Oil reservoir monitoring index early warning method based on deep learning

Abstract

The invention provides an oil reservoir monitoring index early warning method based on deep learning. The oil reservoir monitoring index early warning method based on deep learning comprises the following steps: step 1, collecting historical data; step 2, establishing a convolutional neural network model, taking the bottomhole flowing pressure value as an input value, and taking the maximum value and the minimum value of the corresponding bottomhole temperature as output values; step 3, respectively training the two neural networks by using the label data corresponding to the maximum and minimum bottom hole temperatures to obtain two prediction models; and step 4, obtaining the maximum value and the minimum value of the predicted bottom hole temperature according to the prediction model, thereby quantifying the uncertainty interval of the maximum value and the minimum value. According to the oil reservoir monitoring index early warning method based on deep learning, a reliable data driving basis is provided for oil reservoir dynamic early warning; meanwhile, compared with a full-connection neural network, the deep convolutional neural network structure for predicting the upper and lower bounds of the bottom hole temperature has the advantages of good anti-noise effect and high accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of oilfield development, and in particular to an oil reservoir monitoring indicator early warning method based on deep learning. Background Art

[0002] After years of development, dynamic monitoring technology has gradually formed technologies in basic experiments, monitoring scheme design, monitoring data collection, data interpretation and comprehensive application, becoming another important means to understand and evaluate the dynamics and effects of reservoir development in addition to reservoir engineering and reservoir numerical simulation. The traditional reservoir engineering method based on the analytical solution of the seepage mechanics equation requires a significant simplification of the reservoir model and cannot fully consider the influence of reservoir heterogeneity and complex flow mechanisms. Reservoir numerical simulation is a conventional method for reservoir dynamic analysis and prediction. Its advantage is that the correlation between parameters based on the mechanism is clear, and after years of development, the method is mature and can be implemented based on commercial software. Its main limitation is that it relies on accurate geological models and seepage mechanisms. For blocks with unclear geological models or seepage mechanisms, the error is large, and only part of the monitoring data can be considered. The data-driven method uses machine learning and deep learning methods to directly fit the nonlinear relationship between data and perform corresponding analysis and prediction. It does not require a clear seepage mechanism or geological model, but it has high requirements for the quality of the collected data.

[0003] Data analysis based on machine learning mainly includes correlation analysis, denoising and dimensionality reduction, classification and regression, prediction, inversion and control, and has formed algorithms such as k-nearest neighbor method, support vector machine, logistic regression and neural network. How to combine machine learning algorithms with specific engineering practices and establish corresponding data-driven analysis technologies is the focus of industrial big data research. For reservoir monitoring data, establishing corresponding data analysis algorithms based on machine learning is conducive to fully exploring the characteristics of monitoring data and comprehensively considering the relationship between various dimensions of data, so that monitoring data can play a greater role in guiding the efficient development of oil fields, which is of great significance to the construction of intelligent oil and gas fields. The existing reservoir monitoring indicator early warning mainly relies on subjective experience and seepage mechanism, and lacks a data-driven early warning method.

[0004] In the Chinese patent application with application number: CN202011615557.8, a method and system for dynamic analysis of oilfield production based on big data are involved, including respectively obtaining basic data of a target well and a nearest neighboring well of the target well, the basic data including static data, dynamic data and associated data, performing data preprocessing on the acquired basic data, fusing the preprocessed data to obtain index data; establishing a distributed storage data warehouse, and sending the index data to the data warehouse for storage through a data interface; using the index data in the data warehouse as modeling parameters to construct an oilfield production diagnosis and early warning model; using data visualization technology to visualize the oilfield production diagnosis and early warning model, and issuing an early warning based on the oilfield production diagnosis and early warning model.

[0005] In the Chinese patent application with application number: CN202011608819.8, it involves an intelligent fault diagnosis method, system, storage medium, and equipment for offshore oil, gas and water wells, including: cleaning oil well data; data analysis; model building; model evaluation. The present invention takes offshore oil, gas and water wells as the research object, and proposes an intelligent fault diagnosis method for offshore oil, gas and water wells, which performs data cleaning, feature analysis, fault diagnosis, and fault prediction on development well data, which is conducive to improving collaborative efficiency and reducing communication costs. Through the research of this invention, a development well history data integration and intelligent early warning system is designed, which mainly uses statistical analysis technology to monitor the production process of oil and gas wells in real time, and to issue early warnings for abnormal trends in the production process; at the same time, it identifies possible faults and risks in the production practices of different oil and gas fields, so that production management personnel can take timely countermeasures, thereby achieving the purpose of improving production time and reducing maintenance costs.

[0006] In the Chinese patent application with application number: CN202211297047.X, a multi-parameter abnormal diagnosis and early warning method for oil wells based on machine learning is involved, which is carried out in the following steps: pre-processing the oil well production data, and taking threshold-based abnormal data processing for abnormal data; then analyzing the production parameters, after identifying the abnormalities of the long-term trend of a single parameter in the oil well, establishing a fault classification model to identify multi-parameter abnormalities for the oil well; formulating fault identification rules for the oil well and the identification rules for the fault type to give recommended disposal methods; forming a complete oil well fault case library through deep identification and manual inspection; and giving early warnings for oil well abnormalities that may occur in the future based on the long-term operating status index. The invention can collect oil well data in real time, complete accurate diagnosis of possible fault types, and make effective deep early warnings, thereby effectively and long-term improving the production and operation efficiency of oil wells.

[0007] In the Chinese patent application with application number: CN202111029072.5, a method for predicting a time-series dynamometer diagram of a pumping well based on deep learning is involved, and the method includes: (1) constructing a training data set; (2) normalization and interpolation processing; (3) the first stage: determining the working trend of the dynamometer diagram; (4) the second stage: predicting the dynamometer diagram; (5) the third stage: diagnosis and early warning; The invention constructs a training data set by slicing, normalizing and interpolating the historical dynamometer diagram, and applies a multi-channel convolutional neural network CNN to perform time-series trend working condition determination - wax deposition, insufficient fluid supply, gas lock and pump valve leakage, etc., and applies the LSTM sequence-to-sequence model to capture the changing trend characteristics and changing speed characteristics of the dynamometer diagram in the current and past period of time, and makes quantitative predictions of the dynamometer diagram in the future period of time, so that the working condition warning information is more intuitive and more interpretable.

[0008] The above existing technologies are significantly different from the present invention and fail to solve the technical problem we want to solve. For this reason, we have invented a new reservoir monitoring indicator early warning method based on deep learning. Summary of the invention

[0009] The purpose of the present invention is to provide a deep learning-based reservoir monitoring indicator early warning method with good noise resistance and high accuracy.

[0010] The purpose of the present invention can be achieved through the following technical measures: an oil reservoir monitoring indicator early warning method based on deep learning, the oil reservoir monitoring indicator early warning method based on deep learning includes:

[0011] Step 1, collect historical data;

[0012] Step 2, establish a convolutional neural network model, with the bottom hole flow pressure value as the input value and the corresponding maximum and minimum bottom hole temperature values ​​as the output value;

[0013] Step 3, using the label data corresponding to the maximum and minimum bottom hole temperature, respectively train two neural networks to obtain two prediction models;

[0014] Step 4: According to the prediction model, the maximum and minimum values ​​of the predicted bottom hole temperature are obtained, thereby quantifying its uncertainty range.

[0015] The purpose of the present invention can also be achieved by the following technical measures:

[0016] In step 1, daily bottom hole flowing pressure values ​​of the oil wells and maximum and minimum bottom hole temperatures corresponding to the bottom hole flowing pressure values ​​are collected.

[0017] In step 1, the daily bottom hole flowing pressure values ​​of the oil wells and all possible bottom hole temperature values ​​corresponding to the bottom hole flowing pressure values ​​are collected, and three sets of data are obtained by taking the maximum values, namely the maximum and minimum values ​​of the bottom hole flowing pressure and the bottom hole temperature corresponding to the bottom hole flowing pressure, as label data for neural network training.

[0018] In step 2, the convolutional neural network contains multiple modules and is finally output to the required dimension by a fully connected layer after dimension transformation.

[0019] In step 2, each module consists of a sequentially arranged convolutional layer, a normalization layer, a ReLU activation function layer, and a maximum pooling layer.

[0020] In step 2, the role of the convolution layer is to extract data features through convolution operations; the role of the normalization layer is to normalize the network parameters to improve training stability and convergence; the role of the maximum pooling layer is to reduce the dimension to improve training efficiency; the matrix after the maximum pooling layer is deformed into one dimension and output to the dimension of bottom hole temperature through the linear layer.

[0021] In step 3, the Adam optimizer is used for training, and the label data corresponding to the maximum and minimum bottom hole temperature are used to train two neural networks to obtain two prediction models.

[0022] In step 3, the prediction model obtained is:

[0023]

[0024]

[0025]

[0026] Among them, p wf =p wf (t) is the bottom hole flowing pressure value, is the approximate maximum bottom hole temperature, is an approximation of the minimum bottom hole temperature,

[0027] f is the bottom hole pressure value p wf Approximate maximum bottom hole temperature The neural network operator of the mapping,

[0028] g is the bottom hole pressure value p wf Approximate minimum bottom hole temperature Mapped neural network operators;

[0029] The bottom hole temperature variation range corresponding to the bottom hole flowing pressure value.

[0030] In step 4, during prediction, the measured values ​​of bottom hole flow pressure are input into two prediction models to obtain the predicted maximum and minimum values ​​of bottom hole temperature, thereby quantifying its uncertainty interval.

[0031] In step 4, if the measured bottom hole temperature value continues to be outside the uncertainty interval, an early warning is issued.

[0032] The deep learning-based reservoir monitoring index early warning method of the present invention collects daily bottom hole flowing pressure values ​​of oil wells and the maximum and minimum bottom hole temperature values ​​corresponding to the bottom hole flowing pressure values; then establishes a convolutional neural network model, specifically using the bottom hole flowing pressure value as an input value and the corresponding maximum and minimum bottom hole temperature values ​​as output values, and uses the Adam optimizer for training to obtain a neural network model; finally, the measured bottom hole flowing pressure value is input into the neural network model to obtain the bottom hole temperature value, and an early warning is issued if the bottom hole temperature value continues to be outside the bottom hole temperature variation range.

[0033] The present invention adopts a data-driven algorithm, does not rely on subjective experience and seepage mechanism, and can automatically learn based on historical data to obtain the uncertainty interval of bottom hole temperature corresponding to different bottom hole flow pressures, providing a reliable data-driven basis for dynamic early warning of oil reservoirs; at the same time, the deep convolutional neural network structure used to predict the upper and lower limits of bottom hole temperature in the present invention has the advantages of good noise resistance and high accuracy compared with the fully connected neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of maximum value fitting of historical data of bottom hole temperature based on convolutional neural network in a specific embodiment of the present invention;

[0035] Figure 2 It is a schematic diagram of minimum value fitting of historical data of bottom hole temperature based on convolutional neural network in a specific embodiment of the present invention;

[0036] Figure 3 It is a schematic diagram of predicting the upper limit of the uncertainty interval of the bottom hole temperature in a specific embodiment of the present invention;

[0037] Figure 4 It is a schematic diagram of predicting the lower limit of the uncertainty interval of the bottom hole temperature in a specific embodiment of the present invention;

[0038] Figure 5 A schematic diagram of the structure of a neural network model in a specific embodiment of the present invention;

[0039] Figure 6 The present invention is a flowchart of a specific embodiment of the deep learning-based reservoir monitoring indicator early warning method of the present invention. DETAILED DESCRIPTION

[0040] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0041] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations and / or combinations thereof.

[0042] The reservoir monitoring indicator early warning method based on deep learning of the present invention includes:

[0043] Step 1, collect historical data;

[0044] Step 2, establish a convolutional neural network model, with the bottom hole flow pressure value as the input value and the corresponding maximum and minimum bottom hole temperature values ​​as the output value;

[0045] Step 3, using the label data corresponding to the maximum and minimum bottom hole temperature, respectively train two neural networks to obtain two prediction models;

[0046] Step 4: According to the prediction model, the maximum and minimum values ​​of the predicted bottom hole temperature are obtained, thereby quantifying its uncertainty range.

[0047] The following are specific embodiments of the present invention.

[0048] In a specific embodiment 1 of the present invention, Figure 6 As shown, Figure 6 The flowchart of the oil reservoir monitoring indicator early warning method based on deep learning of the present invention. The oil reservoir monitoring indicator early warning method based on deep learning includes:

[0049] S101. Collect historical data

[0050] Collect daily bottom hole flowing pressure values ​​of oil wells and all possible bottom hole temperature values ​​corresponding to the bottom hole flowing pressure values, and obtain three sets of data by taking the maximum values, namely the maximum and minimum values ​​of the bottom hole flowing pressure and the bottom hole temperature corresponding to the bottom hole flowing pressure, as label data for neural network training;

[0051] S102. Establish a convolutional neural network model, with the bottom hole flow pressure value as the input value, and the corresponding maximum and minimum values ​​of the bottom hole temperature as the output value. The convolutional neural network contains multiple modules, and is ultimately output to the required dimension by the fully connected layer after dimensional transformation. Each module consists of a convolutional layer, a normalization layer, a ReLU activation function layer, and a maximum pooling layer arranged in sequence. The function of the convolutional layer is to extract data features through convolution operations; the function of the normalization layer is to normalize the network parameters to improve training stability and convergence; the function of the maximum pooling layer is to reduce the dimension to improve training efficiency. The matrix after the maximum pooling layer is deformed into one dimension, and is output as the dimension of the bottom hole temperature through the linear layer.

[0052] S103, using the Adam optimizer for training, using the label data corresponding to the maximum and minimum bottom hole temperature to train two neural networks respectively to obtain two prediction models:

[0053]

[0054]

[0055]

[0056] Among them, p wf =p wf (t) is the bottom hole flowing pressure value, is the approximate maximum bottom hole temperature, is an approximation of the minimum bottom hole temperature,

[0057] f is the bottom hole pressure value p wf Approximate maximum bottom hole temperature The neural network operator of the mapping,

[0058] g is the bottom hole pressure value p wf Approximate minimum bottom hole temperature Mapped neural network operators;

[0059] The bottom hole temperature variation range corresponding to the bottom hole flowing pressure value;

[0060] S104. During prediction, the measured value of the bottom hole flow pressure is input into the two neural network models to obtain the predicted maximum and minimum values ​​of the bottom hole temperature, thereby quantifying its uncertainty interval. If the measured bottom hole temperature value continues to be outside the uncertainty interval, an early warning is issued.

[0061] In a specific embodiment 2 of the present invention, the reservoir monitoring indicator early warning method based on deep learning includes the following steps:

[0062] S1. Collect historical data

[0063] Collect daily bottom hole flowing pressure values ​​of oil wells and maximum and minimum bottom hole temperatures corresponding to the bottom hole flowing pressure values;

[0064] S2. Establish a convolutional neural network model, with the bottom hole flow pressure as the input value and the corresponding maximum and minimum bottom hole temperature as the output value

[0065] S3, using the Adam optimizer for training, obtains two neural network prediction models (such as Figure 4 shown):

[0066]

[0067]

[0068]

[0069] Among them, p wf =p wf (t) is the bottom hole flowing pressure value, is an approximate value of the maximum bottom hole temperature, such as Figure 1 As shown; is an approximation of the minimum bottom hole temperature, such as Figure 2 As shown,

[0070] f is the bottom hole pressure value p wf Approximate maximum bottom hole temperature The neural network operator of the mapping,

[0071] g is the bottom hole pressure value p wf Approximate minimum bottom hole temperature Mapped neural network operators;

[0072] The bottom hole temperature variation range corresponding to the bottom hole flowing pressure value is as follows: Figure 3 As shown;

[0073] S4, input the measured value of the bottom hole flow pressure into the two neural network models to obtain the maximum and minimum values ​​of the bottom hole temperature value. If the bottom hole temperature value continues to be outside the bottom hole temperature uncertainty interval (i.e., the upper and lower limits described in the figure), an early warning is issued, such as Figure 4 shown.

[0074] In a specific embodiment 3 of the present invention, as Figure 5As shown in the figure, the convolutional neural network contains multiple modules. Each module consists of a sequentially arranged convolution layer, a normalization layer, a ReLU activation function layer, and a maximum pooling layer. The function of the convolution layer is to extract data features through convolution operations; the function of the normalization layer is to normalize the network parameters to improve training stability and convergence; the function of the maximum pooling layer is to reduce the dimension to improve training efficiency. The matrix after the maximum pooling layer is deformed into one dimension, and the output is the dimension of the bottom hole temperature through the linear layer. The input of the neural network is the bottom hole flow pressure time series data, and the output is the maximum or minimum bottom hole temperature time series data. The training method is supervised training based on labeled data, using the Adam optimizer and the mean square error function.

[0075] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions recorded in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0076] Except for the technical features described in the specification, all other technical features are known technologies to those skilled in the art.

Claims

1. The reservoir monitoring indicator early warning method based on deep learning is characterized by: The deep learning-based reservoir monitoring indicator early warning method includes: Step 1, collect historical data; Step 2, establish a convolutional neural network model, with the bottom hole flow pressure value as the input value and the corresponding maximum and minimum bottom hole temperature values ​​as the output value; Step 3, using the label data corresponding to the maximum and minimum bottom hole temperature, respectively train two neural networks to obtain two prediction models; Step 4: According to the prediction model, the maximum and minimum values ​​of the predicted bottom hole temperature are obtained, thereby quantifying its uncertainty range.

2. The reservoir monitoring indicator early warning method based on deep learning according to claim 1 is characterized in that: In step 1, daily bottom hole flowing pressure values ​​of the oil wells and maximum and minimum bottom hole temperatures corresponding to the bottom hole flowing pressure values ​​are collected.

3. The reservoir monitoring indicator early warning method based on deep learning according to claim 2 is characterized in that: In step 1, the daily bottom hole flowing pressure values ​​of the oil wells and all possible bottom hole temperature values ​​corresponding to the bottom hole flowing pressure values ​​are collected, and three sets of data are obtained by taking the maximum values, namely the maximum and minimum values ​​of the bottom hole flowing pressure and the bottom hole temperature corresponding to the bottom hole flowing pressure, as label data for neural network training.

4. The reservoir monitoring indicator early warning method based on deep learning according to claim 1 is characterized in that: In step 2, the convolutional neural network contains multiple modules and is finally output to the required dimension by a fully connected layer after dimension transformation.

5. The reservoir monitoring indicator early warning method based on deep learning according to claim 4 is characterized in that: In step 2, each module consists of a sequentially arranged convolutional layer, a normalization layer, a ReLU activation function layer, and a maximum pooling layer.

6. The reservoir monitoring indicator early warning method based on deep learning according to claim 5 is characterized in that: In step 2, the role of the convolution layer is to extract data features through convolution operations; the role of the normalization layer is to normalize the network parameters to improve training stability and convergence; the role of the maximum pooling layer is to reduce the dimension to improve training efficiency; the matrix after the maximum pooling layer is deformed into one dimension and output to the dimension of bottom hole temperature through the linear layer.

7. The reservoir monitoring indicator early warning method based on deep learning according to claim 1 is characterized in that: In step 3, the Adam optimizer is used for training, and the label data corresponding to the maximum and minimum bottom hole temperature are used to train two neural networks to obtain two prediction models.

8. The reservoir monitoring indicator early warning method based on deep learning according to claim 7 is characterized in that: In step 3, the prediction model obtained is: Among them, p wf =p wf (t) is the bottom hole flowing pressure value, is the approximate maximum bottom hole temperature, is an approximation of the minimum bottom hole temperature, f is the bottom hole pressure value p wf Approximate maximum bottom hole temperature The neural network operator of the mapping, g is the bottom hole pressure value p wf Approximate minimum bottom hole temperature Mapped neural network operators; The bottom hole temperature variation range corresponding to the bottom hole flowing pressure value.

9. The reservoir monitoring indicator early warning method based on deep learning according to claim 1 is characterized in that: In step 4, during prediction, the measured values ​​of bottom hole flow pressure are input into two prediction models to obtain the predicted maximum and minimum values ​​of bottom hole temperature, thereby quantifying its uncertainty interval.

10. The reservoir monitoring indicator early warning method based on deep learning according to claim 9 is characterized in that: In step 4, if the measured bottom hole temperature value continues to be outside the uncertainty interval, an early warning is issued.

Images