Intelligent diagnosis method for oil well working condition combining number and shape

Abstract

The application provides a number-shape combined oil well working condition intelligent diagnosis method, comprising the following steps: step 1, establishing an oil well working condition sample library of large-scale indicator diagrams and multi-dimensional time sequence data; step 2, proposing and establishing a CNN+DBN oil well working condition diagnosis neural network according to the characteristics of the oil well working condition diagnosis problem, completing the learning of the working condition sample library, and further improving the algorithm performance through repeated optimization; step 3, establishing an oil well working condition intelligent monitoring system, realizing the real-time connection-analysis-pushing of the intelligent monitoring neural network and the oil well production database, and designing an oil well working condition intelligent monitoring client front-end webpage system. The number-shape combined oil well working condition intelligent diagnosis method combines the intuitiveness, timeliness and effectiveness of graphics organically to determine the number-shape combined intelligent diagnosis scheme of "indicator diagram image + other time sequence data + CNN + DBN", and improves the oil well production dynamic intelligent diagnosis coincidence rate.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis and automated data processing technology, and in particular to an intelligent diagnostic method for oil well conditions that combines numerical and graphical methods. Background Technology

[0002] In the petroleum industry, deep convolutional neural network (DNN) dynamometer card recognition technology has remained in the experimental stage due to a lack of data, without significant breakthroughs. Only with the continuous advancement of oilfield modernization and industrialization has a solid foundation been laid for the research and development of intelligent dynamometer card recognition and control technologies using neural networks. However, traditional deep convolutional neural network (DNN) operating condition recognition methods have certain drawbacks: 1. While they learn quickly from images, their accuracy in distinguishing similar operating conditions is insufficient; 2. Some operating conditions occur through slow evolution, which cannot be diagnosed using a single-moment dynamometer card, indicating that the neural network does not adequately consider the evolutionary characteristics of time-series data.

[0003] In his 2015 master's thesis, "Intelligent Recognition of Oil Well Dynamograms Based on Deep Learning," Li Penghui constructed oil well dynamogram recognition models based on deep learning structures, using Convolutional Neural Networks (CNN) and Deep Belief Networks (DBN) respectively. The experimental results of these two deep learning-based oil well dynamogram recognition methods were compared and analyzed. However, this thesis only used dynamogram data for oil well condition diagnosis, without comprehensively considering other time-series data, making it difficult to distinguish between different operating conditions with similar dynamograms. Furthermore, the thesis used CNN and DBN separately for modeling and comparison, without coupling the two methods.

[0004] The article "Dysfunction Diagram Diagnosis Technology Based on CNN Convolutional Neural Network," published in the *Journal of Xi'an Petroleum University (Natural Science Edition)* (Issue 5, 2018), describes a dysfunction diagram diagnosis model based on digital descriptions of oil well parameters and dysfunction diagrams. It combines this with Convolutional Neural Network (CNN) technology to develop a corresponding calculation program, achieving intelligent diagnosis of pumping unit well conditions. However, this paper only utilizes CNN technology and the single data from the digital dysfunction diagram for diagnosis, without processing the time-series data beyond the dysfunction diagram.

[0005] Chinese patent application CN201810711850.0 discloses an intelligent monitoring method for rod pump operating conditions based on a dynamometer. This method intelligently monitors the operating conditions of rod pumps using a dynamometer, addressing the limitations of existing technologies in quickly identifying and diagnosing pumping wells experiencing gradual failures, and the limited room for improvement in diagnostic accuracy due to errors in comparing sample data with actual data. It enables the rapid identification and early warning of pumping wells with gradual failures, thereby improving the accuracy of rod pump operating condition monitoring and the intelligence and effectiveness of neural networks. However, this patent only uses CNN technology, considering the trend of dynamometer diagram changes over time. It struggles to efficiently analyze and process other time-series data beyond the dynamometer diagram image, and still cannot solve the problem of operating condition diagnosis under similar dynamometer diagram conditions.

[0006] The existing technologies described above are significantly different from the present invention and have failed to solve the technical problem we want to address. Therefore, we have invented a new intelligent diagnostic method for oil well conditions that combines numerical and graphical methods. Summary of the Invention

[0007] The purpose of this invention is to provide a digital-graphical intelligent diagnostic method for oil well conditions that improves the accuracy of intelligent diagnostics of oil well production dynamics by organically integrating the intuitiveness, timeliness, and effectiveness of graphics through problem analysis and technical research to determine a digital-graphical combination intelligent diagnostic scheme of "dynamometer image + other time-series data + CNN + DBN".

[0008] The objective of this invention can be achieved through the following technical measures: a digital-graphical combined intelligent diagnosis method for oil well operating conditions, comprising:

[0009] Step 1: Establish a large-scale dynamometer diagram and multi-dimensional time-series data sample library for oil well operating conditions;

[0010] Step 2: In view of the characteristics of oil well condition diagnosis problem, a neural network for oil well condition diagnosis coupled with convolutional neural network and deep belief network (CNN+DBN) is proposed and established to complete the learning of the condition sample database, and the algorithm performance is further improved through repeated optimization.

[0011] Step 3: Establish an intelligent monitoring system for oil well operating conditions, realize real-time connection, analysis, and push between the intelligent monitoring neural network and the oil well production database, and design a client-side web page system for intelligent monitoring of oil well operating conditions.

[0012] The objective of this invention can also be achieved through the following technical measures:

[0013] In step 1, a large-scale oil well condition sample library is established by collecting, organizing, cleaning, extracting features, and sorting production monitoring data of tens of millions of oil wells, and classifying and classifying the data. This library consists of dynamometer maps and multi-dimensional time-series data.

[0014] Step 1 includes:

[0015] 1) Collect oil well production monitoring data from oil production plants, and carry out data sorting, screening, and cleaning to improve data quality;

[0016] 2) Combining big data on oil well production monitoring, oil well idle reports, and field experience, the oil well production conditions are classified and categorized from a technical perspective into reservoir problems, tubing problems, rod problems, pump problems, surface problems, and data problems;

[0017] 3) Based on the results of the technical classification of oil well operating conditions and the actual needs of oilfield production management, oil well production conditions are divided into three categories from the perspective of production operation: healthy production, sub-healthy production, and abnormal production. Differentiated management measures are specified for different categories.

[0018] 4) To address the issue that relying solely on dynamometer diagrams is insufficient for accurate diagnosis of certain operating conditions, the system analyzes data indicators in the Production Control System (PCS) database and selects multi-dimensional time-series data for operating condition diagnosis.

[0019] 5) Conduct data divergence and correlation analysis on more than 60 types of multidimensional time series data, and further select data indicators with strong representativeness;

[0020] 6) For the selected multidimensional time series data, determine the time window for sample use: transient, short-term, medium-term, and long-term. Use time series feature extraction algorithm to extract its intrinsic features as operating condition analysis parameters.

[0021] Step 2 includes:

[0022] Step 21: Based on the characteristics of oil well condition diagnosis problems, conduct an in-depth analysis of the applicability of Convolutional Neural Network (CNN) and Deep Belief Network (DBN).

[0023] Step 22: Design the CNN system architecture for intelligent diagnosis of oil well conditions and perform relevant deep convolutional neural network core calculations;

[0024] Step 23: Design the DBN system architecture for intelligent diagnosis of oil well conditions and perform relevant deep convolutional neural network core calculations;

[0025] Step 24: Design the CNN+DBN system architecture for intelligent diagnosis of oil well conditions;

[0026] Step 25: Select representative datasets to analyze the effectiveness of the CNN+DBN system architecture for intelligent diagnosis of oil well conditions;

[0027] Step 26: Set the default learning parameters and use the established oil well condition sample library to train and test the CNN+DBN neural network for intelligent diagnosis of oil well conditions;

[0028] Step 27: Use techniques such as architecture optimization, iterative training, and Dropout to further optimize the neural network, thereby improving its training efficiency and generalization ability.

[0029] In step 21, through problem analysis and investigation, the intuitiveness and timeliness of graphics and the accuracy and effectiveness of data are organically integrated to determine the CNN+DBN intelligent diagnostic solution that combines numbers and shapes.

[0030] In step 22, the system architecture has a total of 14 layers, including 5 convolutional layers, 3 pooling layers, 3 hidden layers, 1 input layer, 1 rasterization layer, 1 output layer, and 2,254,592 neurons;

[0031] Convolution kernel operation formula:

[0032]

[0033] In the formula O j — Output data of the j-th layer neuron, φ — ReLU activation function, b j —Bias of neurons in layer j, —Calculate the weighted sum of each neuron from i to j, where The summation operation is performed sequentially from the neurons in layer i, X. i —Input variables of layer i neurons, V—Weight coefficients, Θ j —Filter mapping to layer j neurons;

[0034] Pooling core operation formula:

[0035] S j =downsample(O j )

[0036] In the formula S j — Output results of pooling processing of layer j neurons, O j —Output data of neurons in layer j;

[0037] Rasterization core operation formula:

[0038]

[0039] In the formula X jmn —Input variables: two-dimensional spatial coordinates of neurons in layer j; T—projected coordinate mapping; O k — Output rasterized pixel coordinates.

[0040] In step 23, the joint probability distribution of the energy function for the input variable x and the latent variable h is defined as follows:

[0041]

[0042] In the formula, p represents the model probability distribution, e represents the natural logarithm, and Z = ∑ x,h e-energy(x,h) is called the partition function, where energy(x,h) is the energy function, and ∑ x,h e -θ —The negative logarithmic summation of input variable x and latent variable h is used to perform forward stacked RBM learning from low to high level and backward fine-tuning learning from high to low level.

[0043] In step 24, the CNN branch is used to extract the power map information, and the DBN branch is used to input numerical data. After the rasterization layer is used for numerical processing, the processed data is concatenated into vectors in the core concatenation layer and input into the fully connected neural network to give the final output, thereby realizing the coupling of CNN and DBN and performing the core calculation of the deep hybrid neural network.

[0044] Step 3 includes:

[0045] Step 31: Design the overall architecture of the intelligent monitoring software system for oil well operating conditions;

[0046] Step 32: Perform intelligent monitoring and calculation of oil well operating conditions to achieve remote data acquisition, analysis, and storage;

[0047] Step 33: Design and compile the front-end web interface for intelligent monitoring of oil well operating conditions.

[0048] The intelligent diagnosis method for oil well operating conditions combining numerical and graphical methods in this invention is based on on-site production monitoring data. Data preprocessing is performed, data indicators in the PCS database are analyzed, and multidimensional time-series data for operating condition diagnosis is selected. Divergence and correlation analyses are conducted on the data to optimize the selection. A time-series feature extraction algorithm is used to extract intrinsic features as operating condition analysis parameters. A CNN+DBN system architecture for intelligent diagnosis of oil well operating conditions is designed, and relevant deep hybrid neural network core calculations are performed. Default learning parameters are set, and the CNN+DBN neural network for intelligent diagnosis of oil well operating conditions is trained and tested using an established oil well operating condition sample library. The neural network is further optimized using techniques such as architecture optimization, iterative training, and Dropout to improve the diagnostic accuracy and generalization ability of the neural network, achieving a single-well operating condition diagnosis rate of 90%. This improves the oil well production rate and efficiency, increases oil well output, and improves the operating conditions of the oil well. Attached Figure Description

[0049] Figure 1 A flowchart of a specific embodiment of the intelligent diagnostic method for oil well conditions combining numerical and graphical methods of the present invention;

[0050] Figure 2This is a schematic diagram of the CNN+DBN system architecture for intelligent diagnosis of oil well conditions in a specific embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram illustrating the accuracy and verification accuracy of the "CNN+DBN" neural network for intelligent diagnosis of oil well conditions in a specific embodiment of the present invention after learning the training set of dynamometer diagram + multidimensional time series data;

[0052] Figure 4 This is a diagram illustrating an optimization scheme for overfitting a neural network to a large task using Dropout technology in a specific embodiment of the present invention.

[0053] Figure 5 This is a flowchart of intelligent monitoring and calculation of oil well operating conditions in a specific embodiment of the present invention. Detailed Implementation

[0054] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0055] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, and / or combinations thereof.

[0056] The intelligent diagnostic method for oil well operating conditions combining numerical and graphical methods of the present invention includes the following steps:

[0057] Step 1: By collecting, organizing, cleaning, extracting features, and sorting production monitoring data from tens of millions of oil wells, a large-scale oil well condition sample library of "dynamometer charts + multi-dimensional time series data" is established in a hierarchical and classified manner.

[0058] The establishment of the large-scale oil well operating condition sample library consisting of "dynamometer diagrams + multi-dimensional time-series data" includes:

[0059] 1) Collect oil well production monitoring data from oil production plants, and carry out data sorting, screening, and cleaning to improve data quality;

[0060] 2) Combining big data on oil well production monitoring, oil well idle reports, and field experience, the oil well production conditions are classified and categorized from a technical perspective into issues such as reservoir problems, tubing problems, rod problems, pump problems, surface problems, and data problems;

[0061] 3) Based on the results of the technical classification of oil well operating conditions and the actual needs of oilfield production management, oil well production conditions are divided into three categories from the perspective of production operation: healthy production, sub-healthy production, and abnormal production. Differentiated management measures are specified for different categories.

[0062] 4) To address the issue that relying solely on dynamometer diagrams is insufficient for accurate diagnosis of certain operating conditions, the system analyzes data indicators in the Production Control System (PCS) database and selects multi-dimensional time-series data for operating condition diagnosis.

[0063] 5) Conduct data divergence and correlation analysis on more than 60 types of multidimensional time series data, and further select data indicators with strong representativeness;

[0064] 6) For the selected multidimensional time series data, determine the time window for sample use: transient, short-term, medium-term, and long-term. Use time series feature extraction algorithm to extract its intrinsic features as operating condition analysis parameters.

[0065] Step 2: Based on the characteristics of oil well operating condition diagnosis problems, we conduct an in-depth analysis of the applicability of deep learning algorithms such as CNN and DBN, propose and establish a "CNN+DBN" oil well operating condition diagnosis neural network, complete the learning of the operating condition sample database, and further improve the algorithm performance through repeated optimization.

[0066] The aforementioned "CNN+DBN" oil well condition diagnosis neural network includes:

[0067] 1) Based on the characteristics of oil well condition diagnosis, this paper analyzes the applicability of Convolutional Neural Networks (CNN) and Deep Belief Networks (DBN). Regarding data input format, DBN requires significantly less storage space than CNN: CNN uses images as input, with a single dynamometer card containing up to 40,000 images, while DBN uses arrays, with a single dynamometer card containing only 400 images. The training library file size for CNN is 9GB, nearly 100 times that of DBN. In terms of accuracy and efficiency, DBN's training accuracy for dynamometer card data is far lower than that of CNN, making it difficult to find a suitable DBN network structure for dynamometer card analysis; however, DBN's computational efficiency is far higher than that of CNN. Through problem analysis and research, the intuitiveness and timeliness of graphics are organically integrated with the accuracy and effectiveness of data, leading to the determination of the "CNN + DBN combined graphical intelligent diagnosis" scheme.

[0068] 2) Design a CNN system architecture for intelligent diagnosis of oil well conditions, and perform relevant deep convolutional neural network core calculations. The system architecture has a total of 14 layers, including 5 convolutional layers, 3 pooling layers, 3 hidden layers, 1 input layer, 1 rasterization layer, 1 output layer, and 2,254,592 neurons.

[0069] Convolution kernel operation formula:

[0070]

[0071] In the formula O j — Output data of the j-th layer neuron, φ — ReLU activation function, b j —Bias of neurons in layer j, —Calculate the weighted sum of each neuron from i to j, where The summation operation is performed sequentially from the neurons in layer i, X. i —Input variables of layer i neurons, V—Weight coefficients, Θ j —Filter mapping to layer j neurons.

[0072] Pooling core operation formula:

[0073] S j =downsample(O j )

[0074] In the formula S j — Output results of pooling processing of layer j neurons, O j — Output data of neurons in layer j.

[0075] Rasterization core operation formula:

[0076]

[0077] In the formula X jmn —Input variables: two-dimensional spatial coordinates of neurons in layer j; T—projected coordinate mapping; O k — Output rasterized pixel coordinates;

[0078] 3) Design a DBN system architecture for intelligent diagnosis of oil well conditions, and perform relevant deep convolutional neural network core calculations. The joint probability distribution of the energy function of the input variable x and the hidden variable h is defined as follows:

[0079]

[0080] In the formula, p represents the model probability distribution, e represents the natural logarithm, and Z = ∑ x,h e-energy(x,h) is called the partition function, where energy(x,h) is the energy function, and ∑ x,h e -θ —The negative logarithmic summation of input variable x and latent variable h is used to perform forward stacked RBM learning from low to high level and backward fine-tuning learning from high to low level;

[0081] 4) Design a CNN+DBN system architecture for intelligent diagnosis of oil well conditions. The CNN branch is used to extract dynamometer card information, and the DBN branch is used to input numerical data. After numerical processing by the rasterization layer, the processed data is vectorized in the core stitching layer and input into the fully connected neural network to give the final output, realizing the coupling of CNN and DBN and performing deep hybrid neural network core calculation.

[0082] 5) Select representative datasets to analyze the effectiveness of the CNN+DBN system architecture for intelligent diagnosis of oil well conditions;

[0083] 6) Set default learning parameters and use the established oil well condition sample library to train and test the "CNN+DBN" neural network for intelligent diagnosis of oil well conditions;

[0084] 7) Use techniques such as architecture optimization, iterative training, and Dropout to further optimize the neural network, thereby improving its training efficiency and generalization ability.

[0085] Step 3: Establish an intelligent monitoring system for oil well operating conditions, realize real-time connection, analysis, and push between the intelligent monitoring neural network and the oil well production database, and design a client-side web page system for intelligent monitoring of oil well operating conditions.

[0086] The aforementioned intelligent monitoring system for oil well operating conditions includes:

[0087] 1) Design the overall architecture of the intelligent monitoring software system for oil well operating conditions;

[0088] 2) Perform intelligent monitoring and calculation of oil well operating conditions to achieve remote data acquisition, analysis, and storage;

[0089] 3) Design the front-end web interface for intelligent monitoring of oil well operating conditions.

[0090] In a specific embodiment 1 of the present invention, such as Figure 1 As shown, Figure 1 This is a flowchart of the intelligent diagnostic method for oil well conditions that combines numerical and graphical methods according to the present invention.

[0091] Step 101 involves collecting, organizing, cleaning, extracting features, and sorting production monitoring data from tens of millions of oil wells, and establishing a large-scale oil well condition sample library consisting of "dynamometer diagrams + multi-dimensional time-series data" in a hierarchical and categorized manner.

[0092] The establishment of a large-scale oil well operating condition sample library consisting of "dynamometer diagrams + multi-dimensional time-series data" includes:

[0093] 1) Collect oil well production monitoring data from oil production plants, carry out data sorting, screening, cleaning, and data preprocessing, and realize batch automatic standardized processing of data to improve data quality;

[0094] 2) Combining big data on oil well production monitoring, oil well idle reports, and field experience, the oil well production conditions are categorized from a technical perspective into major categories such as reservoir problems, tubing problems, rod problems, pump problems, surface problems, and data problems;

[0095] 3) Based on the results of the technical classification of oil well operating conditions and the actual needs of oilfield production management, oil well production conditions are divided into three categories from the perspective of production operation: healthy production, sub-healthy production, and abnormal production. Differentiated management measures are specified for different categories.

[0096] 4) To address the problem that relying solely on dynamometer charts is insufficient for accurate diagnosis of certain operating conditions, the system analyzes the dynamometer chart acquisition table, dynamometer chart analysis table, and oil well real-time table in the Production Control System (PCS) database, and selects multi-dimensional time-series data indicators for operating condition diagnosis.

[0097] 5) Conduct data divergence and correlation analysis on more than 60 types of multidimensional time series data, further select data indicators with strong representativeness, and realize the representation of all indicators with the fewest indicators for use in working condition diagnosis analysis.

[0098] 6) For the selected multidimensional time-series data, for indicators in the PCS database that are collected every 30 minutes, the relative rate of change before and after 30 minutes is directly used to characterize their instantaneous change trend.

[0099]

[0100] DX > 0: The indicator increases in the two data collections; DX = 0: The indicator remains unchanged in the two data collections; DX < 0: The indicator decreases in the two data collections. The time window used for the sample is determined to be transient, short-term, medium-term, or long-term. The time series feature extraction algorithm is used to extract its intrinsic features as parameters for working condition analysis.

[0101] Step 102: Based on the characteristics of oil well operating condition diagnosis problems, the applicability of deep learning algorithms such as CNN and DBN is analyzed in depth. A "CNN+DBN" oil well operating condition diagnosis neural network is proposed and established to complete the learning of the operating condition sample database. The algorithm performance is further improved through repeated optimization.

[0102] The aforementioned "CNN+DBN" oil well condition diagnosis neural network includes:

[0103] 1) Addressing the characteristics of oil well condition diagnosis, this study delves into the applicability of Convolutional Neural Networks (CNNs) and Deep Belief Networks (DBNs). Regarding data input format, DBNs require significantly less storage space than CNNs: CNNs use image input, with a single dynamometer card containing up to 40,000 images, while DBNs use array input, with a single dynamometer card containing only 400 images. The training library file for CNNs is 9GB in size, nearly 100 times that of DBNs. In terms of accuracy and efficiency, DBNs have a much lower training accuracy for dynamometer card data than CNNs, making it difficult to find a suitable DBN network structure for dynamometer card analysis; however, DBNs have a much higher computational efficiency than CNNs. Through problem analysis and research, the intuitiveness and timeliness of graphics are organically integrated with the accuracy and effectiveness of data, leading to the determination of the "CNN+DBN Digital Graphical Intelligent Diagnosis" scheme.

[0104] 2) Design a CNN system architecture for intelligent diagnosis of oil well conditions. The system architecture has a total of 14 layers, including 5 convolutional layers, 3 pooling layers, 3 hidden layers, 1 input layer, 1 rasterization layer, 1 output layer, and 2,254,592 neurons. The core convolution operation formula is as follows:

[0105]

[0106] In the formula O j — Output data of the j-th layer neuron, φ — ReLU activation function, b j —Bias of neurons in layer j, —Calculate the weighted sum of each neuron from i to j, where The summation operation is performed sequentially from the neurons in layer i, X. i —Input variables of layer i neurons, V—Weight coefficients, Θ j —Filter mapping to layer j neurons.

[0107] Pooling core operation formula:

[0108] S j =downsample(O j )

[0109] In the formula S j — Output results of pooling processing of layer j neurons, O j — Output data of neurons in layer j.

[0110] Rasterization core operation formula:

[0111]

[0112] In the formula X jmn —Input variables: two-dimensional spatial coordinates of neurons in layer j; T—projected coordinate mapping; O k— Output rasterized pixel coordinates;

[0113] 3) Design a DBN system architecture for intelligent diagnosis of oil well operating conditions. The joint probability distribution of the energy function of the input variable x and the latent variable h is defined as follows:

[0114]

[0115] In the formula, p represents the model probability distribution, e represents the natural logarithm, and Z = ∑ x,h e -energy(x,h) This is called the partition function, where energy(x,h) is the energy function, and ∑ x,h e -θ —The negative logarithmic summation of input variable x and latent variable h is used to perform forward stacked RBM learning from low to high level and backward fine-tuning learning from high to low level;

[0116] 4) such as Figure 2 The design proposes a CNN+DBN system architecture for intelligent diagnosis of oil well conditions. The CNN branch is used to extract dynamometer card information, and the DBN branch is used to input numerical data. After numerical processing by the rasterization layer, the processed data is vectorized and concatenated in the core stitching layer. The data is then input into a fully connected neural network to provide the final output, thereby achieving coupling between CNN and DBN and performing core calculations of a deep hybrid neural network.

[0117] 5) Select representative datasets to analyze the effectiveness of the CNN+DBN system architecture for intelligent diagnosis of oil well conditions.

[0118] Learning from a database of working conditions and further improving algorithm performance through repeated optimization includes the following steps:

[0119] 1) such as Figure 3 By setting default learning parameters, the "CNN+DBN" neural network for intelligent diagnosis of oil well conditions was trained and tested using an established oil well condition sample library. 80% of the data was used for training and 20% for validation. The initial training and validation accuracy reached 91.6%, and the diagnostic efficiency was as high as 49.3ms / time.

[0120] 2) such as Figure 4 Further optimize neural networks using techniques such as architecture optimization, iterative training, and Dropout to improve their training efficiency and generalization ability.

[0121] Step 103: Establish an intelligent monitoring system for oil well operating conditions, realize real-time connection, analysis, and push between the intelligent monitoring neural network and the oil well production database, and establish a customer front-end web page system for intelligent monitoring of oil well operating conditions.

[0122] The aforementioned establishment of an intelligent monitoring system for oil well operating conditions includes:

[0123] 1) Design the overall architecture of the intelligent monitoring software system for oil well operating conditions;

[0124] 2) such as Figure 5 Perform intelligent monitoring and calculation of oil well operating conditions to achieve remote data capture, analysis and storage;

[0125] Write a data query statement and submit it to the PCS database. If there is new data in the PCS database, parse the new data. On the one hand, load the working condition diagnostic card drawing function to draw the working condition diagnostic card. On the other hand, generate the diagnostic data body through steps 5) and 6) in step 101. Load the neural network calculation core obtained in step 102, pass the diagnostic data body into the neural network, and call the neural network to perform calculations. Design a diagnostic database and an alarm database, and store the calculation results in the diagnostic database. Design an alarm strategy, and determine whether the calculation results meet the alarm strategy. If they do, store the alarm data; if they do not, skip this step.

[0126] 3) Design the front-end web interface for intelligent monitoring of oil well operating conditions.

[0127] Example 2:

[0128] In a specific embodiment 2 of the present invention, step 1 involves collecting, organizing, cleaning, extracting features, and sorting production monitoring data from tens of millions of oil wells to establish a large-scale oil well condition sample library consisting of "dynamometer diagrams + multi-dimensional time-series data" in a hierarchical and classified manner.

[0129] The establishment of the large-scale oil well operating condition sample library consisting of "dynamometer diagrams + multi-dimensional time-series data" includes:

[0130] 1) Collect oil well production monitoring data from oil production plants, and carry out data sorting, screening, and cleaning to improve data quality;

[0131] 2) Combining big data on oil well production monitoring, oil well idle reports, and field experience, the oil well production conditions are classified and categorized from a technical perspective into issues such as reservoir problems, tubing problems, rod problems, pump problems, surface problems, and data problems;

[0132] 3) Based on the results of the technical classification of oil well operating conditions and the actual needs of oilfield production management, oil well production conditions are divided into three categories from the perspective of production operation: healthy production, sub-healthy production, and abnormal production. Differentiated management measures are specified for different categories.

[0133] 4) To address the issue that relying solely on dynamometer diagrams is insufficient for accurate diagnosis of certain operating conditions, the system analyzes data indicators in the Production Control System (PCS) database and selects multi-dimensional time-series data for operating condition diagnosis.

[0134] 5) Conduct data divergence and correlation analysis on more than 60 types of multidimensional time series data, and further select data indicators with strong representativeness;

[0135] 6) For the selected multidimensional time series data, determine the time window for sample use: transient, short-term, medium-term, and long-term. Use time series feature extraction algorithm to extract its intrinsic features as operating condition analysis parameters.

[0136] Step 2: Based on the characteristics of oil well operating condition diagnosis problems, we conduct an in-depth analysis of the applicability of deep learning algorithms such as CNN and DBN, propose and establish a "CNN+DBN" oil well operating condition diagnosis neural network, complete the learning of the operating condition sample database, and further improve the algorithm performance through repeated optimization.

[0137] The aforementioned "CNN+DBN" oil well condition diagnosis neural network includes:

[0138] 1) Based on the characteristics of oil well condition diagnosis, this paper analyzes the applicability of Convolutional Neural Networks (CNN) and Deep Belief Networks (DBN). Regarding data input format, DBN requires significantly less storage space than CNN: CNN uses images as input, with a single dynamometer card containing up to 40,000 images, while DBN uses arrays, with a single dynamometer card containing only 400 images. The training library file size for CNN is 9GB, nearly 100 times that of DBN. In terms of accuracy and efficiency, DBN's training accuracy for dynamometer card data is far lower than that of CNN, making it difficult to find a suitable DBN network structure for dynamometer card analysis; however, DBN's computational efficiency is far higher than that of CNN. Through problem analysis and research, the intuitiveness and timeliness of graphics are organically integrated with the accuracy and effectiveness of data, leading to the determination of the "CNN + DBN graphical intelligent diagnosis" scheme.

[0139] 2) Design a CNN system architecture for intelligent diagnosis of oil well conditions. The system architecture has a total of 14 layers, including 2 convolutional layers, 2 pooling layers, 2 hidden layers, 1 input layer, 1 rasterization layer, 1 output layer, and 1,142,816 neurons. The core convolution operation formula is as follows:

[0140]

[0141] In the formula O j — Output data of the j-th layer neuron, φ — ReLU activation function, b j —Bias of neurons in layer j, —Calculate the weighted sum of each neuron from i to j, where The summation operation is performed sequentially from the neurons in layer i, X. i —Input variables of layer i neurons, V—Weight coefficients, Θ j —Filter mapping to layer j neurons.

[0142] Pooling core operation formula:

[0143] Sj =downsample(O j )

[0144] In the formula S j — Output results of pooling processing of layer j neurons, O j — Output data of neurons in layer j.

[0145] Rasterization core operation formula:

[0146]

[0147] In the formula X jmn —Input variables: two-dimensional spatial coordinates of neurons in layer j; T—projected coordinate mapping; O k — Output rasterized pixel coordinates;

[0148] 3) Design a DBN system architecture for intelligent diagnosis of oil well conditions, and perform relevant deep convolutional neural network core calculations. The joint probability distribution of the energy function of the input variable x and the hidden variable h is defined as follows:

[0149]

[0150] In the formula, p represents the model probability distribution, e represents the natural logarithm, and Z = ∑ x,h e -energy(x,h) This is called the partition function, where energy(x,h) is the energy function, and ∑ x,h e -θ —The negative logarithmic summation of input variable x and latent variable h is used to perform forward stacked RBM learning from low to high level and backward fine-tuning learning from high to low level;

[0151] 4) Design a CNN+DBN system architecture for intelligent diagnosis of oil well conditions. The CNN branch is used to extract dynamometer card information, and the DBN branch is used to input numerical data. After numerical processing by the rasterization layer, the processed data is vectorized in the core stitching layer and input into the fully connected neural network to give the final output, realizing the coupling of CNN and DBN and performing deep hybrid neural network core calculation.

[0152] 5) Select representative datasets to analyze the effectiveness of the CNN+DBN system architecture for intelligent diagnosis of oil well conditions;

[0153] 6) Set the default learning parameters and use the established oil well condition sample library to train and test the "CNN+DBN" neural network for intelligent diagnosis of oil well conditions. 80% of the data is used for training and 20% for validation. The initial training and validation accuracy reached 87.3%, and the diagnostic efficiency was as high as 23.5ms / time.

[0154] 7) Use techniques such as architecture optimization, iterative training, and Dropout to further optimize the neural network, thereby improving its training efficiency and generalization ability.

[0155] Step 3: Establish an intelligent monitoring system for oil well operating conditions, realize real-time connection, analysis, and push between the intelligent monitoring neural network and the oil well production database, and design a client-side web page system for intelligent monitoring of oil well operating conditions.

[0156] The aforementioned intelligent monitoring system for oil well operating conditions includes:

[0157] 1) Design the overall architecture of the intelligent monitoring software system for oil well operating conditions;

[0158] 2) Perform intelligent monitoring and calculation of oil well operating conditions to achieve remote data acquisition, analysis, and storage;

[0159] 3) Design the front-end web interface for intelligent monitoring of oil well operating conditions.

[0160] Example 3:

[0161] In a specific embodiment 3 of the present invention, step 1 involves collecting, organizing, cleaning, extracting features, and sorting production monitoring data from tens of millions of oil wells to establish a large-scale oil well condition sample library consisting of "dynamometer diagrams + multi-dimensional time-series data" in a hierarchical and classified manner.

[0162] The establishment of the large-scale oil well operating condition sample library consisting of "dynamometer diagrams + multi-dimensional time-series data" includes:

[0163] 1) Collect oil well production monitoring data from oil production plants, and carry out data sorting, screening, and cleaning to improve data quality;

[0164] 2) Combining big data on oil well production monitoring, oil well idle reports, and field experience, the oil well production conditions are classified and categorized from a technical perspective into issues such as reservoir problems, tubing problems, rod problems, pump problems, surface problems, and data problems;

[0165] 3) Based on the results of the technical classification of oil well operating conditions and the actual needs of oilfield production management, oil well production conditions are divided into three categories from the perspective of production operation: healthy production, sub-healthy production, and abnormal production. Differentiated management measures are specified for different categories.

[0166] 4) To address the issue that relying solely on dynamometer diagrams is insufficient for accurate diagnosis of certain operating conditions, the system analyzes data indicators in the Production Control System (PCS) database and selects multi-dimensional time-series data for operating condition diagnosis.

[0167] 5) Conduct data divergence and correlation analysis on more than 60 types of multidimensional time series data, and further select data indicators with strong representativeness;

[0168] 6) For the selected multidimensional time series data, determine the time window for sample use: transient, short-term, medium-term, and long-term. Use time series feature extraction algorithm to extract its intrinsic features as operating condition analysis parameters.

[0169] Step 2: Based on the characteristics of oil well operating condition diagnosis problems, we conduct an in-depth analysis of the applicability of deep learning algorithms such as CNN and DBN, propose and establish a "CNN+DBN" oil well operating condition diagnosis neural network, complete the learning of the operating condition sample database, and further improve the algorithm performance through repeated optimization.

[0170] The aforementioned "CNN+DBN" oil well condition diagnosis neural network includes:

[0171] 1) Based on the characteristics of oil well condition diagnosis, this paper analyzes the applicability of Convolutional Neural Networks (CNN) and Deep Belief Networks (DBN). Regarding data input format, DBN requires significantly less storage space than CNN: CNN uses images as input, with a single dynamometer card containing up to 40,000 images, while DBN uses arrays, with a single dynamometer card containing only 400 images. The training library file size for CNN is 9GB, nearly 100 times that of DBN. In terms of accuracy and efficiency, DBN's training accuracy for dynamometer card data is far lower than that of CNN, making it difficult to find a suitable DBN network structure for dynamometer card analysis; however, DBN's computational efficiency is far higher than that of CNN. Through problem analysis and research, the intuitiveness and timeliness of graphics are organically integrated with the accuracy and effectiveness of data, leading to the determination of the "CNN + DBN graphical intelligent diagnosis" scheme.

[0172] 2) Design a CNN system architecture for intelligent diagnosis of oil well conditions. The system architecture has a total of 14 layers, including 3 convolutional layers, 3 pooling layers, 3 hidden layers, 1 input layer, 1 rasterization layer, 1 output layer, and 1,215,136 neurons. The core convolution operation formula is as follows:

[0173]

[0174] In the formula O j — Output data of the j-th layer neuron, φ — ReLU activation function, b j —Bias of neurons in layer j, —Calculate the weighted sum of each neuron from i to j, where The summation operation is performed sequentially from the neurons in layer i, X. i —Input variables of layer i neurons, V—Weight coefficients, Θ j —Filter mapping to layer j neurons.

[0175] Pooling core operation formula:

[0176] S j=downsample(O j )

[0177] In the formula S j — Output results of pooling processing of layer j neurons, O j — Output data of neurons in layer j.

[0178] Rasterization core operation formula:

[0179]

[0180] In the formula X jmn —Input variables: two-dimensional spatial coordinates of neurons in layer j; T—projected coordinate mapping; O k — Output rasterized pixel coordinates;

[0181] 3) Design a DBN system architecture for intelligent diagnosis of oil well conditions, and perform relevant deep convolutional neural network core calculations. The joint probability distribution of the energy function of the input variable x and the hidden variable h is defined as follows:

[0182]

[0183] In the formula, p represents the model probability distribution, e represents the natural logarithm, and Z = ∑ x,h e -energy(x,h) This is called the partition function, where energy(x,h) is the energy function, and ∑ x,h e -θ —The negative logarithmic summation of input variable x and latent variable h is used to perform forward stacked RBM learning from low to high level and backward fine-tuning learning from high to low level;

[0184] 4) Design a CNN+DBN system architecture for intelligent diagnosis of oil well conditions. The CNN branch is used to extract dynamometer card information, and the DBN branch is used to input numerical data. After numerical processing by the rasterization layer, the processed data is vectorized in the core stitching layer and input into the fully connected neural network to give the final output, realizing the coupling of CNN and DBN and performing deep hybrid neural network core calculation.

[0185] 5) Select representative datasets to analyze the effectiveness of the CNN+DBN system architecture for intelligent diagnosis of oil well conditions;

[0186] 6) Set the default learning parameters and use the established oil well condition sample library to train and test the "CNN+DBN" neural network for intelligent diagnosis of oil well conditions. 80% of the data is used for training and 20% for validation. The initial training and validation accuracy reached 94.2%, and the diagnostic efficiency reached 31.9ms / time.

[0187] 7) Use techniques such as architecture optimization, iterative training, and Dropout to further optimize the neural network, thereby improving its training efficiency and generalization ability.

[0188] Step 3: Establish an intelligent monitoring system for oil well operating conditions, realize real-time connection, analysis, and push between the intelligent monitoring neural network and the oil well production database, and design a client-side web page system for intelligent monitoring of oil well operating conditions.

[0189] The aforementioned intelligent monitoring system for oil well operating conditions includes:

[0190] 1) Design the overall architecture of the intelligent monitoring software system for oil well operating conditions;

[0191] 2) Perform intelligent monitoring and calculation of oil well operating conditions to achieve remote data acquisition, analysis, and storage;

[0192] 3) Design the front-end web interface for intelligent monitoring of oil well operating conditions.

[0193] Finally, it should be noted that the above description is merely 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 foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0194] Except for the technical features described in the specification, all other technologies are known to those skilled in the art.

Claims

1. A digital-graphical combined intelligent diagnostic method for oil well operating conditions, characterized in that, The method includes: Step 1: Establish a large-scale dynamometer diagram and multi-dimensional time-series data oil well operating condition sample library; including: 1) Collect oil well production monitoring data from oil production plants, and carry out data sorting, screening, and cleaning to improve data quality; 2) Combining big data on oil well production monitoring, oil well idle reports, and field experience, the oil well production conditions are classified and categorized from a technical perspective into reservoir problems, tubing problems, rod problems, pump problems, surface problems, and data problems; 3) Based on the results of the technical classification of oil well operating conditions and the actual needs of oilfield production management, oil well production conditions are divided into three categories from the perspective of production operation: healthy production, sub-healthy production, and abnormal production. Differentiated management measures are specified for different categories. 4) To address the issue that relying solely on dynamometer diagrams is insufficient for accurate diagnosis of certain operating conditions, the system analyzes data indicators in the Production Command System (PCS) database and selects multi-dimensional time-series data for operating condition diagnosis. 5) Conduct data divergence and correlation analysis on more than 60 types of multidimensional time series data, and further screen data indicators with strong representativeness for use in working condition diagnosis and analysis; 6) For the selected multidimensional time series data, determine the time window for the sample to be transient, short-term, medium-term, or long-term, and use the time series feature extraction algorithm to extract its intrinsic features as operating condition analysis parameters; In step 1, a large-scale oil well condition sample library of dynamometer diagrams and multi-dimensional time series data is established by collecting, organizing, cleaning, extracting features and sorting the production monitoring data of tens of millions of oil wells. Step 2: Addressing the characteristics of oil well condition diagnosis, a CNN+DBN-based oil well condition diagnosis neural network is proposed and established. This network learns from the condition sample database and its performance is further improved through repeated optimization. This includes: Step 2-1: Based on the characteristics of oil well condition diagnosis problems, conduct an in-depth analysis of the applicability of Convolutional Neural Network (CNN) and Deep Belief Network (DBN). In step 2-1, through the analysis and investigation of the problem, the intuitiveness and timeliness of the graphics and the accuracy and effectiveness of the data are organically integrated to determine the CNN+DBN intelligent diagnostic solution that combines numbers and graphics. Step 2-2: Design the CNN system architecture for intelligent diagnosis of oil well conditions and perform relevant deep convolutional neural network core calculations; In step 2-2, the system architecture has a total of 14 layers, including 5 convolutional layers, 3 pooling layers, 3 hidden layers, 1 input layer, 1 rasterization layer, 1 output layer, and 2,254,592 neurons; Convolution kernel operation formula: In the formula —Output data of neurons in layer j —ReLU activation function —Bias of neurons in layer j, —Calculate the weighted sum of each neuron from i to j, where The summation operation is performed sequentially from the neurons in layer i. —Input variables of layer i neurons, V—Weight coefficients, —Filter mapping to layer j neurons; Pooling core operation formula: In the formula —Pooling output of layer j neurons —Output data of neurons in layer j; Rasterization core operation formula: In the formula —Input variables: two-dimensional spatial coordinates of neurons in layer j; T—projected coordinate mapping — Output rasterized pixel coordinates; Steps 2-3: Design the DBN system architecture for intelligent diagnosis of oil well operating conditions; In steps 2-3, the joint probability distribution of the energy function for the input variable x and the latent variable h is defined as follows: In the formula, p represents the model probability distribution, e represents the natural constant, and e represents the normalization constant. This is called the partition function, where energy(x,h) is the energy function. —The negative logarithmic summation of input variable x and latent variable h is used to perform forward stacked RBM learning from low to high level and backward fine-tuning learning from high to low level; Steps 2-4: Design the CNN+DBN system architecture for intelligent diagnosis of oil well conditions; In steps 2-4, a CNN branch is used to input the dynamometer card to extract dynamometer card information, and a DBN branch is used to input numerical data, which are the working condition analysis parameters of the multi-dimensional time series data extracted in step 1. After numerical processing by the rasterization layer, the processed data from the CNN branch and the DBN branch are vector-stitched together in the core stitching layer. The stitched vector is input into a fully connected neural network to obtain the oil well working condition diagnosis output, realizing the coupling of CNN and DBN, and performing deep hybrid neural network core calculations. Steps 2-5: Select representative datasets to analyze the effectiveness of the CNN+DBN system architecture for intelligent diagnosis of oil well conditions; Steps 2-6: Set default learning parameters and train and test the CNN+DBN neural network for intelligent diagnosis of oil well conditions using the established oil well condition sample library; Steps 2-7 involve using techniques such as architecture optimization, iterative training, and Dropout to further optimize the neural network, thereby improving its training efficiency and generalization ability. Step 3: Establish an intelligent monitoring system for oil well operating conditions, achieving real-time connection, analysis, and push notifications between the intelligent monitoring neural network and the oil well production database; design a client-side web page system for intelligent oil well operating condition monitoring; including: Step 3-1: Design the overall architecture of the intelligent monitoring software system for oil well operating conditions; Step 3-2: Perform intelligent monitoring and calculation of oil well operating conditions to achieve remote data acquisition, analysis, and storage; Step 3-3: Design and compile the front-end web interface for intelligent monitoring of oil well operating conditions.

Images