A shale oil in-situ development stage production prediction method and related device

Abstract

The present application belongs to the field of unconventional oil and gas resource development, and discloses a shale oil in-situ development stage production prediction method and related device. Through obtaining multi-source dynamic monitoring data such as temperature field, pressure field, fluid component change and reservoir property in shale oil in-situ development, the development process is divided into early, middle and late stages according to the preset standard, and the key characteristic variables of each stage are identified. Then, the time series analysis method is used to extract and reduce the data characteristics to form a characteristic vector. Then, the characteristic vector is used to train a unified adaptive framework integrating independent prediction models for early, middle and late stages, wherein each model is trained using the key characteristic variables of the corresponding stage. Finally, the real-time monitoring data is input to output the production prediction result. The present method significantly enhances the pertinence, generalization ability and interpretability of the prediction, and realizes high-precision production prediction for the whole cycle of in-situ development.

Description

Technical Field

[0001] This invention belongs to the field of unconventional oil and gas resource development technology, and in particular relates to a method and related apparatus for predicting the production of shale oil in the in-situ development stage. Background Technology

[0002] With the continued advancement of unconventional oil and gas resource development, shale oil in-situ development technology, with its advantages in efficiently utilizing low-permeability and difficult-to-access reserves, has become one of the core development directions in the current oil and gas exploration and development field. In the entire process of shale oil in-situ development, accurate production prediction at different development stages is a key technical support for optimizing development scheme design, improving resource recovery rates, and reducing investment and operational risks, directly affecting the economic benefits and implementation effectiveness of development projects. In recent years, machine learning methods, due to their powerful nonlinear fitting capabilities and good adaptability to complex geological engineering data, have been gradually introduced into the field of oil and gas production prediction, providing a new technical path to overcome the limitations of traditional prediction methods.

[0003] Existing machine learning-based production prediction technologies still have significant shortcomings in the specific scenario of shale oil in-situ development. For example, Chinese Patent Publication No. CN116151480A proposes a method and device for predicting shale oil well production, which achieves iterative prediction at arbitrary step lengths through dynamic and static parameter fusion and embedding mechanisms. However, this method is mainly aimed at the production stage of conventional shale oil wells and does not fully consider key factors unique to shale oil in-situ development, such as the coupling of multiple physical fields (thermal-mechanical-chemical), heating cycle changes, phase transition processes, and dynamic evolution of reservoir properties. This results in a lack of specificity and physical consistency in production prediction during the early or middle stages of in-situ development, and it relies on a large amount of historical daily oil production data for training, making it difficult to effectively model when data is sparse or missing in the early stages of in-situ development. Another example is the AI-based oil well production prediction system and method proposed in Chinese Patent Publication No. CN120373562A. Although it uses a recurrent neural network combined with an artificial lemming optimization algorithm to improve prediction accuracy and efficiency, this method belongs to a general oil well production prediction framework. It does not carry out feature engineering design or model structure adaptation for special working conditions such as electric heating, steam injection, and in-situ cracking in shale oil development, nor does it integrate key dynamic parameters such as temperature field, pressure field, and fluid composition changes. When applied to shale oil in-situ development scenarios, it is difficult to capture the intrinsic mechanism of production changes, and the interpretability and generalization ability of the prediction results are limited.

[0004] It is evident that existing machine learning-based production prediction methods generally suffer from technical problems such as insufficient integration of physical mechanisms, lack of stage feature modeling, and poor adaptability to small samples in the special scenario of shale oil in-situ development, which is highly unsteady, strongly coupled, and has sparse data. These problems make it difficult to meet the actual needs of accurate production prediction in shale oil in-situ development. Summary of the Invention

[0005] This invention provides a method and related apparatus for predicting production during the in-situ development stage of shale oil. This method can effectively solve the problems of insufficient integration of physical mechanisms, lack of stage feature modeling, and poor adaptability to small samples in existing machine learning-based production prediction methods. This method can meet the actual needs of accurate production prediction in in-situ shale oil development.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for predicting shale oil production during in-situ development includes:

[0008] Acquire multi-source dynamic monitoring data during the in-situ development of shale oil; the multi-source dynamic monitoring data includes temperature field, pressure field, fluid composition changes, and reservoir physical property parameters;

[0009] Based on the pre-defined development phase division criteria, the development phase is divided into three stages: early, mid, and late; and the key characteristic variables corresponding to each stage are determined.

[0010] Time series analysis was used to extract features and reduce dimensionality of multi-source dynamic monitoring data to obtain feature vectors.

[0011] The pre-constructed adaptive prediction model for the entire development cycle is trained using feature vectors; wherein, the adaptive prediction model for the entire development cycle includes an early stage prediction model, a mid-stage prediction model, and a late stage prediction model integrated into a unified framework; the early stage prediction model, the mid-stage prediction model, and the late stage prediction model are trained using key feature variables corresponding to each stage in the feature vectors, respectively, to output the trained adaptive prediction model for the entire development cycle.

[0012] The collected real-time monitoring data during the development phase is input into an adaptive prediction model that covers the entire development cycle, and the output is the production prediction result for the in-situ development phase of shale oil.

[0013] Furthermore, the acquisition of multi-source dynamic monitoring data during the in-situ development of shale oil includes:

[0014] During the in-situ development of shale oil, multi-source dynamic monitoring data is acquired, including: temperature field data acquired through distributed fiber optic sensing technology, pressure field data recorded through downhole pressure gauges, fluid component change data acquired through gas chromatography analysis, and reservoir physical parameters acquired through laboratory tests or field logging data.

[0015] Furthermore, in the step of dividing the development stage into three stages—early, mid, and late—according to a preset development stage division standard, the development stage division standard is as follows:

[0016] When the average reservoir temperature reaches the preset temperature threshold, it indicates that the development stage has entered the middle stage from the early stage.

[0017] When the production growth rate is lower than the preset growth rate, it indicates that the development stage has moved from the middle stage to the late stage.

[0018] Furthermore, the step of using time series analysis to extract features and reduce the dimensionality of multi-source dynamic monitoring data to obtain feature vectors includes:

[0019] A sliding window method is used to segment multi-source dynamic monitoring data into time series data; each window includes data for a preset time span.

[0020] Principal component analysis was used to reduce the dimensionality of the data within the window, and feature vectors reflecting the main trends in data change were extracted.

[0021] Furthermore, the step of training a pre-built adaptive prediction model for the entire development cycle using feature vectors includes:

[0022] An adaptive prediction model for the entire development cycle is constructed, which integrates early-stage, mid-stage, and late-stage prediction models into a unified framework. The early-stage prediction model is based on a support vector regression model; the mid-stage prediction model is based on a random forest model; and the late-stage prediction model is based on a long short-term memory network model.

[0023] In the process of integrating the early-stage, mid-stage, and late-stage prediction models into a unified framework, a weighted fusion strategy is used to integrate the prediction results output by the early-stage, mid-stage, and late-stage prediction models. The specific formula is as follows:

[0024] y=∑w i ·y i

[0025] In the formula, y represents the final prediction result. i w represents the prediction result of the i-th sub-model. i These are the corresponding weighting coefficients, which are automatically adjusted based on cross-validation or Bayesian optimization algorithms.

[0026] Furthermore, the step of inputting the collected real-time monitoring data from the development phase into an adaptive prediction model oriented towards the entire development cycle, and outputting the shale oil in-situ development phase production prediction results, includes:

[0027] Identify the development stage corresponding to the real-time monitoring data during the development phase;

[0028] The system invokes the early-stage, mid-stage, or late-stage prediction models from the adaptive prediction model for the entire development cycle to predict real-time monitoring data during the development stage and outputs the production prediction results for the in-situ development stage of shale oil.

[0029] Furthermore, after inputting the collected real-time monitoring data from the development phase into an adaptive prediction model oriented towards the entire development cycle and outputting the shale oil in-situ development phase production prediction results, the process also includes:

[0030] By analyzing the error between the production prediction results and the actual monitoring results during the in-situ development stage of shale oil, the adaptive prediction model for the entire development cycle is corrected.

[0031] A visualization report is generated based on the output production forecast results of the in-situ development stage of shale oil; wherein, the visualization report includes a production change curve over time, a prediction error distribution map for each stage, and an importance ranking map of key feature variables.

[0032] A shale oil production prediction system for in-situ development includes:

[0033] The data acquisition module is used to acquire multi-source dynamic monitoring data during the in-situ development of shale oil; the multi-source dynamic monitoring data includes temperature field, pressure field, fluid composition changes and reservoir physical parameters;

[0034] The segmentation module is used to divide the development stage into three stages—early, mid, and late—based on a preset development stage segmentation standard; and to determine the key feature variables corresponding to each stage.

[0035] The feature extraction module is used to extract and reduce the dimensions of multi-source dynamic monitoring data using time series analysis methods to obtain feature vectors.

[0036] The model training module is used to train a pre-built adaptive prediction model for the entire development cycle using feature vectors. The adaptive prediction model for the entire development cycle includes an early-stage prediction model, a mid-stage prediction model, and a late-stage prediction model integrated into a unified framework. The early-stage prediction model, the mid-stage prediction model, and the late-stage prediction model are trained using key feature variables corresponding to each stage in the feature vectors, respectively, to output the trained adaptive prediction model for the entire development cycle.

[0037] The production forecasting module is used to input the real-time monitoring data collected during the development phase into an adaptive forecasting model that covers the entire development cycle, and output the production forecast results for the in-situ development phase of shale oil.

[0038] A shale oil production prediction device for in-situ development, comprising:

[0039] Memory, used to store computer programs;

[0040] A processor is used to execute the computer program to implement the steps of the above-described shale oil in-situ development stage production prediction method.

[0041] A computer-readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of the above-described method for predicting production during the in-situ development stage of shale oil.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] This invention provides a method for predicting shale oil production during in-situ development. It acquires multi-source dynamic monitoring data, including temperature, pressure, fluid composition variations, and reservoir properties, from shale oil in-situ development. Based on preset standards, the development process is divided into early, middle, and late stages, and key characteristic variables for each stage are identified. Subsequently, time series analysis is used to extract and reduce the dimensionality of data features, forming feature vectors. These feature vectors are then used to train a unified adaptive framework integrating independent prediction models for the early, middle, and late stages, where each model is trained using the key characteristic variables of its corresponding stage. Finally, real-time monitoring data is input to output the production prediction results. This method incorporates the unique mechanisms of in-situ development, such as the coupling of thermo-mechanical-chemical multi-physics fields, heating cycle changes, phase transition processes, and reservoir dynamic evolution, into the model structure through a phased modeling strategy. Simultaneously, by leveraging feature extraction and dimensionality reduction to address data sparsity, it ensures effective capture of the inherent laws governing production changes under small sample conditions, improving the model's physical consistency and stage adaptability. This method significantly enhances the specificity, generalization ability, and interpretability of predictions. It overcomes the problems of insufficient integration of physical mechanisms, lack of stage feature modeling, and poor adaptability to small samples in existing technologies, and achieves high-precision yield prediction for the entire in-situ development cycle. Attached Figure Description

[0044] Figure 1 A flowchart illustrating the implementation of a shale oil production prediction method during in-situ development, as provided in this embodiment of the invention.

[0045] Figure 2 A comparison chart of early development stage production forecast results and actual monitoring results provided for embodiments of the present invention;

[0046] Figure 3 This is a comparison chart of the mid-term development stage production forecast results and actual monitoring results provided in an embodiment of the present invention;

[0047] Figure 4 This is a schematic diagram of the sub-model integration framework for different development stages provided in the embodiments of the present invention;

[0048] Figure 5 A flowchart illustrating a method for predicting shale oil production during in-situ development, as provided in this embodiment of the invention;

[0049] Figure 6 This is a schematic diagram of a shale oil production prediction system for the in-situ development stage, provided as an embodiment of the present invention. Detailed Implementation

[0050] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0051] The technical terms involved in this invention are explained below:

[0052] PCA stands for Principal Component Analysis.

[0053] KPCA stands for Kernel Principal Component Analysis.

[0054] RBF stands for Radial Basis Function.

[0055] SVR stands for Support Vector Regression.

[0056] LSTM stands for Long Short-Term Memory.

[0057] RF stands for Random Forest.

[0058] RNN stands for Recurrent Neural Networks.

[0059] As mentioned in the background section, in the in-situ development of shale oil, traditional production prediction methods struggle to accurately capture production characteristics at different development stages due to factors such as complex reservoir properties, significant coupling effects of thermo-mechanical-chemical multi-physics fields, and dynamic changes in heating cycles. While existing machine learning-based production prediction methods have achieved some success in conventional oil and gas wells or shale gas wells, they generally suffer from shortcomings in the highly unsteady, strongly coupled, and data-sparse special scenario of shale oil in-situ development. These shortcomings include insufficient integration of physical mechanisms, lack of stage feature modeling, and poor adaptability to small samples. Therefore, there is an urgent need for a production prediction method that can deeply integrate multi-source dynamic monitoring data and physical evolution laws in in-situ development to improve prediction accuracy, interpretability, and engineering applicability, providing technical support for efficient in-situ development of shale oil.

[0060] To achieve the above objectives, this embodiment provides a method for predicting shale oil production during in-situ development. This method uses dynamic feature extraction and phased modeling techniques to accurately capture complex thermo-mechanical-chemical coupling effects, and combines physical constraints and error feedback mechanisms to improve the model's adaptability and long-term prediction stability under small sample conditions.

[0061] like Figure 5 As shown in the figure, this embodiment provides a method for predicting shale oil production during the in-situ development stage, including:

[0062] Acquire multi-source dynamic monitoring data during the in-situ development of shale oil; the multi-source dynamic monitoring data includes temperature field, pressure field, fluid composition changes, and reservoir physical property parameters;

[0063] Based on the pre-defined development phase division criteria, the development phase is divided into three stages: early, mid, and late; and the key characteristic variables corresponding to each stage are determined.

[0064] Time series analysis was used to extract features and reduce dimensionality of multi-source dynamic monitoring data to obtain feature vectors.

[0065] The pre-constructed adaptive prediction model for the entire development cycle is trained using feature vectors; wherein, the adaptive prediction model for the entire development cycle includes an early stage prediction model, a mid-stage prediction model, and a late stage prediction model integrated into a unified framework; the early stage prediction model, the mid-stage prediction model, and the late stage prediction model are trained using key feature variables corresponding to each stage in the feature vectors, respectively, to output the trained adaptive prediction model for the entire development cycle.

[0066] The collected real-time monitoring data during the development phase is input into an adaptive prediction model that covers the entire development cycle, and the output is the production prediction result for the in-situ development phase of shale oil.

[0067] The prediction method provided in this embodiment will be further explained below with reference to the accompanying drawings:

[0068] like Figure 1 As shown in the figure, this embodiment provides a method for predicting shale oil production during the in-situ development stage. It can be seen that the overall process of this method includes data collection, development stage division, dynamic feature extraction, sub-model training, model integration, real-time prediction, and error correction. In practical applications, these steps unfold sequentially and are interconnected, forming a complete prediction framework. Taking a shale oil in-situ development project as an example, assuming the project is located in a certain region with complex reservoir properties and significant dynamic changes in the heating cycle, traditional methods are difficult to accurately predict production. To solve the above problems, the specific implementation steps of this shale oil in-situ development stage production prediction method are as follows:

[0069] Step 1: Obtain multi-source dynamic monitoring data during the in-situ development of shale oil:

[0070] Multi-source dynamic monitoring data mainly includes temperature field, pressure field, fluid composition changes, and reservoir physical parameters. For example, in actual operation, the spatiotemporal distribution data of the reservoir temperature field can be obtained through distributed fiber optic sensing technology, the changes in the pressure field can be recorded using downhole pressure gauges, and the changing trends of fluid components can be analyzed using gas chromatography. In addition, reservoir physical parameters such as porosity and permeability can be obtained through laboratory tests or field logging data. These data form the basis for subsequent modeling, and their quality and completeness directly affect the prediction accuracy. To facilitate subsequent processing, all data need to be preprocessed, including removing outliers, filling missing values, and normalization. For example, for temperature field data, this embodiment can use linear interpolation to fill missing values ​​and scale it to the [0, 1] interval using the min-max normalization method.

[0071] Step 2: Based on the pre-defined development phase division criteria, divide the development phase into three stages: early, mid, and late; determine the key characteristic variables corresponding to each stage:

[0072] According to the development stage classification criteria, the entire development process is divided into three stages: early, middle, and late, and key characteristic variables for each stage are identified. Specifically, the early stage typically corresponds to the initial heating phase, with low reservoir temperature and poor fluid flow; the middle stage is characterized by a gradual increase in reservoir temperature and the beginning of fluid flow; and the late stage is characterized by a stable reservoir temperature and a gradual decrease in production. In this embodiment, analysis of historical data reveals that when the average reservoir temperature reaches a certain preset temperature threshold, it can be considered as transitioning from the early to the middle stage, while when the production growth rate falls below a certain preset growth rate, it is considered as entering the late stage. For each stage, key characteristic variables are further identified. For example, in the early stage, initial reservoir temperature and pressure are the main influencing factors; in the middle stage, changes in fluid composition and reservoir permeability become key variables; and in the late stage, the reservoir pressure decay rate and fluid viscosity dominate.

[0073] Step 3: Construct a machine learning model based on dynamic feature extraction. Use time series analysis methods to extract features and reduce the dimensionality of multi-source dynamic monitoring data to obtain feature vectors.

[0074] First, the time series data is segmented using a sliding window technique, with each window containing data points over a certain time span. Then, Principal Component Analysis (PCA) is used to reduce the dimensionality of the data within each window, extracting feature vectors that reflect the main trends in data change. For example, assuming the original data dimension is n, after PCA processing, the first k principal components are retained as input features, where k is much smaller than n. Furthermore, to enhance the model's nonlinear expressive power, this embodiment of PCA can also incorporate Kernel Principal Component Analysis (KPCA), projecting the data to a higher-dimensional space via kernel function mapping before dimensionality reduction. In this embodiment, the Radial Basis Function (RBF) is chosen as the kernel function, with the mathematical expression: K(x,y)=exp(-||xy||2 / (2σ2)), where σ is the kernel width parameter, which needs to be adjusted according to the specific data; x and y represent two different data points; K(x,y) represents the kernel function; and exp(·) represents the natural exponential function. It should be noted that the machine learning model is constructed as an adaptive prediction model for the entire development cycle. The adaptive prediction model for the entire development cycle includes sub-models integrated into a unified framework: early-stage prediction model, mid-stage prediction model, and late-stage prediction model.

[0075] Step 4: For each development stage, train the corresponding sub-model and correct the model output using physical constraints.

[0076] For the early stages, a Support Vector Regression (SVR) model is used, aiming to minimize prediction error while satisfying sparsity requirements. The objective function of the SVR model can be expressed as: min 1 / 2 ||w||2 + C∑ξ i Where w is the weight vector, C is the regularization parameter, and ξ i These are slack variables. For the intermediate stage, due to the relatively large amount of data and strong nonlinear relationships, a random forest (RF) model is chosen for modeling. The RF model improves prediction accuracy by constructing multiple decision trees and voting or averaging the results. For the late stage, considering the data sparsity and long-term prediction requirements, a long short-term memory (LSTM) network model is used for modeling. The LSTM model effectively solves the gradient vanishing problem in traditional recurrent neural networks (RNNs) by introducing a gating mechanism. Its core formula includes the forgetting gate f. t =σ(W f ·[h t-1 ,x t ]+b f ), Input gate i t =σ(W i ·[h t-1 ,x t ]+b i Output gate o t =σ(W o ·[h t-1 ,x t ]+b o And the unit state update formula C t =f t ⊙C t-1 +i t ⊙tanh(W c ·[h t-1 ,x t ]+b c ), where σ is the activation function, and ⊙ represents element-wise multiplication; where h t-1 Indicates the hidden state of the previous time step; x t Indicates input data; b f b i b o b c These represent the bias terms for the forget gate, input gate, output gate, and candidate cell states, respectively; W f W i W o W c Represent the weight matrices for the forget gate, input gate, output gate, and candidate unit states, respectively; tanh represents the hyperbolic tangent activation function; C t-1 Indicates the cell state at the previous time step; C tThis represents the updated unit state. During training, physical constraints are introduced to correct the model output. For example, for reservoir pressure prediction, the mass conservation equation can be added as a constraint, i.e. ρ / t+ ·(ρv)=q, where ρ is density, v is velocity field, and q is source and sink term.

[0077] Step 5: Integrate the above sub-models into a unified framework and train them to obtain an adaptive prediction model for the entire development cycle.

[0078] A weighted fusion strategy is used to integrate the prediction results of each sub-model, and its mathematical expression is y=∑w i ×y i , where y is the final prediction result, y i For the prediction result of the i-th sub-model, w i These are the corresponding weighting coefficients. The weighting coefficients can be determined automatically based on cross-validation or Bayesian optimization algorithms. Furthermore, to improve the model's robustness, a model uncertainty assessment mechanism can be introduced, quantifying the model's confidence level by calculating the standard deviation of the prediction results. In this embodiment, experimental verification shows that the weighted fusion strategy significantly improves prediction accuracy compared to a single model, especially in the transition region between different development stages.

[0079] Step 6: Input the collected real-time monitoring data from the development phase into the adaptive prediction model for the entire development cycle, and output the production prediction results for the in-situ development phase of shale oil:

[0080] In practical applications, based on the real-time monitoring data input at the current development stage, the system invokes the corresponding sub-model for yield prediction. For example, assuming the current stage is mid-stage, the system will automatically identify and invoke the mid-stage random forest model for prediction. During the prediction process, real-time monitoring data is preprocessed and feature extracted before being input into the model. The model outputs prediction results and generates a visualization report. To facilitate rapid response from on-site technicians, the system also provides an interactive interface where users can view prediction results, adjust input parameters, and export report files.

[0081] Step 7: Correct the adaptive prediction model for the entire development cycle by analyzing the error between the predicted production results and the actual monitoring results during the in-situ development stage of shale oil.

[0082] By introducing an error feedback mechanism, the prediction results are dynamically corrected, ensuring the model's stability in long-term predictions. Specifically, the error between the prediction results and the actual monitoring data is first calculated, and then the error information is fed back into the model for parameter updates. For example, for an LSTM model, the weight matrix W and bias terms can be adjusted using the backpropagation algorithm, with the update formula being W = W - η· L / W, where η is the learning rate and L is the loss function. Furthermore, to prevent overfitting, a regularization term can be introduced to constrain the parameters. In this embodiment, experimental verification shows that the error feedback mechanism can effectively reduce long-term prediction errors, especially when data fluctuations are significant.

[0083] Step 8: Output the prediction results and generate a visualization report to provide decision support for optimizing the development plan.

[0084] The forecast results are presented in graphical form, including curves showing output changes over time, distribution maps of forecast errors at each stage, and ranking maps of the importance of key feature variables. For example, Figure 2 and Figure 3 The figures shown are comparisons between production forecasts and actual monitoring data for the early and mid-stages of development, respectively. They demonstrate a high degree of agreement between the forecasts and actual data. Furthermore, as... Figure 4 As shown, Figure 4 This diagram illustrates the integration framework of sub-models at different development stages. Figure 4 The reports clearly show the collaborative relationships between the sub-models and the data flow paths; these visualizations not only help technical personnel to intuitively understand the prediction results, but also provide important references for optimizing subsequent development plans.

[0085] Therefore, this embodiment provides a method for predicting shale oil production during the in-situ development stage. This method, through dynamic feature extraction and phased modeling, combined with physical constraints and an error feedback mechanism, achieves accurate prediction of shale oil production during the in-situ development stage. In practical applications, this method demonstrates strong adaptability and stability, meeting the needs of field technicians for rapid response. It also provides technical support for heating strategy optimization, reservoir stimulation design, and economic benefit assessment, possessing significant engineering practical value.

[0086] like Figure 6 As shown, this embodiment also provides a shale oil production prediction system for the in-situ development stage, including:

[0087] The data acquisition module is used to acquire multi-source dynamic monitoring data during the in-situ development of shale oil; the multi-source dynamic monitoring data includes temperature field, pressure field, fluid composition changes and reservoir physical parameters;

[0088] The segmentation module is used to divide the development stage into three stages—early, mid, and late—based on a preset development stage segmentation standard; and to determine the key feature variables corresponding to each stage.

[0089] The feature extraction module is used to extract and reduce the dimensions of multi-source dynamic monitoring data using time series analysis methods to obtain feature vectors.

[0090] The model training module is used to train a pre-built adaptive prediction model for the entire development cycle using feature vectors. The adaptive prediction model for the entire development cycle includes an early-stage prediction model, a mid-stage prediction model, and a late-stage prediction model integrated into a unified framework. The early-stage prediction model, the mid-stage prediction model, and the late-stage prediction model are trained using key feature variables corresponding to each stage in the feature vectors, respectively, to output the trained adaptive prediction model for the entire development cycle.

[0091] The production forecasting module is used to input the real-time monitoring data collected during the development phase into an adaptive forecasting model that covers the entire development cycle, and output the production forecast results for the in-situ development phase of shale oil.

[0092] The present invention also provides a shale oil production prediction device for the in-situ development stage, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the shale oil production prediction method for the in-situ development stage.

[0093] The present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the shale oil in-situ development stage production prediction method.

[0094] When the processor executes the computer program, it implements the steps of production prediction in the in-situ development stage of shale oil, such as: acquiring multi-source dynamic monitoring data during the in-situ development of shale oil; the multi-source dynamic monitoring data includes temperature field, pressure field, fluid composition changes and reservoir physical parameters;

[0095] Based on the pre-defined development phase division criteria, the development phase is divided into three stages: early, mid, and late; and the key characteristic variables corresponding to each stage are determined.

[0096] Time series analysis was used to extract features and reduce dimensionality of multi-source dynamic monitoring data to obtain feature vectors.

[0097] The pre-constructed adaptive prediction model for the entire development cycle is trained using feature vectors; wherein, the adaptive prediction model for the entire development cycle includes an early stage prediction model, a mid-stage prediction model, and a late stage prediction model integrated into a unified framework; the early stage prediction model, the mid-stage prediction model, and the late stage prediction model are trained using key feature variables corresponding to each stage in the feature vectors, respectively, to output the trained adaptive prediction model for the entire development cycle.

[0098] The collected real-time monitoring data during the development phase is input into an adaptive prediction model that covers the entire development cycle, and the output is the production prediction result for the in-situ development phase of shale oil.

[0099] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing preset functions, wherein the instruction segments describe the execution process of the computer program in the shale oil in-situ development stage production prediction equipment. For example, the computer program can be divided into a data acquisition module, a partitioning module, a feature extraction module, a model training module, and a production prediction module, with the following specific functions: The data acquisition module is used to acquire multi-source dynamic monitoring data during the shale oil in-situ development process; the multi-source dynamic monitoring data includes temperature field, pressure field, fluid composition changes, and reservoir physical parameters;

[0100] The segmentation module is used to divide the development stage into three stages—early, mid, and late—based on a preset development stage segmentation standard; and to determine the key feature variables corresponding to each stage.

[0101] The feature extraction module is used to extract and reduce the dimensions of multi-source dynamic monitoring data using time series analysis methods to obtain feature vectors.

[0102] The model training module is used to train a pre-built adaptive prediction model for the entire development cycle using feature vectors. The adaptive prediction model for the entire development cycle includes an early-stage prediction model, a mid-stage prediction model, and a late-stage prediction model integrated into a unified framework. The early-stage prediction model, the mid-stage prediction model, and the late-stage prediction model are trained using key feature variables corresponding to each stage in the feature vectors, respectively, to output the trained adaptive prediction model for the entire development cycle.

[0103] The production forecasting module is used to input the real-time monitoring data collected during the development phase into an adaptive forecasting model that covers the entire development cycle, and output the production forecast results for the in-situ development phase of shale oil.

[0104] The shale oil in-situ development stage production prediction equipment can be a desktop computer, laptop, handheld computer, or cloud server, etc. This equipment may include, but is not limited to, processors and memory. Those skilled in the art will understand that the above are examples of shale oil in-situ development stage production prediction equipment and do not constitute a limitation on such equipment. It may include more components than described above, or combine certain components, or use different components. For example, the shale oil in-situ development stage production prediction equipment may also include input / output devices, network access devices, buses, etc.

[0105] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or any conventional processor. This processor is the control center for shale oil in-situ development stage production prediction, connecting various parts of the shale oil in-situ development stage production prediction equipment via various interfaces and lines.

[0106] The memory can be used to store the computer program and / or modules. The processor realizes various functions of the shale oil in-situ development stage production prediction equipment by running or executing the computer program and / or modules stored in the memory and calling the data stored in the memory.

[0107] The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function (such as sound playback, image playback, etc.). The data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). Furthermore, the memory may include high-speed random access memory and non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital cards (SD cards), flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0108] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for predicting production during the in-situ development stage of shale oil.

[0109] If the modules / units integrated in the shale oil in-situ development stage production prediction system are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0110] Based on this understanding, the present invention can implement all or part of the processes in the above-mentioned shale oil in-situ development stage production prediction method, or it can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above-mentioned shale oil in-situ development stage production prediction method. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or a preset intermediate form, etc.

[0111] The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0112] It should be noted that the content contained in the computer-readable storage medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electrical carrier signals and telecommunication signals.

[0113] In summary, this embodiment provides a method for predicting shale oil production during the in-situ development stage. Compared with traditional prediction methods, this method has the following significant advantages:

[0114] This method includes steps such as collecting multi-source dynamic monitoring data, dividing development into stages, extracting dynamic features, training sub-models in stages, integrating adaptive prediction models, introducing an error feedback mechanism, and outputting prediction results. This method can effectively capture the thermo-mechanical-chemical coupling effect through dynamic feature extraction and staged modeling, improving prediction accuracy, effectively solving the modeling challenges under small sample conditions, and possessing high computational efficiency and wide applicability. It can provide technical support for the optimization of shale oil development and has significant engineering application value.

[0115] The above embodiments are merely one of the implementation methods for achieving the technical solution of the present invention. The scope of protection claimed by the present invention is not limited to this embodiment, but also includes any variations, substitutions and other implementation methods that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting shale oil production during in-situ development, characterized in that, include: Acquire multi-source dynamic monitoring data during the in-situ development of shale oil; the multi-source dynamic monitoring data includes temperature field, pressure field, fluid composition changes, and reservoir physical property parameters; Based on the pre-defined development phase division criteria, the development phase is divided into three stages: early, mid, and late; and the key characteristic variables corresponding to each stage are determined. Time series analysis was used to extract features and reduce dimensionality of multi-source dynamic monitoring data to obtain feature vectors. The pre-constructed adaptive prediction model for the entire development cycle is trained using feature vectors; wherein, the adaptive prediction model for the entire development cycle includes an early stage prediction model, a mid-stage prediction model, and a late stage prediction model integrated into a unified framework; the early stage prediction model, the mid-stage prediction model, and the late stage prediction model are trained using key feature variables corresponding to each stage in the feature vectors, respectively, to output the trained adaptive prediction model for the entire development cycle. The collected real-time monitoring data during the development phase is input into an adaptive prediction model that covers the entire development cycle, and the output is the production prediction result for the in-situ development phase of shale oil. The acquisition of multi-source dynamic monitoring data during the in-situ development of shale oil includes: During the in-situ development of shale oil, multi-source dynamic monitoring data is acquired, including: temperature field data is acquired through distributed optical fiber sensing technology, pressure field data is recorded through downhole pressure gauges, fluid component change data is acquired through gas chromatography analysis, and reservoir physical parameters are acquired through laboratory tests or field logging data. In the step of dividing the development stage into three stages—early, mid, and late—according to a preset development stage division standard, the development stage division standard is as follows: When the average reservoir temperature reaches the preset temperature threshold, it indicates that the development stage has entered the middle stage from the early stage. When the production growth rate is lower than the preset growth rate, it indicates that the development stage has moved from the middle stage to the late stage. The process of training a pre-built adaptive prediction model for the entire development cycle using feature vectors includes: An adaptive prediction model for the entire development cycle is constructed, which integrates early-stage, mid-stage, and late-stage prediction models into a unified framework. The early-stage prediction model is based on a support vector regression model; the mid-stage prediction model is based on a random forest model; and the late-stage prediction model is based on a long short-term memory network model. In the process of integrating the early-stage, mid-stage, and late-stage prediction models into a unified framework, a weighted fusion strategy is used to integrate the prediction results output by the early-stage, mid-stage, and late-stage prediction models. The specific formula is as follows: y=∑w i ·and i In the formula, y represents the final prediction result. i w represents the prediction result of the i-th sub-model. i These are the corresponding weighting coefficients, which are automatically adjusted based on cross-validation or Bayesian optimization algorithms. The process of inputting the collected real-time monitoring data from the development phase into an adaptive prediction model oriented towards the entire development cycle, and outputting the shale oil in-situ development phase production prediction results, also includes: By analyzing the error between the production prediction results and the actual monitoring results during the in-situ development stage of shale oil, the adaptive prediction model for the entire development cycle is corrected. A visualization report is generated based on the output production forecast results of the in-situ development stage of shale oil; wherein, the visualization report includes a production change curve over time, a prediction error distribution map for each stage, and an importance ranking map of key feature variables.

2. The method for predicting shale oil production during in-situ development according to claim 1, characterized in that, The method employs time series analysis to extract features and reduce the dimensionality of multi-source dynamic monitoring data to obtain feature vectors, including: A sliding window method is used to segment multi-source dynamic monitoring data into time series data; each window includes data for a preset time span. Principal component analysis was used to reduce the dimensionality of the data within the window, and feature vectors reflecting the main trends in data change were extracted.

3. The method for predicting shale oil production during in-situ development according to claim 1, characterized in that, The process involves inputting real-time monitoring data collected during the development phase into an adaptive prediction model that covers the entire development cycle, and outputting shale oil in-situ development production prediction results, including: Identify the development stage corresponding to the real-time monitoring data during the development phase; The system invokes the early-stage, mid-stage, or late-stage prediction models from the adaptive prediction model for the entire development cycle to predict real-time monitoring data during the development stage and outputs the production prediction results for the in-situ development stage of shale oil.

4. A shale oil in-situ development stage production prediction system, used to implement the shale oil in-situ development stage production prediction method according to any one of claims 1-3, characterized in that, include: The data acquisition module is used to acquire multi-source dynamic monitoring data during the in-situ development of shale oil. The multi-source dynamic monitoring data includes temperature field, pressure field, fluid composition changes, and reservoir physical parameters; The segmentation module is used to divide the development stage into three stages—early, mid, and late—based on a preset development stage segmentation standard; and to determine the key feature variables corresponding to each stage. The feature extraction module is used to extract and reduce the dimensions of multi-source dynamic monitoring data using time series analysis methods to obtain feature vectors. The model training module is used to train a pre-built adaptive prediction model for the entire development cycle using feature vectors. The adaptive prediction model for the entire development cycle includes an early-stage prediction model, a mid-stage prediction model, and a late-stage prediction model integrated into a unified framework. The early-stage prediction model, the mid-stage prediction model, and the late-stage prediction model are trained using key feature variables corresponding to each stage in the feature vectors, respectively, to output the trained adaptive prediction model for the entire development cycle. The production forecasting module is used to input the real-time monitoring data collected during the development phase into an adaptive forecasting model that covers the entire development cycle, and output the production forecast results for the in-situ development phase of shale oil.

5. A production prediction device for in-situ shale oil development, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the shale oil in-situ development stage production prediction method according to any one of claims 1-3.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it is used to implement the steps of the shale oil in-situ development stage production prediction method according to any one of claims 1-3.

Images