-source parameter module 310 is used to construct an original high-dimensional time-series data matrix based on the multi-source sensor monitoring parameters collected during the oilfield production process. The factor extraction module 320 is used to perform factor extraction on the original high-dimensional time series data matrix to obtain the number of potential factors N, and extract N potential factors. The causal discovery module 330 is used to use the N potential factors as candidate causal variables and the oil production sequence as the target variable, execute a time-series causal discovery algorithm, screen out M causal factors that have a direct causal impact on oil production from the N potential factors, and determine the causal effect coefficient corresponding to each causal factor. The attribution analysis module 340 is used to train an intermediate prediction model using the time series of N latent factors as model input features, and to perform attribution analysis on the intermediate prediction model using the attribution analysis method to output the attribution contribution ratio of each latent factor. The two-dimensional quadrant diagnosis module 350 is used to perform two-dimensional quadrant diagnosis based on the causal effect coefficient and attribution contribution ratio of each potential factor, and to determine the combined weight of each causal factor based on the diagnosis results. The model training module 360 ​​is used to train the oil production prediction model by using the time series of M causal factors as model input features and the combined weights of the causal factors.

[0045] In one alternative implementation, the two-dimensional quadrant diagnostic module 350 includes: A two-dimensional diagnostic space unit is used to map each of the causal factors to a two-dimensional diagnostic space with the causal effect coefficient as the horizontal axis and the attribution contribution ratio as the vertical axis; the two-dimensional diagnostic space includes four quadrants: the first quadrant corresponds to the causal effect > And attribution contribution > The second quadrant corresponds to causal effects > And attribution contribution ≤ The third quadrant corresponds to causal effects ≤ And attribution contribution > The fourth quadrant corresponds to causal effects ≤ And attribution contribution ≤ The and These are the preset causal significance threshold and the preset uniform contribution threshold, respectively; The combined weighting unit is used to determine the combined weight of each causal factor based on the quadrant to which each causal factor belongs, the attribution contribution ratio of the causal factor, and the causal effect coefficient.

[0046] In one alternative implementation, the combined weighting unit is specifically used for: Based on the quadrant to which each causal factor belongs, the combined weight of each causal factor is determined as follows: If the causal factor belongs to the first quadrant, then the attribution contribution ratio of the causal factor shall be used as the combined weight of the causal factor. If the causal factor belongs to the second or third quadrant, the product of the attribution contribution ratio of the causal factor and the absolute value of the causal effect coefficient shall be used as the combined weight of the causal factor.

[0047] In one optional embodiment, the factor extraction module 320 includes: The sample covariance unit is used to calculate the sample covariance matrix of the original high-dimensional time series data matrix, and to perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors arranged in descending order. The latent factor quantity unit is used to determine the number of latent factors N by combining the Bayesian information criterion with the cumulative explained variance threshold. The value of the latent factor quantity is the minimum number of factors when the cumulative explained variance reaches a preset threshold. The factor loading unit is used to construct the factor loading matrix and perform orthogonal rotation on the factor loading matrix using the maximum variance method, so that the factor loadings are polarized after rotation. The latent factor extraction unit is used to perform a linear mapping on the original data through the rotated loading matrix to extract N latent factors.

[0048] In one alternative implementation, the causal discovery module 330 includes: A stabilization processing unit is used to stabilize the N potential factors and the oil production sequence respectively, to obtain the stabilized candidate causal variable sequence and the target variable sequence; Causal linking units are used to set the maximum lag period and construct a set of candidate causal links; The recursive condition verification unit is used to perform recursive condition independence verification on each candidate link in the candidate causal link set during the recursive condition independence verification stage. It adopts partial correlation verification as the condition independence verification method and removes insignificant candidate links at a preset significance level to obtain a reduced candidate link set. The multi-condition verification unit is used to, during the multi-condition independence verification stage, add the estimated parent node set of the target variable and the estimated parent node set of the candidate cause variable to the condition set for each candidate link in the reduced candidate link set, perform condition independence verification, and use the corresponding preset significance level threshold to filter the significant causal links that pass the verification to form a directed acyclic causal graph. The causal factor unit is used to extract all directed edges that directly point to the target variable from the directed acyclic causal graph, determine the latent factors connected by the directed edges as causal factors, and remove the latent factors that fail the multi-condition independence test as non-causal factors. The causal effect coefficient unit is used to take the conditional partial correlation coefficients corresponding to each directed edge that has passed the multi-conditional independence test as the causal effect coefficients of the corresponding causal factors.

[0049] In one alternative implementation, the model training module 360 ​​includes: The production forecasting unit is used to input the time series of M causal factors into the oil production forecasting model and obtain the predicted oil production output by the model, wherein the oil production forecasting model is a tree ensemble model. The weighted loss function unit is used to construct a weighted loss function based on the actual oil production corresponding to the time series of M causal factors, the predicted oil production, and the combined weights of each causal factor. The model training unit is used to minimize the weighted loss function to train the model parameters and obtain the trained oil production prediction model.

[0050] The oil production prediction model construction apparatus provided in this application can execute the oil production prediction model construction method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects of executing the oil production prediction model construction method.

[0051] According to embodiments of the present invention, the present invention also provides an electronic device, a readable storage medium, and a computer program product.

[0052] Example 4 Figure 4 This is a schematic diagram of the structure of an electronic device 410 that implements the oil production prediction model construction method of the embodiments of this application. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0053] like Figure 4As shown, the electronic device 410 includes at least one processor 411 and a memory, such as a read-only memory (ROM) 412 or a random access memory (RAM) 413, communicatively connected to the at least one processor 411. The memory stores computer programs executable by the at least one processor. The processor 411 can perform various appropriate actions and processes based on the computer program stored in the ROM 412 or loaded from storage unit 418 into the RAM 413. The RAM 413 may also store various programs and data required for the operation of the electronic device 410. The processor 411, ROM 412, and RAM 413 are interconnected via a bus 414. An input / output (I / O) interface 415 is also connected to the bus 414.

[0054] Multiple components in electronic device 410 are connected to I / O interface 415, including: input unit 416, such as keyboard, mouse, etc.; output unit 417, such as various types of displays, speakers, etc.; storage unit 418, such as disk, optical disk, etc.; and communication unit 419, such as network card, modem, wireless transceiver, etc. Communication unit 419 allows electronic device 410 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0055] Processor 411 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 411 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 411 performs the various methods and processes described above, such as the oil production forecasting model building method.

[0056] In some embodiments, the oil production forecasting model construction method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 418. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 410 via ROM 412 and / or communication unit 419. When the computer program is loaded into RAM 413 and executed by processor 411, one or more steps of the oil production forecasting model construction method described above may be performed. Alternatively, in other embodiments, processor 411 may be configured as the oil production forecasting model construction method by any other suitable means (e.g., by means of firmware).

[0057] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0058] Computer programs used to implement the methods of this application may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable device, such that when executed by the processor, the computer programs cause the functions / operations specified in the path diagrams and / or block diagrams to be implemented. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0059] In the context of this application, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0060] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0061] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0062] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0063] It should be understood that the various paths shown above can be used to reorder, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this application can be achieved, and this is not limited herein.

[0064] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for constructing an oil production prediction model, characterized in that, include: Based on the multi-source sensor monitoring parameters collected during the oilfield production process, an original high-dimensional time series data matrix is ​​constructed. Factor extraction is performed on the original high-dimensional time series data matrix to obtain the number of latent factors N, and N latent factors are extracted. Using the N potential factors as candidate causal variables and the oil production sequence as the target variable, a time-series causal discovery algorithm is executed to screen out M causal factors that have a direct causal impact on oil production from the N potential factors, and to determine the causal effect coefficient corresponding to each causal factor. An intermediate prediction model is obtained by training the model using the time series of N latent factors as input features. Attribution analysis is then performed on the intermediate prediction model to output the attribution contribution ratio of each latent factor. Two-dimensional quadrant diagnosis is performed based on the causal effect coefficient and attribution contribution ratio of each potential factor, and the combined weight of each causal factor is determined based on the diagnosis results. Using the time series of M causal factors as the model input features, and training with the combined weights of the causal factors, a trained oil production prediction model is obtained.

2. The method according to claim 1, characterized in that, The two-dimensional quadrant diagnosis based on the causal effect coefficient and attribution contribution ratio of each potential factor, and the determination of the combined weight of each causal factor based on the diagnosis results, includes: Each causal factor is mapped to a two-dimensional diagnostic space with the causal effect coefficient as the horizontal axis and the attribution contribution ratio as the vertical axis; the two-dimensional diagnostic space includes four quadrants: the first quadrant corresponds to the causal effect > And attribution contribution > The second quadrant corresponds to causal effects > And attribution contribution ≤ The third quadrant corresponds to causal effects ≤ And attribution contribution > The fourth quadrant corresponds to a causal effect ≤ And attribution contribution ≤ The and These are the preset causal significance threshold and the preset uniform contribution threshold, respectively; The combined weight of each causal factor is determined based on the quadrant to which each causal factor belongs, the attribution contribution ratio of each causal factor, and the causal effect coefficient.

3. The method according to claim 2, characterized in that, The determination of the combined weight of each causal factor based on the quadrant to which each causal factor belongs, the attribution contribution ratio of each causal factor, and the causal effect coefficient includes: Based on the quadrant to which each causal factor belongs, the combined weight of each causal factor is determined as follows: If the causal factor belongs to the first quadrant, then the attribution contribution ratio of the causal factor shall be used as the combined weight of the causal factor. If the causal factor belongs to the second or third quadrant, the product of the attribution contribution ratio of the causal factor and the absolute value of the causal effect coefficient shall be used as the combined weight of the causal factor.

4. The method according to claim 1, characterized in that, The process involves performing factor extraction on the original high-dimensional time-series data matrix to obtain the number of latent factors N, and extracting N latent factors, including: Calculate the sample covariance matrix of the original high-dimensional time series data matrix, and perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors arranged in descending order; The number of latent factors N is determined by combining the Bayesian information criterion with the cumulative explained variance threshold. The number of latent factors is the minimum number of factors when the cumulative explained variance reaches a preset threshold. Construct a factor loading matrix, and perform orthogonal rotation on the factor loading matrix using the maximum variance method to make the factor loadings polarized after rotation. The original data is linearly mapped using the rotated loading matrix to extract N latent factors.

5. The method according to claim 1, characterized in that, The process involves using N potential factors as candidate causal variables and the oil production series as the target variable, executing a time-series causal discovery algorithm to select M causal factors that have a direct causal impact on oil production from the N potential factors, and determining the causal effect coefficient corresponding to each causal factor, including: The N potential factors and the oil production sequence are respectively stabilized to obtain the stabilized candidate causal variable sequence and target variable sequence; Set a maximum lag period and construct a set of candidate causal links; In the recursive conditional independence verification stage, a recursive conditional independence verification is performed on each candidate link in the candidate causal link set. Partial correlation verification is used as the conditional independence verification method. Insignificant candidate links are removed at a preset significance level to obtain a reduced candidate link set. In the multi-condition independence verification stage, for each candidate link in the reduced candidate link set, the estimated parent node set of the target variable and the estimated parent node set of the candidate cause variable are added to the condition set at the same time. Condition independence verification is performed, and significant causal links that pass the verification are selected using the corresponding preset significance level threshold to form a directed acyclic causal graph. Extract all directed edges that directly point to the target variable from the directed acyclic causal graph. The latent factors connected by the directed edges are determined to be causal factors, and the latent factors that fail the multi-condition independence test are determined to be non-causal factors and are removed. The conditional partial correlation coefficients corresponding to each directed edge that passes the multi-conditional independence test are used as the causal effect coefficients of the corresponding causal factors.

6. The method according to claim 1, characterized in that, The method of using time series data of M causal factors as model input features and training the model with combined weights of the causal factors to obtain a trained oil production prediction model includes: The time series of M causal factors are input into the oil production prediction model to obtain the predicted oil production output by the model. The oil production prediction model is a tree ensemble model. A weighted loss function is constructed based on the actual oil production corresponding to the time series of M causal factors, the predicted oil production, and the combined weights of each causal factor. The trained oil production prediction model is obtained by minimizing the weighted loss function to train the model parameters.

7. An apparatus for constructing an oil production prediction model, characterized in that, include: The multi-source parameter module is used to construct the original high-dimensional time-series data matrix based on the multi-source sensor monitoring parameters collected during the oilfield production process. The factor extraction module is used to perform factor extraction on the original high-dimensional time series data matrix to obtain the number of potential factors N, and extract N potential factors; The causal discovery module is used to use the N potential factors as candidate causal variables and the oil production sequence as the target variable, execute a time-series causal discovery algorithm, screen out M causal factors that have a direct causal impact on oil production from the N potential factors, and determine the causal effect coefficient corresponding to each causal factor. The attribution analysis module is used to train an intermediate prediction model using the time series of N latent factors as model input features, and to perform attribution analysis on the intermediate prediction model using attribution analysis methods to output the attribution contribution ratio of each latent factor. The two-dimensional quadrant diagnosis module is used to perform two-dimensional quadrant diagnosis based on the causal effect coefficient and attribution contribution ratio of each potential factor, and to determine the combined weight of each causal factor based on the diagnosis results. The model training module is used to train the oil production prediction model by using the time series of M causal factors as model input features and the combined weights of the causal factors.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the oil production prediction model construction method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the oil production prediction model construction method as described in any one of claims 1-6.

10. A computer program product comprising a computer program that, when executed by a processor, implements the oil production prediction model construction method as described in any one of claims 1-6.

Images