Oil well working condition intelligent diagnosis method based on deep learning and multi-feature fusion
By using deep learning and multi-feature fusion, a VGGNet model was constructed to extract pump dynamometer features and combine them with expert engineering features. This solved the problems of high misjudgment rate and poor generalization ability in existing oil well condition diagnosis, achieving high-accuracy intelligent diagnosis and improving oilfield production efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NORTHEAST GASOLINEEUM UNIV
- Filing Date
- 2026-02-03
- Publication Date
- 2026-05-15
AI Technical Summary
Existing oil well condition diagnosis methods rely on manual analysis, resulting in a high misjudgment rate, poor generalization ability, and an inability to effectively extract the shape features of dynamometer cards, leading to low oilfield production efficiency.
A deep learning-based, multi-feature fusion approach is adopted. By constructing a VGGNet deep convolutional neural network model to extract the contour features of the pump dynamometer, and combining them with expert engineering features, a multi-feature fusion-based operating condition diagnostic model is established to achieve intelligent diagnosis of real-time operating conditions.
It improves the accuracy and generalization ability of oil well condition diagnosis, ensures the stable operation of oil well equipment, and reduces equipment failure rate and safety risks.
Smart Images
Figure CN122049591A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas development technology, and more specifically to an intelligent diagnostic method for oil well conditions based on deep learning and multi-feature fusion. Background Technology
[0002] Currently, most oilfields have entered the mid-to-late stages of development, with decreasing production efficiency. Rod pumping units still dominate crude oil extraction. However, because many critical components of these pumping units are located hundreds of meters deep within the wellbore, in constant contact with a mixture of oil, gas, water, and even sand, the harsh downhole environment leads to a high equipment failure rate. If pumping units operate under faulty conditions for extended periods, it will result in high energy consumption, mechanical damage, and even safety accidents. Therefore, timely and accurate monitoring of the pumping unit's operating status, along with fault analysis and handling, is one of the key issues in oilfield development.
[0003] In recent years, machine learning-based fault diagnosis methods have addressed the difficulty of reusing manual analysis methods. For example, expert systems, spectrum analysis, support vector machines, and artificial neural network models have been applied to the operational status diagnosis of pumping wells, gradually replacing traditional manual analysis methods and achieving some practical application in oilfields. However, in the process of feature extraction from dynamometer cards using conventional artificial neural network models, the image data of the dynamometer card is converted into a set of low-dimensional features, inevitably resulting in the loss of a large amount of effective information. This prevents the model from directly and effectively learning the shape features of the dynamometer card, and existing methods still cannot adequately meet the requirements of actual oilfield production. Therefore, how to provide an intelligent diagnostic method for oil well operating conditions based on deep learning and multi-feature fusion is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0004] In view of this, the present invention provides an intelligent diagnosis method for oil well conditions based on deep learning and multi-feature fusion, which solves the problems of reliance on manual labor, high misjudgment rate and poor generalization ability in existing oil well condition diagnosis methods.
[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent diagnosis of oil well operating conditions based on deep learning and multi-feature fusion includes the following steps: S1. Collect historical dynamometer data of the oil well, including suspension point displacement and load data; S2. Preprocess the historical dynamometer card data of oil wells, including normalization, binarization and expansion processing, and extract effective features; S3. Based on the preprocessed historical dynamometer card data of oil wells, establish a sample library containing typical operating conditions, and divide the sample library into a training set and a test set. S4. Construct a deep convolutional neural network model based on VGGNet as a feature extraction network to extract the contour features of the pump power map. S5. Extract the contour and position features of the pump power diagram and perform feature fusion to obtain the graphic features of the pump power diagram. Combine these with expert engineering features to establish a multi-feature fusion working condition diagnostic model. S6. Train the working condition diagnosis model using the training set, and verify the performance of the working condition diagnosis model using the test set to achieve intelligent diagnosis of real-time working conditions.
[0006] Optionally, in S1, the displacement and load data of the suspension point of the oil well are collected, and the historical dynamometer data of the oil well are obtained through inversion.
[0007] Optional, typical operating conditions in S3 include normal, insufficient liquid supply, gas influence, waxing, piston disengagement from working cylinder, continuous pumping and spraying, floating valve leakage, pump collision, sand discharge, rod breakage, fixed valve leakage, and pump jamming.
[0008] Optionally, the feature extraction network in S4 includes 5 layers of single convolutional neural networks and 3 layers of fully connected layers, with the output being a one-dimensional vector.
[0009] Optionally, expert engineering features in S5 include fill factor, load difference, curvature features, peak count, and smoothness index, which are used to help distinguish similar working conditions.
[0010] Optionally, the extraction method for expert engineering features in S5 is as follows: calculate the smoothness and fullness of the lower stroke section of the indicator diagram to distinguish between gas influence and insufficient liquid supply; analyze the peak number of the upper stroke section of the indicator diagram to determine the sand discharge condition; calculate the curvature characteristics to identify fixed valve leakage; detect the sudden drop in load value to determine if the piston has disengaged from the working cylinder.
[0011] As can be seen from the above technical solutions, compared with the prior art, the present invention provides an intelligent diagnosis method for oil well operating conditions based on deep learning and multi-feature fusion, which has the following beneficial effects: The multi-feature fusion diagnosis method of the present invention integrates deep learning and expert engineering features, and improves the accuracy and generalization ability of oil well operating condition diagnosis by combining pump power diagram contour features and engineering rule features. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0013] Figure 1 This is a flowchart of the intelligent diagnostic method for oil well operating conditions of the present invention; Figure 2 This is a schematic diagram of a typical working condition of the present invention; Figure 3 This is a flowchart of the indicator diagram preprocessing process of the present invention; Figure 4 This is a schematic diagram of the pumping unit operating condition diagnostic model of the present invention; Figure 5 This is a schematic diagram of the multi-feature fusion working condition diagnostic model of the present invention; Figure 6 This is a schematic diagram of the diagnostic results under normal operating conditions in an embodiment of the present invention; Figure 7 This is a schematic diagram of the diagnostic results for insufficient liquid supply in an embodiment of the present invention; Figure 8 This is a schematic diagram of the gas-affected operating condition diagnosis results in an embodiment of the present invention; Figure 9 This is a schematic diagram of the diagnostic results for a fixed valve leak in an embodiment of the present invention; Figure 10 This is a schematic diagram of the diagnostic results for the piston disengagement working condition in an embodiment of the present invention; Figure 11 This is a schematic diagram of the rod breakage diagnosis results in an embodiment of the present invention; Figure 12 This is a schematic diagram of the diagnostic results for the leaking condition of the floating valve in an embodiment of the present invention; Figure 13 This is a schematic diagram of the diagnostic results for the continuous spraying and pumping operation in an embodiment of the present invention; Figure 14 This is a schematic diagram of the sand discharge condition diagnosis results in an embodiment of the present invention; Figure 15 This is a schematic diagram of the diagnostic results of wax deposition conditions in an embodiment of the present invention; Figure 16 This is a schematic diagram of the pump operating condition diagnosis results in an embodiment of the present invention; Figure 17 This is a schematic diagram of the pump collision condition diagnosis results in an embodiment of the present invention; Figure 18 This is a schematic diagram of the statistical results of fault classification accuracy in an embodiment of the present invention; Figure 19 This is a performance comparison curve of the model in the embodiments of the present invention. Detailed Implementation
[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.
[0015] This invention discloses an intelligent diagnostic method for oil well conditions based on deep learning and multi-feature fusion, such as... Figure 1 As shown, it includes the following steps: S1. Collect historical dynamometer data of the oil well, including suspension point displacement and load data; S2. Preprocess the historical dynamometer card data of oil wells, including normalization, binarization and expansion processing, and extract effective features; S3. Based on the preprocessed historical dynamometer card data of oil wells, establish a sample library containing typical operating conditions, and divide the sample library into a training set and a test set. S4. Construct a deep convolutional neural network model based on VGGNet as a feature extraction network to extract the contour features of the pump power map. S5. Extract the contour and position features of the pump power diagram and perform feature fusion to obtain the graphic features of the pump power diagram. Combine these with expert engineering features to establish a multi-feature fusion working condition diagnostic model. S6. Train the working condition diagnosis model using the training set, and verify the performance of the working condition diagnosis model using the test set to achieve intelligent diagnosis of real-time working conditions.
[0016] Furthermore, in S1, the displacement and load data of the suspension point of the oil well are collected, and the historical dynamometer data of the oil well are obtained through inversion.
[0017] In this embodiment, as Figure 3 As shown, S2 is specifically: S21. Data Normalization: Normalize the original displacement and load data of the oil well to the [0,1] interval. Calculation formula: ; In the formula, and The x and y axes of the dynamometer diagram after data normalization are: and For pump displacement and load data, and This provides the displacement data of the pump at its highest and lowest points within one motion cycle. and These are the maximum and minimum values of the load data; S22. Image binarization: Convert the pump power graph into a binary image, so that the entire pump power graph image presents a black and white effect, the image becomes simpler, and the amount of data is reduced; S23. Image dilation: Dilates a binary image to highlight its contour information.
[0018] Furthermore, such as Figure 2 As shown, typical operating conditions in S3 include normal, insufficient liquid supply, gas influence, wax formation, piston dislodging from the working cylinder, continuous pumping and spraying, floating valve leakage, pump collision, sand discharge, rod breakage, fixed valve leakage, and pump jamming.
[0019] Furthermore, the feature extraction network in S4 consists of 5 layers of single convolutional neural networks and 3 layers of fully connected layers, with the output being a one-dimensional vector.
[0020] In this embodiment, as Figure 4 As shown, the input image size of the feature extraction network is (180, 180). The first 5 layers are used to extract image features, and the last 3 layers are used to map image features to image categories. Here, Conv3-64+ReLU represents a convolution operation with a (3, 3) kernel, outputting 64 feature maps, followed by a non-linear mapping operation using ReLU as the non-linear mapping function. maxpool represents a downsampling operation using the max pooling method. FC+Dropout represents a fully connected layer containing a Dropout operation, and FC+Softmax represents a fully connected layer that then uses the Softmax method to output the classification result.
[0021] like Figure 5 As shown, the working condition diagnosis model first removes dynamometer diagrams with abnormal data and graphics, extracts the contour features and position features of the pump dynamometer diagram and performs feature fusion to obtain the graphic features of the pump dynamometer diagram. Based on the graphic features of the pump dynamometer diagram, the initial pumping unit working condition is calculated, and the pumping unit working condition is further subdivided by combining engineering features to obtain the final working condition.
[0022] Furthermore, the expert engineering features in S5 include fill factor, load difference, curvature features, peak count, and smoothness index, which are used to help distinguish similar working conditions.
[0023] Furthermore, the extraction method for expert engineering features in S5 is as follows: calculate the smoothness and fullness of the lower stroke section of the indicator diagram to distinguish between gas influence and insufficient liquid supply; analyze the peak number of the upper stroke section of the indicator diagram to determine the sand-producing condition; calculate the curvature characteristics to identify fixed valve leakage; detect the sudden drop in load value to determine if the piston has disengaged from the working cylinder.
[0024] In this embodiment of the invention, an indicator diagram dataset of a rod-operated pumping system was used for the experiment. The input data pixel size was 180×180. The program was written based on the PyCharm platform, using Tersonflow as the framework and Python as the basic language. The model was trained for 100 epochs, and the results are as follows. Figures 6-17 The model ultimately converged with an accuracy of 94% on the training set and 91% on the test set. The loss on the training set was 0.08 and the loss on the test set was 0.24 when the model converged. The operational condition diagnostic model incorporating engineering features was compared with four other classic convolutional neural networks under the same experimental conditions, including VGG, LeNet5, AlexNet, and ResNet as the comparison experimental group. The results are as follows: Figure 18 As shown. Figure 18 The statistics show the test accuracy of all techniques in 10 experiments. The black crosses within the boxes in the figure correspond to the average fault classification accuracy of the methods shown. It can be seen that the method proposed in this embodiment is superior to the reference fault diagnosis model used in the comparative experimental group. Specifically, the average fault classification accuracy of the method proposed in this embodiment is higher than that of other models in 10 experiments, with a fault classification accuracy exceeding 93.5%.
[0025] like Figure 19 As shown, in terms of model accuracy convergence, the working condition diagnosis model after incorporating engineering features tends to converge after about 40 epochs, ranking among the top in convergence speed among all models, and can train a better model in a shorter time.
[0026] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0027] Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for intelligent diagnosis of oil well operating conditions based on deep learning and multi-feature fusion, characterized in that, Includes the following steps: S1. Collect historical dynamometer data of the oil well, including suspension point displacement and load data; S2. Preprocess the historical dynamometer card data of oil wells, including normalization, binarization and expansion processing, and extract effective features; S3. Based on the preprocessed historical dynamometer card data of oil wells, establish a sample library containing typical operating conditions, and divide the sample library into a training set and a test set. S4. Construct a deep convolutional neural network model based on VGGNet as a feature extraction network to extract the contour features of the pump power map. S5. Extract the contour and position features of the pump power diagram and perform feature fusion to obtain the graphic features of the pump power diagram. Combine these with expert engineering features to establish a multi-feature fusion working condition diagnostic model. S6. Train the working condition diagnosis model using the training set, and verify the performance of the working condition diagnosis model using the test set to achieve intelligent diagnosis of real-time working conditions.
2. The intelligent diagnostic method for oil well operating conditions based on deep learning and multi-feature fusion according to claim 1, characterized in that, S1 collects the suspension point displacement and load data of the oil well, and obtains the historical dynamometer data of the oil well through inversion.
3. The intelligent diagnostic method for oil well operating conditions based on deep learning and multi-feature fusion according to claim 1, characterized in that, Typical operating conditions in S3 include normal, insufficient liquid supply, gas influence, waxing, piston dislodging from the working cylinder, continuous pumping and spraying, floating valve leakage, pump collision, sand discharge, rod breakage, fixed valve leakage, and pump jamming.
4. The intelligent diagnostic method for oil well operating conditions based on deep learning and multi-feature fusion according to claim 1, characterized in that, The feature extraction network in S4 consists of 5 layers of single convolutional neural networks and 3 layers of fully connected layers, with the output being a one-dimensional vector.
5. The intelligent diagnostic method for oil well operating conditions based on deep learning and multi-feature fusion according to claim 1, characterized in that, In S5, expert engineering features include fill factor, load difference, curvature characteristics, peak count, and smoothness index, which are used to help distinguish similar working conditions.
6. The intelligent diagnostic method for oil well operating conditions based on deep learning and multi-feature fusion according to claim 1, characterized in that, The specific method for extracting expert engineering features in S5 is as follows: calculate the smoothness and fullness of the lower stroke section of the indicator diagram to distinguish between gas influence and insufficient liquid supply; analyze the peak number of the upper stroke section of the indicator diagram to determine the sand-producing condition; calculate the curvature characteristics to identify fixed valve leakage; and detect the sudden drop in load to determine if the piston has disengaged from the working cylinder.