Wellbore effusion intelligent early warning method based on effusion round and AI conjoint analysis

Abstract

The invention discloses a wellbore effusion intelligent early warning method based on effusion round and AI conjoint analysis, which comprises the following steps: acquiring gas well production basic data, daily production dynamic data and real-time production data; according to daily production dynamic data, hydrops turns are divided, data processing is carried out according to the turns, and convolutional neural network modeling is carried out; on the basis of real-time production data, the preliminary model is adopted for hydrops early warning; and S2 is repeated for the latest finished round of production data, model updating is completed, and the next round of effusion early warning application is started according to the step S3. According to the method, the advantages of liquid accumulation-liquid drainage periodical characteristics and mass production real-time data in the gas well production process are fully utilized, and the model is autonomously and intelligently updated in each round, so that the model can be fully matched with gas well production dynamic changes, the accuracy and timeliness of gas well shaft liquid accumulation diagnosis are improved, and the working efficiency is improved. Therefore, technical support is provided for improving gas well yield and recovery efficiency.

Description

Technical Field

[0001] This invention belongs to the field of intelligent oil and gas extraction technology, and in particular relates to an intelligent early warning method for wellbore fluid accumulation based on the combined analysis of fluid accumulation cycles and AI. Background Technology

[0002] Currently, with the continuous development of information technology, the digital transformation of gas wells has achieved phased results. At present, a large amount of real-time wellhead data can be acquired, supporting a certain degree of digital management and intelligent applications. However, how to utilize real-time data for more intelligent and effective early warning of wellbore liquid accumulation is an urgent need for intelligent gas production.

[0003] Traditional pressure gradient methods, which rely primarily on testing, are too costly to be widely adopted. Methods based on real-time testing of wellhead gas and water production are not only expensive but also produce delayed early warnings, hindering further intelligent applications. Methods that primarily rely on wellhead parameters combined with multiphase flow calculations in the wellbore are overly dependent on high-pressure fluid properties, making them inconvenient to use.

[0004] Once a gas well enters the drainage and production phase, the production process typically unfolds in cycles of normal production, fluid accumulation, and fluid removal. Domestic and international methods for fluid accumulation early warning, whether based on mechanistic models or AI models, rarely fully utilize the cyclical nature of gas well fluid accumulation in conjunction with real-time production data for modeling. Furthermore, current fluid accumulation model algorithms rarely consider the edge-deployment and operational needs of intelligent gas well production, and cannot perform autonomous model updates. Therefore, the accuracy and timeliness of wellbore fluid accumulation early warnings often fail to meet the demands of field production. Summary of the Invention

[0005] The problem this invention aims to solve is to provide an intelligent early warning method for wellbore fluid accumulation based on the joint analysis of fluid accumulation cycles and AI. This method utilizes the production characteristics of fluid accumulation cycles to model the early warning method for fluid accumulation, which is of great significance for improving the accuracy and timeliness of wellbore fluid accumulation early warning.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: an intelligent early warning method for wellbore fluid accumulation based on the combined analysis of fluid accumulation cycles and AI, comprising the following steps: S1: Acquire basic gas well production data, daily production dynamic data, and real-time production data; S2: Based on the daily production dynamic data, divide the liquid accumulation cycle, and perform data processing and convolutional neural network modeling according to the cycle; S3: Based on the real-time production data, a preliminary model is used to provide early warning of liquid accumulation; S4: Repeat S2 for the latest completed round of production data to update the model, and then start the next round of liquid accumulation warning application according to step S3.

[0007] Furthermore, in S1, the basic production data includes well depth trajectory, casing size, tubing size, and tubing running depth; the daily production dynamic data includes daily water production, daily gas production, and production status notes; and the real-time production data includes oil pressure, casing pressure, back pressure, and wellhead temperature.

[0008] Furthermore, S2 includes the following steps: S21: Using AI algorithms, the keywords "liquid accumulation" and "liquid drainage" are automatically extracted from the daily production dynamic data, and the data is divided into rounds according to the time sequence, with real-time data at the minute level as the main component, forming a round time-series dataset. S22: Perform data cleaning, data normalization, data merging, and dataset classification by round; S23: Convolutional Neural Network Modeling.

[0009] Furthermore, in S22, the data cleaning includes the following steps: S221: Find the threshold records of each variable in the production environment that are similar to the data, compare each variable with its threshold, and regard the data that is outside the threshold range as abnormal data; S222: Handling abnormal data; S223: Each time a new batch of production data is added, the range of values ​​of the variables is compared with the historical range of changes recorded in the threshold library, and the original variable threshold relationship in the library is updated, thereby obtaining the range of values ​​of each variable under different environmental and production conditions.

[0010] Furthermore, in S222, the spatiotemporal distribution test method is adopted. First, the changes of other variables in the record where the abnormal variable is located are compared with the changes of the data before and after it. If the three are similar, the average of the corresponding variables before and after the abnormal value is taken. If the difference is large, the value of the one that is similar is taken.

[0011] Furthermore, in the data normalization process of S22, the min-max normalization method is used to linearly transform the original data for the data within the round range, so that the result value is mapped to the range [0 - 1]. The transformation function is shown in formula (1). (1) Where: max is the maximum value of the sample data, and min is the minimum value of the sample data.

[0012] Furthermore, in the data merging process in S22, the normalized round data are merged to form a complete dataset.

[0013] Furthermore, in the dataset classification in S22, the dataset is divided into a training set and a test set by random extraction in an 8:2 ratio.

[0014] Furthermore, S23 includes the following steps: S231: The temporal parameters are arranged into a matrix, and the convolution kernels in the convolutional layer slide on the input data to perform convolution operations to extract local features. Each convolution kernel will learn to recognize specific feature patterns. The mathematical expression of the convolution operation is shown in formula (2): (2) Where: x is the input data, w is the convolution kernel weight, b is the bias term, and y is the convolution output; S232: Pooling layers are connected after convolutional layers to reduce feature dimensionality and retain important information. After multiple convolutions and pooling, the feature map is flattened and input into the fully connected layer. Finally, the probability distribution of samples belonging to each category is output through the Softmax activation function, completing the initial model creation. S233: Use the test set data to perform preliminary model validation. If the model reaches the preset accuracy of 95%, the next step of inference application can be carried out. Otherwise, return to the modeling process, adjust the training parameters, and retrain until the accuracy requirement is met.

[0015] Furthermore, in S3, for gas wells with abundant training datasets, single-well modeling is performed. Based on an edge computing architecture, it is deployed at the production site to directly process data collected in real time from wellhead sensors, and to perform real-time monitoring and early warning of liquid accumulation in the single wellbore. If it is linked with the liquid drainage process equipment, intelligent early warning and liquid drainage integration can be achieved.

[0016] The advantages and positive effects of this invention are: This invention fully utilizes the periodic characteristics of liquid accumulation and drainage in the gas well production process and the advantages of massive real-time production data. The model is autonomously and intelligently updated in each round, so that the model can fully match the dynamic changes in gas well production, improve the accuracy and timeliness of gas well liquid accumulation diagnosis, and thus provide technical support for improving gas well production and recovery rate. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall process of an embodiment of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The embodiments of the present invention will be further described below with reference to the accompanying drawings: like Figure 1 As shown, a smart early warning method for wellbore fluid accumulation based on joint analysis of fluid accumulation cycles and AI includes the following steps: S1: Acquire basic production data, daily production dynamic data, and real-time production data for gas wells. Specifically, basic production data includes well depth trajectory, casing size, tubing size, and tubing running depth. Daily production dynamic data includes daily water production, daily gas production, and production status notes. Real-time production data includes oil pressure, casing pressure, back pressure, and wellhead temperature.

[0020] S2: Based on daily production dynamic data, divide the liquid accumulation into cycles, and perform data processing and convolutional neural network modeling according to the cycle. Specifically, this includes the following steps: S21: Using AI algorithms, the keywords "liquid accumulation" and "liquid drainage" are automatically extracted from the daily production status notes. The data is divided into rounds according to the time sequence, and real-time data at the minute level, including oil pressure, casing pressure, back pressure, and wellhead temperature, is used to form a round time series dataset.

[0021] S22: Data processing by round includes data cleaning, data normalization, data merging, and dataset classification.

[0022] Specifically, data cleaning includes the following steps: First, find the threshold records of each variable in a production environment similar to the current data. Then, compare each variable with its threshold. Variables that are outside the threshold range are considered abnormal data.

[0023] To handle outlier data, a spatiotemporal distribution test is used. First, the changes of other variables in the record containing the outlier variable are compared with the changes of the data before and after it. If the three are similar, the outlier value is taken as the average of the corresponding variables before and after it; if the difference is large, the value of the closest one is taken.

[0024] Each time a new batch of production data is added, the range of values ​​for the variables is compared with the historical range of changes already recorded in the threshold library, and the original variable threshold relationships in the library are updated, thereby obtaining the range of values ​​for each variable under different environmental and production conditions.

[0025] Data normalization processing includes the following: For data within the round range, the min-max normalization method, also known as deviation normalization, is used. This is a linear transformation of the original data, mapping the result values ​​to the range [0 - 1]. The transformation function is shown in formula (1). (1) Where max represents the maximum value of the sample data, and min represents the minimum value of the sample data. A drawback of this method is that when new data is added, max and min may change, requiring them to be redefined.

[0026] Data merging processing includes the following: The normalized round data are merged to form a complete dataset.

[0027] The dataset classification includes the following: The dataset was divided into training and test sets in an 8:2 ratio using random sampling.

[0028] S23: Convolutional neural network modeling includes the following steps: For the training set data, firstly, time-series parameters including oil pressure, casing pressure, back pressure, and wellhead temperature are composed into a matrix. Then, convolution operations are performed on the input data by sliding convolution kernels (filters) in the convolutional layer to extract local features. Each convolution kernel learns to recognize specific feature patterns, such as sudden changes or periodic fluctuations in parameters. The mathematical expression for the convolution operation is shown in formula (2): (2) Where: x is the input data, w is the convolution kernel weight, b is the bias term, and y is the convolution output.

[0029] Convolutional layers are followed by pooling layers to reduce feature dimensionality while retaining important information. Common pooling operations include max pooling and average pooling. After multiple convolutional and pooling layers, the feature map is flattened and fed into a fully connected layer. Finally, the Softmax activation function outputs the probability distribution of samples belonging to each class, completing the initial model creation.

[0030] Then, the model is validated using test set data. If the model achieves the preset accuracy of 95%, it can proceed to the next step of inference application. Otherwise, it is necessary to return to the modeling process, adjust the training parameters, and retrain to meet the accuracy requirements.

[0031] S3: Based on real-time data such as oil and casing back pressure and wellhead temperature, a preliminary model is used for early warning of fluid accumulation. Specifically, for gas wells with abundant training datasets, single-well modeling can be performed. Based on an edge computing architecture, it can be deployed at the production site to directly process real-time data on oil pressure, casing pressure, back pressure, and wellhead temperature collected from wellhead sensors, enabling real-time monitoring and early warning of fluid accumulation in the single-well borehole. If linked with fluid drainage equipment, intelligent early warning and fluid drainage integration can be achieved.

[0032] S4: After completing one round of drainage measures, the latest round of data is automatically generated. For this round of data, the S2 process is restarted to create the latest model and submit it to the system model library. A new round of S3 model inference application is then initiated.

[0033] If the selected example well is a new well and there is insufficient daily dynamic data and real-time data, it is recommended to use the existing model in the model library for the first round of inference application, based on the principle of the smallest difference in the basic data; then, based on the first round of liquid accumulation data, a preliminary model of the well is rebuilt.

[0034] The advantages and positive effects of this invention are: This invention fully utilizes the periodic characteristics of liquid accumulation and drainage in the gas well production process and the advantages of massive real-time production data. The model is autonomously and intelligently updated in each round, so that the model can fully match the dynamic changes in gas well production, improve the accuracy and timeliness of gas well liquid accumulation diagnosis, and thus provide technical support for improving gas well production and recovery rate.

[0035] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for intelligent early warning of wellbore fluid accumulation based on joint analysis of fluid accumulation cycles and AI, characterized in that: Includes the following steps, S1: Acquire basic gas well production data, daily production dynamic data, and real-time production data; S2: Based on the daily production dynamic data, divide the liquid accumulation cycle, and perform data processing and convolutional neural network modeling according to the cycle; S3: Based on the real-time production data, a preliminary model is used to provide early warning of liquid accumulation; S4: Repeat S2 for the latest completed round of production data to update the model, and then start the next round of liquid accumulation warning application according to step S3.

2. The intelligent early warning method for wellbore fluid accumulation based on joint analysis of fluid accumulation cycles and AI as described in claim 1, characterized in that: In S1, the basic production data includes well depth trajectory, casing size, tubing size, and tubing running depth; the daily production dynamic data includes daily water production, daily gas production, and production status notes; and the real-time production data includes oil pressure, casing pressure, back pressure, and wellhead temperature.

3. A method for intelligent early warning of wellbore fluid accumulation based on joint analysis of fluid accumulation cycles and AI, as described in claim 1 or 2, characterized in that: S2 includes the following steps: S21: Using AI algorithms, the keywords "liquid accumulation" and "liquid drainage" are automatically extracted from the daily production dynamic data, and the data is divided into rounds according to the time sequence, with real-time data at the minute level as the main component, forming a round time-series dataset. S22: Perform data cleaning, data normalization, data merging, and dataset classification by round; S23: Convolutional Neural Network Modeling.

4. The intelligent early warning method for wellbore fluid accumulation based on joint analysis of fluid accumulation cycles and AI as described in claim 3, characterized in that: In step S22, the data cleaning includes the following steps: S221: Find the threshold records of each variable in the production environment that are similar to the data, compare each variable with its threshold, and regard the data that is outside the threshold range as abnormal data; S222: Handling abnormal data; S223: Each time a new batch of production data is added, the range of values ​​of the variables is compared with the historical range of changes recorded in the threshold library, and the original variable threshold relationship in the library is updated, thereby obtaining the range of values ​​of each variable under different environmental and production conditions.

5. The intelligent early warning method for wellbore fluid accumulation based on joint analysis of fluid accumulation cycles and AI as described in claim 4, characterized in that: In S222, the spatiotemporal distribution test method is used. First, the changes of other variables in the record where the outlier variable is located are compared with the changes of the data before and after it. If the three are similar, the outlier value is the average of the corresponding variables before and after it. If the difference is large, the value of the one that is similar is taken.

6. The intelligent early warning method for wellbore fluid accumulation based on joint analysis of fluid accumulation cycles and AI as described in claim 3, characterized in that: In the data normalization process described in S22, the min-max normalization method is used to linearly transform the original data for data within the round range, so that the result value is mapped to the range [0 - 1]. The transformation function is shown in formula (1). （1） Where: max is the maximum value of the sample data, and min is the minimum value of the sample data.

7. The intelligent early warning method for wellbore fluid accumulation based on joint analysis of fluid accumulation cycles and AI as described in claim 3, characterized in that: In the data merging process described in S22, the normalized round data are merged to form a complete dataset.

8. The intelligent early warning method for wellbore fluid accumulation based on joint analysis of fluid accumulation cycles and AI as described in claim 3, characterized in that: In the dataset classification in S22, the dataset is divided into a training set and a test set by random extraction in an 8:2 ratio.

9. The intelligent early warning method for wellbore fluid accumulation based on joint analysis of fluid accumulation cycles and AI as described in claim 3, characterized in that: S23 includes the following steps: S231: The temporal parameters are arranged into a matrix, and the convolution kernels in the convolutional layer slide on the input data to perform convolution operations to extract local features. Each convolution kernel will learn to recognize specific feature patterns. The mathematical expression of the convolution operation is shown in formula (2): （2） Where: x is the input data, w is the convolution kernel weight, b is the bias term, and y is the convolution output; S232: Pooling layers are connected after convolutional layers to reduce feature dimensionality and retain important information. After multiple convolutions and pooling, the feature map is flattened and input into the fully connected layer. Finally, the probability distribution of samples belonging to each category is output through the Softmax activation function, completing the initial model creation. S233: Use the test set data to perform preliminary model validation. If the model reaches the preset accuracy of 95%, the next step of inference application can be carried out. Otherwise, return to the modeling process, adjust the training parameters, and retrain until the accuracy requirement is met.

10. A method for intelligent early warning of wellbore fluid accumulation based on joint analysis of fluid accumulation cycles and AI, as described in claim 1 or 2, characterized in that: In S3, for gas wells with abundant training datasets, single-well modeling is performed. Based on an edge computing architecture, it is deployed at the production site to directly process data collected in real time from wellhead sensors, and to perform real-time monitoring and early warning of fluid accumulation in the single wellbore. If it is linked with the fluid drainage process equipment, intelligent early warning and fluid drainage integration can be achieved.

Images