A method, system and device for evaluating the effect of a stimulation measure based on causal VAE

Abstract

The application discloses a kind of based on CausalVAE's stimulation measure effect evaluation method, system and equipment, it is related to oil and gas field production optimization technical field, the method includes the production data of target oil well and adjacent well in continuous time period is preprocessed, obtains multidimensional time series data;Through the causal representation model of CausalVAE encoder-decoder architecture, multidimensional time series data is decomposed into measure factor and natural factor, the decoupling of the potential causal source of production data is realized;At the moment of the implementation of stimulation, measure factor is zero, only using natural factor generates the counterfactual prediction curve of key production index under not implementing stimulation;By comparing actual observation curve and counterfactual prediction curve, the effect evaluation result of stimulation is obtained, to improve the accuracy of evaluation result.The model can learn complex nonlinear relationship from production data, and still can maintain stable evaluation performance under complex working conditions, and is more suitable.

Description

Technical Field

[0001] This invention relates to the field of oil and gas field production optimization technology, specifically to a method, system, and equipment for evaluating the effectiveness of production enhancement measures based on CausalVAE. Background Technology

[0002] In oil and gas field development, production enhancement and control measures such as acidizing, fracturing, water injection adjustment, and pump rate optimization are commonly used to improve wellbore or near-wellbore formation conditions, thereby increasing production capacity or reducing water cut in the short term. However, the changes in production and water cut observed in the field are not entirely due to the measures themselves, but are also affected by a variety of natural factors, such as: natural recovery after the removal of fluid accumulation in the wellbore, reservoir pressure recovery, interference from injection and production in adjacent wells, and fluctuations caused by reservoir heterogeneity.

[0003] Currently, three main methods are used to evaluate the effectiveness of production enhancement measures in the field. The first is the adjacent well comparison method, which infers the effect of the measures by comparing the production differences between the wells with the measures and those with adjacent wells. However, in blocks with strong well network connectivity, adjacent wells are often affected by the wells with the measures, causing the comparison benchmark to become invalid and the evaluation results to be inaccurate. The second is the decline curve fitting method, which uses historical production before the implementation of measures to fit a decline model to predict the production under the scenario without measures, and then compares it with the actual production. However, this method assumes that the production change is stable and is difficult to handle abrupt fluctuations caused by measures or shutdowns, resulting in a large prediction bias. The third is the numerical simulation method, which constructs a reservoir geological model and performs numerical simulation calculations under the scenarios of "with measures" and "without measures" to evaluate the effect. However, the model is highly dependent on geological parameters and boundary conditions, and the calculation is time-consuming, which cannot meet the needs of rolling applications.

[0004] Therefore, existing methods generally suffer from problems such as difficulty in removing confusion effects, strong reliance on experience, inaccurate assessment, low assessment efficiency, and insufficient adaptability. These technical defects are particularly prominent under complex conditions such as high water cut, multi-well group coupling, and unconventional reservoirs, which seriously restrict the scientific nature and timeliness of oilfield production decisions. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that the confusion effect is difficult to remove, the evaluation is inaccurate, the evaluation efficiency is low and the adaptability is insufficient in the prior art. The purpose is to provide a method, system and equipment for evaluating the effect of production increase measures based on CausalVAE, which solves the above problems.

[0006] This invention is achieved through the following technical solution:

[0007] In a first aspect, the present invention provides a method for evaluating the effectiveness of yield-increasing measures based on CausalVAE, comprising:

[0008] Acquire production data of the target oil well and adjacent wells over a continuous time period, and preprocess the data to obtain structured multidimensional time-series data;

[0009] A CausalVAE causal representation model with an encoder-decoder architecture is constructed, in which the encoder maps the multidimensional time series data into latent variables; the latent variables include measure factors for characterizing human intervention factors and natural factors for characterizing non-human intervention factors.

[0010] At the moment of implementation of the production increase measures, the measure factor is set to zero, and the decoder generates counterfactual prediction curves of key production indicators without the implementation of production increase measures using only the natural factors.

[0011] Obtain actual observation curves of key production indicators under the implementation of the aforementioned production increase measures;

[0012] By comparing the actual observed curves of the key production indicators with the counterfactual predicted curves of the key production indicators, the effectiveness evaluation results of the production increase measures are obtained.

[0013] Optionally, the key production indicators include yield and moisture content; the comparison of the actual observed curves and the counterfactual predicted curves of the key production indicators to obtain the effectiveness evaluation results of the yield-increasing measures includes:

[0014] By comparing the actual observed curve of the output with the counterfactual predicted curve of the output, the effect evaluation results of the output-increasing measures on the output are obtained;

[0015] By comparing the actual observed curve of the moisture content with the counterfactual predicted curve of the moisture content, the effect evaluation results of the yield-increasing measures on the moisture content are obtained.

[0016] Optionally, the effect evaluation result of the production-increasing measures on output includes the daily contribution of the measures; the comparison of the actual observed curve of output and the counterfactual predicted curve of output to obtain the effect evaluation result of the production-increasing measures on output includes:

[0017] For any day within the assessment period, the actual observed output value for that day is determined based on the actual observed output curve, and the counterfactual predicted output value for that day is determined based on the counterfactual predicted output curve.

[0018] The difference between the actual observed output value and the counterfactual predicted output value for any given day is calculated to obtain the contribution of the measures on that day.

[0019] Optionally, the evaluation result of the production-increasing measures on the output also includes the cumulative contribution of the measures over multiple days; after calculating the difference between the actual observed output value and the counterfactual predicted output value for any given day to obtain the contribution of the measures on a single day, the method further includes:

[0020] The cumulative contribution of the measures over multiple days is obtained by summing the daily contribution amounts of the measures taken within a continuous time period during the assessment period.

[0021] Optionally, the effect evaluation result of the yield-increasing measures on moisture content includes the amount of moisture content improvement; the step of comparing the actual observed curve of the moisture content and the counterfactual predicted curve of the moisture content to obtain the effect evaluation result of the yield-increasing measures on moisture content includes:

[0022] For any day within the assessment period, the actual observed moisture content of that day is determined based on the actual observed moisture content curve, and the counterfactual predicted moisture content of that day is determined based on the counterfactual predicted moisture content curve.

[0023] Calculate the difference between the actual observed moisture content and the counterfactual predicted moisture content for any given day, and average the difference over a continuous period within the assessment period to obtain the moisture content improvement.

[0024] Optionally, the construction of the CausalVAE causal representation model with an encoder-decoder architecture includes the following training steps:

[0025] Historical production data of the target oil well and adjacent wells over a continuous period of time are acquired and preprocessed to obtain structured historical multidimensional time-series data.

[0026] Construct a CausalVAE causal representation model with an encoder-decoder architecture, and build a loss function that includes a reconstruction loss term and a KL divergence regularization term;

[0027] Based on the historical multidimensional time-series data, the parameters of the CausalVAE causal representation model are optimized and trained by minimizing the loss function.

[0028] Optionally, the loss function is as follows:

[0029]

[0030] Among them, Q t Indicates the first Actual observed daily yield; The first one, reconstructed by the decoder, represents the... Daily projected output value; W t Indicates the first The actual observed moisture content on that day; The first one, reconstructed by the decoder, represents the... Predicted moisture content for the day; z c This represents the measure factor extracted by the encoder; D represents the natural factor extracted by the encoder; KL (⋅) represents the KL divergence; λ is the weighting coefficient of the KL divergence; X represents the historical multidimensional time series data; T represents the time series length of the historical multidimensional time series data; q(,) represents the posterior distribution; p(,) represents the prior distribution.

[0031] Optionally, the step of acquiring production data of the target oil well and adjacent wells over a continuous time period and preprocessing it to obtain structured multidimensional time-series data includes:

[0032] Collect production data of the target oil well and adjacent wells over a continuous period of time; the production data includes daily oil production, water cut, water injection volume, wellhead pressure, and pump speed;

[0033] The production data is filled with missing values, outliers are removed, and normalized to obtain structured multidimensional time-series data.

[0034] Secondly, the present invention provides a system for evaluating the effectiveness of yield-increasing measures based on CausalVAE, comprising:

[0035] The preprocessing module is used to acquire production data of the target oil well and adjacent wells over a continuous time period, and to preprocess the data to obtain structured multidimensional time-series data.

[0036] The prediction module is used to construct a CausalVAE causal representation model with an encoder-decoder architecture. The encoder maps the multidimensional time series data into latent variables. The latent variables include measure factors that characterize human intervention factors and natural factors that characterize non-human intervention factors. At the time of implementation of production-increasing measures, the measure factors are set to zero, and the decoder uses only the natural factors to generate counterfactual prediction curves of key production indicators without the implementation of production-increasing measures.

[0037] The data acquisition module is used to obtain the actual observation curves of key production indicators under the implementation of the aforementioned production increase measures;

[0038] The effect evaluation module is used to compare the actual observed curves of the key production indicators with the counterfactual prediction curves of the key production indicators to obtain the effect evaluation results of the production increase measures.

[0039] Thirdly, the present invention provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement a method for evaluating the effectiveness of production-increasing measures based on CausalVAE as described in any one of the first aspects.

[0040] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0041] Existing technologies cannot distinguish whether production changes originate from the measures themselves or from natural reservoir dynamics. This application provides a method for evaluating the effectiveness of production enhancement measures based on CausalVAE. By using a CausalVAE causal representation model, the method explicitly decouples the measure factors from natural factors in the potential space. During evaluation, a counterfactual prediction curve is constructed as a scientific benchmark, enabling the quantitative calculation of the true contribution purely attributable to the measures after excluding natural fluctuations. This achieves accurate causal attribution of measure effects, effectively eliminating confounding effects and improving the accuracy of the evaluation results. Furthermore, this method does not rely on strong assumptions such as "stable production decline." Through the CausalVAE causal representation model, it learns complex nonlinear relationships from production data, effectively handling abrupt production fluctuations caused by measures, shutdowns, and other events. Even in complex conditions where traditional methods are ineffective, such as high water cut, strong coupling of multiple well groups, and unconventional reservoirs, it maintains stable evaluation performance and has wider applicability. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0043] Figure 1 A flowchart illustrating a method for evaluating the effectiveness of yield-increasing measures based on CausalVAE, provided in an embodiment of this application;

[0044] Figure 2 This is a schematic diagram of the structure of the CausalVAE causal representation model provided in the embodiments of this application;

[0045] Figure 3 This is a schematic diagram of a device for evaluating the effectiveness of production-increasing measures based on CausalVAE, provided in an embodiment of this application. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0047] To address the problems of difficulty in separating confusion effects, inaccurate evaluation, low evaluation efficiency, and insufficient adaptability in existing technologies, this application provides a method for evaluating the effectiveness of yield-increasing measures based on CausalVAE. Please refer to... Figure 1 This is a flowchart illustrating a method for evaluating the effectiveness of yield-increasing measures based on CausalVAE, as provided in an embodiment of this application. The following is a description of... Figure 1 The method for evaluating the effectiveness of yield-increasing measures based on CausalVAE is introduced.

[0048] S101. Obtain production data of the target oil well and adjacent wells over a continuous time period, and perform preprocessing to obtain structured multidimensional time-series data.

[0049] In one possible embodiment, the specific steps of S101 include:

[0050] Production data of the target oil well and adjacent wells are collected over a continuous period of time. The production data includes daily oil production, water cut, water injection volume, wellhead pressure, and pump speed. Missing values ​​are filled, outliers are removed, and the production data is normalized to obtain structured multidimensional time-series data.

[0051] In the specific implementation process, production data of the target well and adjacent wells are collected from the oilfield production system over a continuous period. The target well refers to a specific well that has undergone production enhancement measures (such as fracturing and acidizing) and requires effect evaluation. Adjacent wells refer to surrounding wells that, based on the oilfield development well network deployment relationship, dynamic connectivity analysis, or production interference testing, have potential hydraulic connectivity or dynamic production interference with the target well. Production data includes daily oil production, water cut, water injection volume, wellhead pressure, pump speed, and flowback flow rate. The data acquisition granularity can be set to daily or hourly levels depending on the actual application scenario.

[0052] Furthermore, for missing values ​​in production data caused by sensor malfunctions or transmission interruptions, linear interpolation can be used to fill in the missing values ​​and maintain the continuity of the time series. For outliers in production data caused by meter fluctuations or production anomalies, the Z-Score method can be used for detection and removal. Specifically, the mean μ and standard deviation σ of the production data series are calculated. For any data point (x), if its Z-Score value (μμμσ ... If the absolute value of a point is greater than a preset threshold, it is determined to be an anomaly and removed. In this embodiment, the preset threshold is preferably set to 3.

[0053] Finally, the production data after missing value imputation and outlier removal can be normalized to zero mean to obtain structured multidimensional time series data, thereby eliminating the influence of different dimensions and orders of magnitude on the model and laying a reliable data foundation for subsequent prediction and counterfactual inference of the CausalVAE causal model.

[0054] S102. Construct the CausalVAE causal representation model with an encoder-decoder architecture, and map multidimensional time series data into latent variables through the encoder.

[0055] In its implementation, the CausalVAE causal representation model, as a generative structure for causal representation learning, employs an encoder-decoder architecture. This architecture captures nonlinear causal relationships under complex operating conditions and maintains robustness to small samples and sudden fluctuations, making it particularly suitable for analyzing operating conditions such as old well restart, injection-production coupling, and high water cut stages. The multidimensional time-series data obtained from S101 is input into the encoder of the CausalVAE causal representation model. The encoder performs nonlinear transformation and feature extraction on the multidimensional time-series data, mapping it from a high-dimensional observation space to a low-dimensional latent space and outputting latent variables. Each latent variable consists of two decoupled components, denoted as (z...). c , z n ). Measures factor z c Natural factor z is used to characterize human intervention factors such as acidizing, fracturing, water injection, or pump speed adjustment. n It is used to characterize non-human intervention factors such as natural decline, pressure recovery, and interference from adjacent wells.

[0056] In one possible embodiment, constructing a CausalVAE causal representation model with an encoder-decoder architecture includes the following training steps:

[0057] Historical production data of the target oil well and adjacent wells over a continuous period of time are acquired and preprocessed to obtain structured historical multidimensional time-series data. A CausalVAE causal representation model with an encoder-decoder architecture is constructed, and a loss function containing a reconstruction loss term and a KL divergence regularization term is built. Based on the historical multidimensional time-series data, the parameters of the CausalVAE causal representation model are optimized and trained by minimizing the loss function.

[0058] In the specific implementation process, historical production data of the target oil well and adjacent wells over a continuous time period are acquired, and preprocessing steps as described in S101 (including missing value imputation, outlier removal, and normalization) are performed to obtain structured historical multidimensional time-series data. This historical multidimensional time-series data is then input into the CausalVAE causal representation model. The loss function is minimized using the backpropagation algorithm, thereby iteratively optimizing the parameters of the encoder and decoder in the CausalVAE causal representation model until the model converges.

[0059] In one possible implementation, the loss function is as follows:

[0060]

[0061] Among them, Q t Indicates the first The actual observed daily production value (i.e., daily oil production, in m³ / d). The first number obtained after reconstruction by the decoder Daily production forecast; W t Indicates the first Actual observed moisture content of the day (in %). The first number obtained after reconstruction by the decoder Daily moisture content prediction; z c This represents the measure factor extracted by the encoder; D represents the natural factor extracted by the encoder. KL (⋅) represents the KL divergence, used to constrain the distribution of latent variables; λ is the weight coefficient of the KL divergence; X represents historical multidimensional time series data; T represents the time series length of historical multidimensional time series data; q(,) represents the posterior distribution; p(,) represents the prior distribution.

[0062] In this embodiment, the loss function consists of two parts: a reconstruction loss term and a KL divergence regularization term. The reconstruction loss term aims to constrain the decoder output of the model to accurately reproduce the original input data, ensuring that the model captures the main variation patterns in the data. The KL divergence regularization term aims to constrain the distribution of the latent variables in the encoder output to approximate a preset prior distribution (usually a standard normal distribution). This KL divergence regularization term is key to enabling the model to learn a latent space with good structure (decoupling) and is the mathematical guarantee for achieving the separation of action factors and natural factors.

[0063] S103. At the moment of implementation of the production increase measures, the measure factor is set to zero, and the decoder generates counterfactual prediction curves of key production indicators without the implementation of production increase measures using only natural factors.

[0064] In the specific implementation process, the first step is to determine the specific time T at which the target oil well will implement production enhancement measures. m Then at that moment T m Set the action factor in the CausalVAE causal representation model to zero, i.e., z c =0, the physical meaning of this operation is that, in the causal system constructed by the model, the impact of the production-increasing measures is artificially "erased". Finally, the measure factor (z) after being set to zero is... c =0) and natural factor z nThis is input together to the decoder. The decoder uses this latent variable (z) as a basis. c =0,z n Data reconstruction is performed to generate future change curves for key production indicators under the hypothetical scenario of "no production increase measures implemented," i.e., counterfactual prediction curves.

[0065] Key production indicators include production rate and water cut. Counterfactual prediction curves include counterfactual prediction curves for production rate and water cut. The counterfactual prediction curve for production rate refers to the curve predicting the future production rate of the target oil well over time under the hypothetical scenario of "no production enhancement measures implemented," using the CausalVAE causal representation model of this invention. The counterfactual prediction curve for water cut refers to the curve predicting the future water cut of the target oil well over time under the hypothetical scenario of "no production enhancement measures implemented," using the CausalVAE causal representation model of this invention.

[0066] S104. Obtain the actual observation curves of key production indicators under the implementation of production increase measures.

[0067] In the specific implementation process, the actual production and actual water cut of the target oil well during the effect evaluation period, starting from the implementation time Tm of the production enhancement measures, are extracted from the oilfield production database to form actual observation curves. These actual observation curves include production and water cut curves. The production curve refers to the curve showing the actual production of the target oil well over time after the implementation of the production enhancement measures. The water cut curve refers to the curve showing the actual water cut of the target oil well over time after the implementation of the production enhancement measures.

[0068] S105. Compare the actual observed curves of key production indicators with the counterfactual predicted curves of key production indicators to obtain the effect evaluation results of production increase measures.

[0069] In one possible embodiment, the specific steps of S105 include:

[0070] S1.1 Compare the actual observed yield curve with the counterfactual yield prediction curve to obtain the effect evaluation results of the yield-increasing measures on the yield.

[0071] The evaluation results of the production increase measures include the contribution of the measures on a single day and the cumulative contribution of the measures over multiple days. The contribution of the measures on a single day is used to finely evaluate the immediate production increase effect of the measures on any single day, while the cumulative contribution of the measures over multiple days is used to evaluate the overall production increase effect of the measures during the evaluation period, and is the direct basis for the economic benefit evaluation.

[0072] For any day within the assessment period, the actual observed output value for that day is determined based on the actual output observation curve, and the counterfactual predicted output value for that day is determined based on the counterfactual predicted output curve. The difference between the actual observed output value and the counterfactual predicted output value for that day is calculated to obtain the daily contribution of the measures.

[0073] The calculation method for the daily contribution of measures is as follows:

[0074]

[0075] Among them, Effect t This indicates that the production increase measures were implemented in the [number]th [year]. Daily contribution; Indicates the first Actual observed daily yield; Indicates the first Daily counterfactual forecast of output value.

[0076] Furthermore, after obtaining the contribution of a single day's measures, the contribution of a single day's measures over a continuous period within the assessment period is summed to obtain the cumulative contribution of measures over multiple days.

[0077] The cumulative contribution of measures taken over multiple days is calculated as follows:

[0078]

[0079] Here, CE represents the cumulative contribution of production increase measures over 30 consecutive days; Effect t This indicates that the production increase measures were implemented in the [number]th [year]. Daily contribution: T m The value represents the time when the production-increasing measures were implemented; N represents the number of days for effect evaluation. In this embodiment, N is preferably set to 30 days.

[0080] In the embodiments of this application, the actual increase in production brought about by the production enhancement measures on each day and even throughout the entire cycle can be accurately calculated. The contribution of the measures can be accurately separated from the mixed production data, effectively solving the problem of distorted evaluation results caused by the confusion effect of traditional methods, and providing reliable data support for oilfield production decisions and measure optimization.

[0081] S1.2. By comparing the actual observed curve of moisture content with the counterfactual predicted curve of moisture content, the effect evaluation results of the yield-increasing measures on moisture content can be obtained.

[0082] The evaluation results of the yield-increasing measures on moisture content include the improvement in moisture content, which is used to quantitatively characterize the net effect of the yield-increasing measures in controlling moisture content. The specific calculation process is as follows:

[0083] For any day within the assessment period, the actual observed moisture content of that day is determined based on the actual observed moisture content curve, and the counterfactual predicted moisture content of that day is determined based on the counterfactual predicted moisture content curve. The difference between the actual observed moisture content and the counterfactual predicted moisture content of that day is calculated, and the average value of the difference over a continuous period within the assessment period is obtained to obtain the moisture content improvement.

[0084] The calculation method for moisture content improvement is as follows:

[0085]

[0086] Wherein, ΔW represents the improvement in moisture content; Indicates the first Daily moisture content observation value, Indicates the first Daily moisture content forecast; T m The value N represents the time when the production-increasing measures were implemented, and N is the number of days for effect evaluation. In this embodiment, N is preferably set to 30 days.

[0087] In the embodiments of this application, the moisture content improvement ΔW obtained through the above process effectively eliminates the influence of non-measure factors such as natural decline and pressure recovery on moisture content changes, thereby scientifically and quantitatively revealing the true control capability of the yield-increasing measures themselves on moisture content.

[0088] In one possible implementation, the evaluation results of production enhancement measures are pushed to the oilfield production evaluation system in JSON format via an API interface, such as / api / effect / attribution. This achieves seamless integration of the evaluation process with the existing production management system, eliminates data silos, and supports real-time decision-making.

[0089] In one possible implementation, the model's training data is updated every 30 days, enabling the model to continuously learn about reservoir dynamics and refresh counterfactual prediction curves and evaluation results. This periodic self-updating mechanism ensures that the CausalVAE causal model always reflects the current reservoir state, forming a complete closed loop of "evaluation-learning-re-evaluation," and realizing the transformation from single evaluation to continuous optimization.

[0090] To further illustrate the CausalVAE-based method for evaluating the effectiveness of yield-increasing measures provided in this application, the following detailed explanation will be provided in conjunction with two specific embodiments.

[0091] Example 1:

[0092] In a horizontal shale oil well with insufficient production, supplementary fracturing was implemented on day 60, with a designed flow rate of 12 m³ / min and a total fluid volume of 1800 m³. The well's production increased after fracturing. To evaluate the true effectiveness of the fracturing measure, the CausalVAE-based production enhancement measure effectiveness evaluation method provided in this application was applied. The specific implementation steps are as follows:

[0093] Step 1: Collect production data for the well and adjacent wells for the past 90 days on a daily basis. Perform missing value filling, outlier removal and normalization on the production data to obtain structured multidimensional time series data.

[0094] Step 2: Input 60 consecutive days of multidimensional time-series data into the pre-built CausalVAE causal representation model. The encoder maps the input data into latent variables (z). c , z n ), where z c z represents the fracturing action factor. n Represents natural factors.

[0095] Step 3: At the moment of implementation of fracturing measures, set the measure factor to zero and obtain the counterfactual prediction curve through the encoder.

[0096] Step 4: Based on the counterfactual prediction curve and the actual observation curve, the daily contribution of the fracturing measures was calculated to be 0.4 m³, and the cumulative contribution over 30 days was 129 m³. Since the well's production increased by approximately 0.5 m³ on the first day after fracturing, with the actual contribution from the fracturing measures being 0.4 m³ and the natural fluctuation contribution being 0.1 m³, the well's cumulative production increase over 30 days after fracturing was 172 m³, with the actual contribution from the fracturing measures being 129 m³ and the natural contribution being approximately 43 m³. The well's actual production increased by 10.5% after fracturing, with the water cut fluctuating between 85% and 1%. The results indicate that the fracturing measures played a major role in increasing the well's production, and the evaluation is consistent with the field findings.

[0097] Example 2:

[0098] In a certain old heavy oil well, long-term high water-cut extraction led to near-wellbore blockage and production depletion, with daily production dropping to approximately 3 m³ / d. Due to fluid accumulation in the wellbore, production was forced to stop for 20 days on day 130. Before resuming production on day 150, acidizing was implemented, with a designed acid volume of 35 m³, a concentration of 10%, and an operating pressure controlled at 12 MPa. In the initial stage of resuming production, daily production increased from 3.2 m³ / d to 4.5 m³ / d, showing a significant increase. To evaluate the actual effectiveness of the acidizing measures, the CausalVAE-based production enhancement measure effectiveness evaluation method provided in this application was applied. The specific implementation steps are as follows:

[0099] Step 1: Collect production data of the well and adjacent wells on a daily basis for 100 days before and after acidizing. Perform missing value filling, outlier removal and normalization on the production data to obtain structured multidimensional time series data.

[0100] Step 2: Input the multidimensional time series data into the pre-built CausalVAE causal representation model. The encoder maps the input data into latent variables (z). c , z n ), where z c The acidification factor, z n Represents natural factors.

[0101] Step 3: At the moment of implementation of the acidification measures, set the acidification measure factor to zero and obtain the counterfactual prediction curve through the decoder.

[0102] Step 4: Based on the counterfactual prediction curve and the actual observation curve, the daily contribution of the measures was calculated to be 0.2 m³, and the cumulative contribution over 30 days was 3 m³. Since the apparent production increase on the first day after acidizing was 1.3 m³, with the acidizing measures contributing only 0.2 m³, the natural pressure recovery contribution was +1.1 m³. Since the cumulative production increase over 30 days after acidizing was 39 m³, with the acidizing measures contributing only 3 m³, the natural contribution was 36 m³. The actual production of this well increased by approximately 40 m³ within 30 days after acidizing, with a cumulative increase rate of approximately 10.8%, but the water cut remained relatively stable between 83% and 85%. The results indicate that the vast majority of the production increase came from the natural pressure recovery after production shutdown, with the acidizing measures contributing only about 8%, indicating a limited effect, which is consistent with the conclusions drawn from field monitoring.

[0103] In summary, this application provides a method for evaluating the effectiveness of production enhancement measures based on CausalVAE. By using the CausalVAE causal representation model, multi-source time-series data inputs such as production, water cut, water injection volume, and pressure are decomposed into measure factors and natural factors, thus decoupling the potential causal sources of production data. Based on the causal decomposition results, a counterfactual prediction strategy is introduced to construct a control curve under the "no measures implemented" scenario. The actual observed curve is compared with the counterfactual prediction curve to quantitatively calculate the contribution of the measures. This method not only quantifies the direct impact of individual measures on production and water cut but also effectively eliminates the confounding effects of natural decline, interference from adjacent wells, and pressure recovery, improving the accuracy of the evaluation. This invention ultimately forms a closed-loop system architecture of "modeling—decomposition—counterfactual prediction—attribution," supporting edge deployment and rolling application, significantly improving the scientific rigor, reliability, and real-time performance of measure effectiveness evaluation. This method is applicable to various common oil and gas production measure scenarios such as acidizing, fracturing, water injection adjustment, and pump rate optimization, and has good prospects for widespread application.

[0104] Based on the same inventive concept, please refer to Figure 3 This application also provides a CausalVAE-based system for evaluating the effectiveness of yield-increasing measures, the system comprising:

[0105] The preprocessing module is used to acquire production data of the target oil well and adjacent wells over a continuous time period, and to preprocess the data to obtain structured multidimensional time-series data.

[0106] The prediction module is used to construct the CausalVAE causal representation model with an encoder-decoder architecture. The encoder maps multidimensional time series data into latent variables. The latent variables include action factors that characterize human intervention factors and natural factors that characterize non-human intervention factors. At the time of implementation of production-increasing measures, the action factors are set to zero, and the decoder uses only natural factors to generate counterfactual prediction curves of key production indicators without the implementation of production-increasing measures.

[0107] The data acquisition module is used to obtain the actual observation curves of key production indicators under the implementation of production increase measures;

[0108] The effectiveness evaluation module is used to compare the actual observed curves of key production indicators with the counterfactual predicted curves of key production indicators to obtain the effectiveness evaluation results of production increase measures.

[0109] Optional key production indicators include yield and moisture content; the performance evaluation module is specifically used for:

[0110] By comparing the actual observed yield curve with the counterfactual yield prediction curve, the effect of yield-increasing measures on yield can be evaluated.

[0111] By comparing the actual observed moisture content curve with the counterfactual predicted moisture content curve, the effect of yield-increasing measures on moisture content can be evaluated.

[0112] Optionally, the evaluation results of the production-increasing measures' impact on output include the daily contribution of the measures; the effect evaluation module is specifically used for:

[0113] For any day within the assessment period, the actual observed output value for that day is determined based on the actual observed output curve, and the counterfactual predicted output value for that day is determined based on the counterfactual predicted output curve.

[0114] The difference between the actual observed output value and the counterfactual predicted output value for any given day is calculated to obtain the contribution of the measures on that day.

[0115] Optionally, the evaluation results of the production increase measures on output also include the cumulative contribution of measures over multiple days; the effect evaluation module is also used to: after calculating the difference between the actual observed output value and the counterfactual predicted output value on any given day to obtain the contribution of measures on a single day, sum the contribution of measures on a single day over a continuous period within the evaluation period to obtain the cumulative contribution of measures over multiple days.

[0116] Optionally, the evaluation results of the yield-increasing measures on moisture content include the amount of moisture content improvement; the effect evaluation module is specifically used for:

[0117] For any day within the assessment period, the actual observed moisture content of any day is determined based on the actual observed moisture content curve, and the counterfactual predicted moisture content of any day is determined based on the counterfactual predicted moisture content curve.

[0118] Calculate the difference between the actual observed moisture content and the counterfactual predicted moisture content for any given day, and average the difference over consecutive time periods within the assessment period to obtain the moisture content improvement.

[0119] Optionally, the device also includes a training module, which is used for:

[0120] Historical production data of the target oil well and adjacent wells over a continuous period of time are acquired and preprocessed to obtain structured historical multidimensional time-series data.

[0121] Construct a CausalVAE causal representation model with an encoder-decoder architecture, and build a loss function that includes a reconstruction loss term and a KL divergence regularization term;

[0122] Based on historical multidimensional time-series data, the parameters of the CausalVAE causal representation model are optimized and trained by minimizing the loss function.

[0123] Optional, the loss function is as follows:

[0124]

[0125] Among them, Q t Indicates the first Actual observed daily yield; The first number obtained after reconstruction by the decoder Daily projected output value; W t Indicates the first The actual observed moisture content on that day; The first number obtained after reconstruction by the decoder Predicted moisture content for the day; z c This represents the measure factor extracted by the encoder; D represents the natural factor extracted by the encoder. KL(⋅) represents the KL divergence; λ is the weight coefficient of the KL divergence; X represents the historical multidimensional time series data; T represents the time series length of the historical multidimensional time series data; q(,) represents the posterior distribution; p(,) represents the prior distribution.

[0126] Optionally, the preprocessing module is specifically used for:

[0127] Collect production data of the target oil well and adjacent wells over a continuous period of time; production data includes daily oil production, water cut, water injection volume, wellhead pressure, and pump speed;

[0128] The production data is filled with missing values, outliers are removed, and normalized to obtain structured multidimensional time-series data.

[0129] It should be noted that each module in the CausalVAE-based yield enhancement measure effect evaluation device in this embodiment corresponds one-to-one with each step in the CausalVAE-based yield enhancement measure effect evaluation method in the aforementioned embodiment. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned CausalVAE-based yield enhancement measure effect evaluation method, which will not be repeated here.

[0130] Based on the same inventive concept, this application also provides a computer device, which includes a processor, a memory, and a computer program stored in the memory. The computer program is executed by the processor to implement the aforementioned method for evaluating the effectiveness of production increase measures based on CausalVAE.

[0131] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0132] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0133] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the effectiveness of yield-increasing measures based on CausalVAE, characterized in that, include: Acquire production data of the target oil well and adjacent wells over a continuous time period, and preprocess the data to obtain structured multidimensional time-series data; A CausalVAE causal representation model with an encoder-decoder architecture is constructed, in which the encoder maps the multidimensional time series data into latent variables; the latent variables include measure factors for characterizing human intervention factors and natural factors for characterizing non-human intervention factors. At the moment of implementation of the production increase measures, the measure factor is set to zero, and the decoder generates counterfactual prediction curves of key production indicators without the implementation of production increase measures using only the natural factors. Obtain actual observation curves of key production indicators under the implementation of the aforementioned production increase measures; By comparing the actual observed curves of the key production indicators with the counterfactual predicted curves of the key production indicators, the effectiveness evaluation results of the production increase measures are obtained. The construction of the CausalVAE causal representation model with an encoder-decoder architecture includes the following training steps: acquiring historical production data of the target oil well and adjacent wells over a continuous time period, and preprocessing it to obtain structured historical multidimensional time-series data; constructing the CausalVAE causal representation model with an encoder-decoder architecture, and constructing a loss function that includes a reconstruction loss term and a KL divergence regularization term; and optimizing the parameters of the CausalVAE causal representation model by minimizing the loss function based on the historical multidimensional time-series data.

2. The method for evaluating the effectiveness of yield-increasing measures based on CausalVAE according to claim 1, characterized in that, The key production indicators include yield and moisture content; the comparison of the actual observed curves and the counterfactual predicted curves of the key production indicators to obtain the effectiveness evaluation results of the yield-increasing measures includes: By comparing the actual observed curve of the output with the counterfactual predicted curve of the output, the effect evaluation results of the output-increasing measures on the output are obtained; By comparing the actual observed curve of the moisture content with the counterfactual predicted curve of the moisture content, the effect evaluation results of the yield-increasing measures on the moisture content are obtained.

3. The method for evaluating the effectiveness of yield-increasing measures based on CausalVAE according to claim 2, characterized in that, The evaluation results of the production-increasing measures' effect on output include the daily contribution of the measures; the comparison of the actual observed output curve and the counterfactual predicted output curve to obtain the evaluation results of the production-increasing measures' effect on output includes: For any day within the assessment period, the actual observed output value for that day is determined based on the actual observed output curve, and the counterfactual predicted output value for that day is determined based on the counterfactual predicted output curve. The difference between the actual observed output value and the counterfactual predicted output value for any given day is calculated to obtain the contribution of the measures on that day.

4. The method for evaluating the effectiveness of yield-increasing measures based on CausalVAE according to claim 3, characterized in that, The evaluation of the effectiveness of the production-increasing measures on output also includes the cumulative contribution of measures over multiple days; after calculating the difference between the actual observed output value and the counterfactual predicted output value for any given day to obtain the contribution of the measures on a single day, the method further includes: The cumulative contribution of the measures over multiple days is obtained by summing the daily contribution amounts of the measures taken within a continuous time period during the assessment period.

5. The method for evaluating the effectiveness of yield-increasing measures based on CausalVAE according to claim 2, characterized in that, The evaluation results of the yield-increasing measures on the moisture content include the amount of moisture content improvement; the comparison of the actual observed moisture content curve and the counterfactual predicted moisture content curve to obtain the evaluation results of the yield-increasing measures on the moisture content includes: For any day within the assessment period, the actual observed moisture content of that day is determined based on the actual observed moisture content curve, and the counterfactual predicted moisture content of that day is determined based on the counterfactual predicted moisture content curve. Calculate the difference between the actual observed moisture content and the counterfactual predicted moisture content for any given day, and average the difference over a continuous period within the assessment period to obtain the moisture content improvement.

6. The method for evaluating the effectiveness of yield-increasing measures based on CausalVAE according to claim 1, characterized in that, The loss function is as follows: Among them, Q t This represents the actual observed yield value on day t; W represents the predicted output value for day t, reconstructed by the decoder. t This represents the actual observed moisture content on day t. This represents the predicted moisture content for day t, reconstructed by the decoder; z c This represents the measure factor extracted by the encoder; z n D represents the natural factor extracted by the encoder; KL (·) represents the KL divergence; λ is the weighting coefficient of the KL divergence; X represents the historical multidimensional time series data; T represents the time series length of the historical multidimensional time series data; q(,) represents the posterior distribution; p(,) represents the prior distribution.

7. The method for evaluating the effectiveness of yield-increasing measures based on CausalVAE according to claim 1, characterized in that, The process of acquiring production data from the target oil well and adjacent wells over a continuous time period, and preprocessing it to obtain structured multidimensional time-series data, includes: Collect production data of the target oil well and adjacent wells over a continuous period of time; the production data includes daily oil production, water cut, water injection volume, wellhead pressure, and pump speed; The production data is filled with missing values, outliers are removed, and normalized to obtain structured multidimensional time-series data.

8. A system for evaluating the effectiveness of yield-increasing measures based on CausalVAE, characterized in that, The method for evaluating the effectiveness of yield-increasing measures based on CausalVAE as described in any one of claims 1-7 is employed; the system comprises: The preprocessing module is used to acquire production data of the target oil well and adjacent wells over a continuous time period, and to preprocess the data to obtain structured multidimensional time-series data. The prediction module is used to construct a CausalVAE causal representation model with an encoder-decoder architecture. The encoder maps the multidimensional time series data into latent variables. The latent variables include measure factors that characterize human intervention factors and natural factors that characterize non-human intervention factors. At the time of implementation of production-increasing measures, the measure factors are set to zero, and the decoder uses only the natural factors to generate counterfactual prediction curves of key production indicators under the condition that no production-increasing measures have been implemented. The data acquisition module is used to obtain the actual observation curves of key production indicators under the implementation of the aforementioned production increase measures; The effect evaluation module is used to compare the actual observed curves of the key production indicators with the counterfactual prediction curves of the key production indicators to obtain the effect evaluation results of the production increase measures.

9. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method for evaluating the effectiveness of production-increasing measures based on CausalVAE as described in any one of claims 1-7.

Images