Automatic clustering method for oil well working condition samples

Abstract

The invention provides an automatic clustering method for oil well working condition samples. The automatic clustering method for the oil well working condition samples comprises the steps that 1, a big data system for fault diagnosis is acquired from an oil field data center; 2, preprocessing the data, and drawing an indicator diagram; 3, performing feature extraction on the drawn indicator diagram to obtain feature data of the sample; step 4, carrying out clustering analysis on the feature data and verifying a clustering result; and 5, training and testing the determined sample set by using a convolutional neural network to obtain an oil well fault diagnosis model. According to the oil well working condition sample automatic clustering method, indicator diagram features are extracted by using a deep learning method, then indicator diagram samples for fault diagnosis are rapidly and automatically divided, the quality of a sample set for fault diagnosis is effectively improved, and time and labor costs are greatly reduced.

Description

Technical Field

[0001] The present invention relates to the fields of oil production technology and computer machine learning technology, and in particular to an automatic clustering method for oil well operating condition samples. Background Art

[0002] The pumping unit is one of the important equipment for oil extraction. Due to the large number of oil wells and their wide distribution, the pumping rod, pump and other equipment are often placed in oil wells thousands of meters deep. It is difficult to visually judge the fault type and to detect the fault in a timely and accurate manner, resulting in serious economic losses. With the continuous deepening of intelligent construction of oil fields, a large number of sensors are installed in the oil well production system to collect various production data of oil wells in real time.

[0003] The dynamometer diagram is the key basis for oil well production fault diagnosis. With the rapid development of machine learning, various methods such as support vector machine, BP neural network, convolutional neural network, etc. have been tried for oil well fault diagnosis. However, these methods rely on high-quality dynamometer diagram sample sets, which requires a lot of manpower and time to convert dynamometer diagram into a high-quality sample set. Figure 1 A classification.

[0004] In the Chinese patent application with application number: CN202010829024.3, a water well working condition diagnosis method based on multi-parameter analysis is involved. The water well working condition diagnosis method based on multi-parameter analysis includes: step 1, analyzing the factors affecting the working condition of the water injection well and establishing an evaluation index system; step 2, optimizing the coordinate system using the big data clustering algorithm; step 3, dividing the boundaries using the support vector classifier under the visual water absorption index-injection completion rate coordinate system; step 4, establishing a dynamic macro-control chart according to the working condition type of the water injection well. The water well working condition diagnosis method based on multi-parameter analysis conducts analysis of the factors affecting the working condition and research on evaluation indicators, adopts big data clustering analysis, and divides the boundaries using the support vector classifier to establish a highly adaptable dynamic macro-control chart for water wells, thereby improving the management level of water injection wells.

[0005] In the Chinese patent application with application number: CN202111623600.X, a clustering-based process industry energy consumption status assessment method is involved, which obtains the global production raw data of the process industry and the corresponding total energy consumption data, and generates a sample data set; the sample data set of the process industry is unsupervisedly classified according to the clustering algorithm to determine the number of stable operating conditions of the process industry production; the sample data set is divided into several clusters according to the number of stable operating conditions of the process industry production, and the energy consumption corresponding to the sample data in each cluster is averaged; the sample data in the cluster with the smallest energy consumption mean is used as the global optimal energy consumption of the process industry; the invention uses a clustering method to analyze the global data, does not need to model the process itself, and greatly reduces the input of scientific researchers in energy consumption analysis.

[0006] In the Chinese patent application with application number: CN201510127783.4, a method for optimizing oil production of a pumping unit based on a multi-operating condition model is involved, including: obtaining a plurality of typical operating condition types according to the historical indicator diagram data of the pumping unit; classifying the historical indicator diagram data according to the plurality of typical operating condition types, and corresponding each indicator diagram data contained in the historical indicator diagram data to each typical operating condition type; generating a neural network model of each typical operating condition type based on the indicator diagram data corresponding to each typical operating condition type; determining the type of real-time operating condition of the pumping unit among the plurality of typical operating condition types; and optimizing the oil production process of the pumping unit based on the neural network model corresponding to the type of real-time operating condition.

[0007] The above existing technologies are significantly different from the present invention and fail to solve the technical problem we want to solve. Therefore, we have invented a new automatic clustering method for oil well operating condition samples. Summary of the invention

[0008] The purpose of the present invention is to provide an automatic clustering method for oil well operating condition samples, which effectively improves the quality of sample sets used for fault diagnosis and greatly reduces time and labor costs.

[0009] The object of the present invention can be achieved by the following technical measures: an automatic clustering method for oil well operating condition samples, the automatic clustering method for oil well operating condition samples comprising:

[0010] Step 1: Obtain the big data system for fault diagnosis from the oil field data center;

[0011] Step 2: Preprocess the data and draw the indicator diagram;

[0012] Step 3: Extract features from the drawn dynamometer diagram to obtain feature data of the sample;

[0013] Step 4: Perform cluster analysis on the feature data and verify the clustering results;

[0014] Step 5: Use the convolutional neural network to train and test the determined sample set to obtain the oil well fault diagnosis model.

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

[0016] In step 1, the big data system of fault diagnosis includes: displacement and load.

[0017] In step 2, check whether the displacement and load are consistent in quantity; whether there is a null value, if any of the above two situations occurs, the data is discarded.

[0018] In step 2, the preprocessed displacement and load each contain N sample points, and the dynamometer diagram can be drawn with the displacement as the horizontal coordinate and the load as the vertical coordinate.

[0019] In step 3, the trained neural network model is loaded and used to extract features of the dynamometer diagram.

[0020] In step 3, a 3-channel image with a resolution of 224*112 is input into the convolutional layer, and the feature map containing local information in the model, including the height, width, and number of channels of the feature map, is mapped to 7*7*512 dimensions.

[0021] In step 4, clustering of the classification number is performed according to the number of clusters to be clustered in the dynamometer diagram and the sum of squares of clustering errors corresponding to different classification numbers are calculated;

[0022] Plot the number of classifications and the sum of squared errors into a line graph and observe the inflection point of the line graph. If an obvious inflection point appears, the corresponding number of classifications is the optimal number of classifications.

[0023] In step 4, the formula for calculating the sum of squared errors is:

[0024]

[0025] Where k is the number of clusters, C i is the i-th cluster, x is C i The point in i is the centroid of the ith cluster.

[0026] In step 4, when verifying the clustering results, the clustering results of the dynamometer diagrams are screened one by one according to the results under the optimal classification and combined with expert experience, and the dynamometer diagrams with inconsistent fault types are removed.

[0027] In step 4, cluster analysis of the feature data includes:

[0028] According to expert experience, set the lower limit m and upper limit n of the data set to be clustered K;

[0029] Use K-means clustering algorithm to cluster the number of categories in [m,n];

[0030] Calculate the sum of squares of clustering errors SSE for each K in [m,n];

[0031] Observe the line graph according to the elbow rule and select the optimal number of clusters K;

[0032] Set k to the optimal number of clusters confirmed by the elbow method and save the clustering results.

[0033] In step 4, the oil well operating condition sample clustering method is to use the VGG16 model to extract the power diagram feature vector, obtain 7*7*512 features, use the K-Means clustering algorithm to perform cluster analysis on the massive power diagram, and automatically divide the samples into different subsets based on the feature differences between the samples.

[0034] In step 4, when verifying the clustering results, after the clustering is completed, quickly check the division of the sample set. If there are individual pictures in a category that are different from other dynamometer diagrams, they need to be removed from the current category to finally determine the sample set.

[0035] The purpose of the present invention can also be achieved through the following technical measures: an automatic clustering system for oil well operating condition samples, which adopts an automatic clustering method for oil well operating condition samples to perform oil well fault diagnosis.

[0036] The automatic clustering method of oil well operating condition samples in the present invention is a method for quickly and automatically dividing the dynamometer diagram of the pumping well according to the fault type. In the context of the initial formation of the big data body of oil well production monitoring, the deep learning method is used to extract the characteristics of the dynamometer diagram, and then the dynamometer diagram samples for fault diagnosis are quickly and automatically divided, which effectively improves the quality of the sample set used for fault diagnosis and greatly reduces the time and labor costs.

[0037] Compared with the prior art, the present invention realizes the rapid and automatic division of the dynamometer diagram. The automatic clustering method of oil well working condition samples provided by the present invention uses a machine learning method to learn and extract the characteristics of the dynamometer diagram when a large number of sample sets are required for the fault diagnosis of a pumping well, and automatically divides the dynamometer diagrams with similar characteristics learned by the computer into the same folder, thus avoiding the problem that manual division consumes a lot of time and is prone to errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flowchart of a specific embodiment of the method for automatically clustering oil well operating condition samples of the present invention;

[0039] Figure 2 It is a work diagram drawn by displacement and load in a specific embodiment of the present invention;

[0040] Figure 3 This is a structural diagram of the network model in a specific embodiment of the present invention;

[0041] Figure 4 A schematic diagram of feature extraction using a convolutional basis in a specific embodiment of the present invention;

[0042] Figure 5 Schematic diagram of the elbow method in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0043] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0044] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations and / or combinations thereof.

[0045] like Figure 1 As shown, Figure 1 The flowchart of the oil well working condition sample automatic clustering method of the present invention. The oil well working condition sample automatic clustering method comprises:

[0046] Step 101: Obtain a big data system for fault diagnosis from an oilfield data center; the big data system for fault diagnosis includes: displacement and load.

[0047] Step 102: pre-processing the data and drawing a dynamometer diagram;

[0048] Check whether the displacement and load are consistent; whether there is a null value. If there is any of the above two situations, the data will be discarded.

[0049] It should be noted that the above situation may occur when errors occur when the sensor transmits the data to the data center or when the data is exported.

[0050] Furthermore, the preprocessed displacement and load each contain N sample points, and the dynamometer diagram can be drawn with the displacement as the abscissa and the load as the ordinate.

[0051] Step 103: extracting features from the drawn dynamometer diagram to obtain feature data of the sample;

[0052] Load the trained neural network model and use it to extract features of the dynamometer diagram.

[0053] The 3-channel image with an input resolution of 224*112 enters the convolution layer. From the model, it can be seen that the feature map containing local information (including the height, width, and number of channels of the feature map) is mapped to 7*7*512 dimensions.

[0054] Step 104: Perform cluster analysis on the feature data and verify the clustering results.

[0055] Cluster the number of classifications according to the number of clusters to be clustered in the indicator diagram and calculate the sum of squares of clustering errors corresponding to different numbers of classifications;

[0056] Plot the number of classifications and the sum of squared errors into a line graph and observe the inflection point of the line graph. If an obvious inflection point appears, the corresponding number of classifications is the optimal number of classifications.

[0057] Step 105: Use a convolutional neural network to train and test the determined sample set to obtain an oil well fault diagnosis model.

[0058] As a preferred solution of the oil well operating condition sample automatic clustering method of the present invention, the error sum of squares is expressed as:

[0059]

[0060] Where k is the number of clusters, C i is the i-th cluster, x is C i The point in i is the centroid of the ith cluster.

[0061] As a preferred solution of the oil well operating condition sample automatic clustering method of the present invention, the clustering result verification includes:

[0062] According to the results of the optimal classification and combined with expert experience, the clustering results of the dynamometer diagrams are screened one by one, and the dynamometer diagrams with inconsistent fault types are removed.

[0063] Furthermore, the automatic clustering method of oil well operating condition samples uses the VGG16 model to extract the power diagram feature vector, obtains 7*7*512 features, and uses the K-Means clustering algorithm to perform cluster analysis on massive power diagrams, and automatically divides the samples into different subsets based on the feature differences between each sample.

[0064] Furthermore, cluster analysis of feature data includes:

[0065] According to expert experience, set the lower limit m and upper limit n of the data set to be clustered K;

[0066] Use K-means clustering algorithm to cluster the number of categories in [m,n];

[0067] Calculate the sum of squares of clustering errors SSE for each K in [m,n];

[0068] The error sum of squares is expressed as:

[0069]

[0070] Draw a line graph of each K and its corresponding classification result, such as Figure 5 ;

[0071] Observe the line graph according to the elbow rule and select the optimal number of clusters K;

[0072] It should be noted that the problem can be transformed into the following using the gradient descent theory tool: in order to ensure the minimum total SSE, each step of K, that is, each increase of K by 1, must be able to reduce SSE to the greatest extent. That is, given a K value step size of 1, we need to find the slope with the fastest SSE decrease rate.

[0073] Set k to the optimal number of clusters confirmed by the elbow method and save the clustering results.

[0074] Furthermore, the clustering results are verified, including:

[0075] After clustering, quickly check the division of the sample set. If there are individual pictures in a category that are different from other dynamometer diagrams, they need to be removed from the current category to finally determine the sample set.

[0076] The following are several specific embodiments of the present invention.

[0077] Example 1

[0078] In a specific embodiment 1 of the present invention, referring to Figure 2-5 This is an embodiment of the present invention. Based on the above method, scientific demonstration is carried out through specific embodiments to verify its beneficial effects.

[0079] According to the big data system of fault diagnosis, the invention needs to obtain: 1000 displacement and load data, each displacement and load data contains 200 sample points respectively. These 400 sample points can constitute a dynamometer diagram (displacement as the horizontal axis, load as the vertical axis). There are a total of 1000 dynamometer diagrams. A displacement and load are shown in Table 1:

[0080] Table 1 Displacement and load table of the power diagram

[0081]

[0082] Preprocess the data to check whether the displacement and load are consistent and whether there are null values. Each displacement and its corresponding load should contain 200 sample points. If there is any of the above two situations, the data will be discarded. The displacement and load after preprocessing contain 200 sample points. The indicator diagram can be drawn with displacement as the horizontal axis and load as the vertical axis. Finally, 1000 indicator diagrams are obtained. The indicator diagram is as follows: Figure 2 shown.

[0083] Load the VGG16 model trained on Imagenet to extract the dynamometer features. Adjust the parameter include_top to False to remove the top output layer. Set the parameter weight to Imagenet to load the pre-trained weights on the dataset Imagenet. Set input_shape to 224*112*3. The default number of classes is 1000. The VGG16 model structure without the classification layer is as follows: Figure 3 The feature extraction structure is shown in Figure 4 As shown in the figure, finally, all the feature maps containing local information (including the height, width, and number of channels of the feature maps) are mapped to 7*7*512 dimensions.

[0084] Furthermore, the automatic clustering method of oil well working condition samples uses the VGG16 model to extract the power diagram feature vector, obtains 7*7*512 features, and uses the K-Means clustering algorithm to perform clustering analysis on the massive power diagram. According to expert experience, the lower limit of the data set to be clustered K is set to 5 and the upper limit is set to 12; the K-means clustering algorithm is used to cluster the number of categories in [5,12]; the corresponding clustering error sum of squares SSE under each K in [5,12] is calculated; the error sum of squares is expressed as:

[0085]

[0086] Draw a line graph of each K and its corresponding classification result, such as Figure 5 ; Observe the line graph according to the elbow rule and select the best cluster number K=8; Set k to 8 and save the clustering results of the indicator diagram.

[0087] It should be noted that the problem can be transformed into the following using the gradient descent theory tool: in order to ensure the minimum total SSE, each step of K, that is, each increase of K by 1, must be able to reduce SSE to the greatest extent. That is, given a K value step size of 1, we need to find the slope with the fastest SSE decrease rate.

[0088] Furthermore, after clustering is completed, the division of the sample set is quickly checked. If there are individual pictures in a category that are different from other dynamometer diagrams, they need to be removed from the current category and the sample set is finally determined.

[0089] Example 2

[0090] In the specific embodiment 2 of the present invention, based on the above method, scientific demonstration is carried out through specific embodiments to verify its beneficial effects.

[0091] According to the big data system of fault diagnosis, the following required data are obtained for the invention: 2000 displacement and load data each, each displacement and load data contains 200 sample points respectively. These 400 sample points can constitute a dynamometer diagram (displacement as the horizontal axis, load as the vertical axis), and there are a total of 2000 dynamometer diagrams.

[0092] Preprocess the data to check whether the displacement and load are consistent and whether there are null values. Each displacement and its corresponding load should contain 200 sample points. If there is any of the above two situations, the data will be discarded. The displacement and load after preprocessing contain 200 sample points. The indicator diagram can be drawn with displacement as the horizontal axis and load as the vertical axis, and finally 2000 indicator diagrams are obtained.

[0093] Load the VGG16 model trained on Imagenet to extract the dynamometer features. Adjust the parameter include_top to False to remove the top output layer. Set the parameter weight to Imagenet to load the pre-trained weights on the dataset Imagenet. Set input_shape to 224*112*3. The default number of classes is 1000. The VGG16 model structure without the classification layer is as follows: Figure 3 The feature extraction structure is shown in Figure 4 As shown in the figure, finally, all the feature maps containing local information (including the height, width, and number of channels of the feature maps) are mapped to 7*7*512 dimensions.

[0094] Furthermore, the automatic clustering method of oil well working condition samples uses the VGG16 model to extract the power diagram feature vector, obtains 7*7*512 features, and uses the K-Means clustering algorithm to perform clustering analysis on the massive power diagram. Based on expert experience, the lower limit of the K of the data set to be clustered is set to 10 and the upper limit is set to 20; the K-means clustering algorithm is used to cluster the number of categories within [10,20]; the corresponding sum of squares of clustering errors SSE under each K in [10,20] is calculated;

[0095] According to the elbow rule, the optimal number of clusters K=15 is selected; k is set to 15 and the clustering result of the indicator diagram is saved.

[0096] Furthermore, after clustering is completed, the division of the sample set is quickly checked. If there are individual pictures in a category that are different from other dynamometer diagrams, they need to be removed from the current category and the sample set is finally determined.

[0097] Example 3

[0098] In specific embodiment 3 of the present invention, based on the above method, scientific demonstration is carried out through specific embodiments to verify its beneficial effects.

[0099] According to the big data system of fault diagnosis, the following required data are obtained for the invention: 5,000 displacement and load data each, each of which contains 200 sample points. These 400 sample points can constitute a dynamometer diagram (with displacement as the horizontal axis and load as the vertical axis), with a total of 5,000 dynamometer diagrams.

[0100] Preprocess the data to check whether the displacement and load are consistent and whether there are null values. Each displacement and its corresponding load should contain 200 sample points. If there is any of the above two situations, the data will be discarded. The preprocessed displacement and load each contain 200 sample points. The indicator diagram can be drawn with displacement as the horizontal axis and load as the vertical axis, and finally 5,000 indicator diagrams are obtained.

[0101] Load the VGG16 model trained on Imagenet to extract the dynamometer features. Adjust the parameter include_top to False to remove the top output layer. Set the parameter weight to Imagenet to load the pre-trained weights on the dataset Imagenet. Set input_shape to 224*112*3. The default number of classes is 1000. The VGG16 model structure without the classification layer is as follows: Figure 3 The feature extraction structure is shown in Figure 4 As shown in the figure, finally, all the feature maps containing local information (including the height, width, and number of channels of the feature maps) are mapped to 7*7*512 dimensions.

[0102] Furthermore, the automatic clustering method of oil well working condition samples uses the VGG16 model to extract the power diagram feature vector, obtains 7*7*512 features, and uses the K-Means clustering algorithm to perform clustering analysis on the massive power diagram. Based on expert experience, the lower limit of the K of the data set to be clustered is set to 20 and the upper limit is set to 50; the K-means clustering algorithm is used to cluster the number of categories within [20,50]; the corresponding clustering error sum of squares SSE under each K in [20,50] is calculated;

[0103] According to the elbow rule, the optimal cluster number K=32 is selected; k is set to 32, and the clustering result of the indicator diagram is saved.

[0104] Furthermore, after clustering is completed, the division of the sample set is quickly checked. If there are individual pictures in a category that are different from other dynamometer diagrams, they need to be removed from the current category and the sample set is finally determined.

[0105] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions recorded in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0106] Except for the technical features described in the specification, all other technical features are known technologies to those skilled in the art.

Claims

1. An automatic clustering method for oil well operating condition samples, characterized in that: The automatic clustering method of oil well operating condition samples includes: Step 1: Obtain the big data system for fault diagnosis from the oil field data center; Step 2: Preprocess the data and draw the indicator diagram; Step 3: Extract features from the drawn dynamometer diagram to obtain feature data of the sample; Step 4: Perform cluster analysis on the feature data and verify the clustering results; Step 5: Use the convolutional neural network to train and test the determined sample set to obtain the oil well fault diagnosis model.

2. The automatic clustering method for oil well operating condition samples according to claim 1, characterized in that: In step 1, the big data system of fault diagnosis includes: displacement and load.

3. The automatic clustering method for oil well operating condition samples according to claim 1, characterized in that: In step 2, check whether the displacement and load are consistent in quantity; whether there is a null value, if any of the above two situations occurs, the data is discarded.

4. The automatic clustering method for oil well operating condition samples according to claim 3, characterized in that: In step 2, the preprocessed displacement and load each contain N sample points, and the dynamometer diagram can be drawn with the displacement as the horizontal coordinate and the load as the vertical coordinate.

5. The automatic clustering method for oil well operating condition samples according to claim 1, characterized in that: In step 3, the trained neural network model is loaded and used to extract features of the dynamometer diagram.

6. The automatic clustering method for oil well operating condition samples according to claim 5, characterized in that: In step 3, a 3-channel image with a resolution of 224*112 is input into the convolutional layer, and the feature map containing local information in the model, including the height, width, and number of channels of the feature map, is mapped to 7*7*512 dimensions.

7. The automatic clustering method for oil well operating condition samples according to claim 1, characterized in that: In step 4, clustering of the classification number is performed according to the number of clusters to be clustered in the dynamometer diagram and the sum of squares of clustering errors corresponding to different classification numbers are calculated; Plot the number of classifications and the sum of squared errors into a line graph and observe the inflection point of the line graph. If an obvious inflection point appears, the corresponding number of classifications is the optimal number of classifications.

8. The automatic clustering method for oil well operating condition samples according to claim 7, characterized in that: In step 4, the formula for calculating the sum of squared errors is: Where k is the number of clusters, C i is the i-th cluster, x is C i The point in i is the centroid of the ith cluster.

9. The automatic clustering method for oil well operating condition samples according to claim 1, characterized in that: In step 4, when verifying the clustering results, the clustering results of the dynamometer diagrams are screened one by one according to the results under the optimal classification and combined with expert experience, and the dynamometer diagrams with inconsistent fault types are removed.

10. The automatic clustering method for oil well operating condition samples according to claim 1, characterized in that: In step 4, cluster analysis of the feature data includes: According to expert experience, set the lower limit m and upper limit n of the data set to be clustered K; Use K-means clustering algorithm to cluster the number of categories in [m,n]; Calculate the sum of squares of clustering errors SSE for each K in [m,n]; Observe the line graph according to the elbow rule and select the optimal number of clusters K; Set k to the optimal number of clusters confirmed by the elbow method and save the clustering results.

11. The automatic clustering method for oil well operating condition samples according to claim 10, characterized in that: In step 4, the oil well operating condition sample clustering method is to use the VGG16 model to extract the power diagram feature vector, obtain 7*7*512 features, use the K-Means clustering algorithm to perform cluster analysis on the massive power diagram, and automatically divide the samples into different subsets based on the feature differences between the samples.

12. The automatic clustering method for oil well operating condition samples according to claim 1, characterized in that: In step 4, when verifying the clustering results, after the clustering is completed, quickly check the division of the sample set. If there are individual pictures in a category that are different from other dynamometer diagrams, they need to be removed from the current category to finally determine the sample set.

13. The automatic clustering system for oil well operating condition samples is characterized by: The oil well operating condition sample automatic clustering system adopts the oil well operating condition sample automatic clustering method described in any one of claims 1 to 12 to perform oil well fault diagnosis.

Images