A method for detecting abnormal pumping wells based on production dynamic data and dynamometer diagram
By combining production dynamic data and power diagrams, and using machine learning models to calculate the fluctuation coefficients, the problems of poor data correlation and large errors in oil well abnormal detection are solved, and the rapid and accurate diagnosis of oil well abnormalities is achieved.
Patent Information
- Application Number
- CN202111488554.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-07
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2041-12-07
AI Technical Summary
The existing oil well abnormal detection methods have poor data correlation, large errors, and cannot accurately reflect changes in oil production and liquid production, and are complex in judgment, so they can only diagnose working conditions rather than abnormal oil wells.
Combining the training classification model of production dynamic data, power diagrams and machine learning, abnormal oil well detection is achieved by calculating the fluctuation coefficient, enhancing data correlation, reducing errors, and improving detection accuracy.
Through data processing and model classification, the defining values of oil production and liquid production are calculated, the fluctuation coefficient is obtained, and abnormal oil wells are quickly and accurately diagnosed, improving the accuracy and simplicity of the detection results.
Smart Images

Figure CN114120043B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of oil field exploration and development, and relates to a method for detecting abnormal oil pumping wells based on production dynamic data and a dynamometer diagram. Background Art
[0002] A certain oil field is an important crude oil development area in my country. However, during the crude oil production process, there are abnormal pumping wells in the area, which brings many inconveniences to crude oil development. The so-called abnormal pumping well refers to abnormal changes in the production conditions of the pumping well, mainly including excessive increase or decrease in liquid volume, or excessive fluctuation in water content.
[0003] At present, some measures and methods have been taken to detect abnormal oil wells, but the following problems exist: (1) During the pumping process, the oil production changes due to human or machine operating errors, which makes the correlation of the detection data worse, affecting the accuracy of the abnormal oil well detection results; (2) The existing analysis methods cannot accurately reflect the changes in the oil and liquid production of the oil wells, and there is a large error between the calculated results and the measured values; (3) The process of judging the abnormality of the oil well is very complicated. The existing analysis methods only diagnose the operating conditions, but not the abnormalities in the oil and liquid production of the oil wells. Summary of the invention
[0004] In view of the technical problems arising in the existing abnormal detection of oil wells, the present invention provides a method for detecting abnormal pumping wells based on production dynamic data and indicator diagrams, which combines the production dynamic data, the indicator diagram and the training classification model in machine learning, and realizes the detection of abnormal oil wells through the calculated fluctuation coefficient, thereby enhancing the relevance of analysis data, reducing analysis errors, and improving the accuracy of detection results. Moreover, the entire detection process is simple, and the diagnosis of abnormal oil wells can be quickly realized.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is:
[0006] A method for detecting abnormal pumping wells based on production dynamic data and dynamometer diagrams comprises the following steps:
[0007] 1) Collect production dynamic data of multiple oil wells and perform missing value filling and denoising;
[0008] 2) From the processed production dynamic data, select the load-displacement data generated by the oil well pumping unit on each production day within a month to obtain the corresponding dynamometer diagram, and classify each dynamometer diagram according to the category; label the dynamometer diagrams of different categories to obtain the corresponding data set;
[0009] 3) Each dynamometer diagram in the data set of step 2) is sequentially input into the convolutional neural network CNN and the support vector machine SVM, and the corresponding dynamometer diagram classification model is obtained after processing;
[0010] 4) According to the classification results of the dynamometer classification model in step 3), the critical value Q of oil production is obtained o The limit value of liquid production Q l ;
[0011] 5) The critical value Q of oil production obtained according to step 4) o And the limit value of liquid production Q l , and combined with the production dynamic data of the oil well in step 1), the fluctuation coefficient δ is obtained;
[0012] 6) Determine whether the oil well is abnormal based on the size of the fluctuation coefficient δ.
[0013] Furthermore, in step 1), the oil well production dynamic data includes the number of oil wells N, the daily oil production of a single oil well x, the monthly oil production of a single oil well z, the daily liquid production of a single oil well y and the total number of production days M of a single oil well in a month, where M is any positive integer between 1 and 31.
[0014] Furthermore, in step 1), the missing values are completed by using a mean filling method.
[0015] Furthermore, in step 2), the categories include category 1, category 2, category 3, category 4, category 5, category 6, category 7, category 8, category 9 and category 10; the labels of the 10 categories are [1,0,0,0,0,0,0,0,0,0], [0,1,0,0,0,0,0,0,0,0], [0,0,1,0,0,0,0,0,0,0,0], [0,0,0,1, 0,0,0,0,0], [0,0,0,0,1,0,0,0,0,0], [0,0,0,0,0,1,0,0,0,0], [0,0,0,0,0,0,1,0,0,0], [0,0,0,0,0,0,0,1,0,0], [0,0,0,0,0,0,0,1,0,0], [0,0,0,0,0,0,0,0,1,0], and [0,0,0,0,0,0,0,0,0,1].
[0016] Furthermore, the specific process in step 3) is:
[0017] 3.1) Fix the size of each dynamometer diagram to 256*256;
[0018] 3.2) Input the dynamometer image into CNN, and obtain image features of size 256*256*32 through convolution operation with a convolution kernel of size 5*5; and then obtain image features of size 128*128*32 through a 2*2 maximum pooling layer;
[0019] 3.3) Repeat the convolution operation and pooling operation described in 3.2) 5 times each to obtain an image feature with a size of 4*4*1024;
[0020] 3.4) Then pass through the fully connected layer to obtain the image features of size 1*1*4096;
[0021] 3.5) Input the image features in step 3.4) into SVM for training and multi-classification operations to obtain a dynamometer diagram classification model.
[0022] Furthermore, in the step 3.5), the dynamometer diagram classification model includes a first category, a second category and a third category; the first category is category 1; the second category includes category 2, category 4, category, category 8 and category 9; the third category includes category 3, category 5, category 6 and category 10.
[0023] Furthermore, in step 4), the oil production limit value Q o And the liquid production limit value Q l The calculation formulas are:
[0024]
[0025]
[0026] Where i refers to the i-th day of production of an oil well, 1≤i≤M; j refers to the j-th oil well, 1≤j≤N; x ij represents the oil production of the jth oil well on the i-th day of production; y ij It indicates the liquid production of the jth oil well on the i-th production day.
[0027] Furthermore, in step 5), the calculation formula of the fluctuation coefficient δ is:
[0028]
[0029] Furthermore, in step 6), when the classification model is the second largest category and the fluctuation coefficient δ is greater than 50%, it is judged as a well with abnormal oil production.
[0030] Furthermore, in step 6), when the classification model is the first category and the third category and the fluctuation coefficient δ is greater than 30%, it is judged as a well with abnormal oil production.
[0031] The beneficial effects of the present invention are:
[0032] 1. The present invention is a method for detecting abnormal pumping wells based on production dynamic data and dynamometer diagrams. The dynamometer diagram classification model is combined with production dynamic data. Through training and model classification, the boundary value is calculated to determine the fluctuation coefficient of daily oil and liquid production, and then the abnormal oil wells are detected based on the fluctuation coefficient. The implementation process of the present invention is simple, and the accuracy is greatly improved compared with the previous method.
[0033] 2. In the present invention, missing value filling and denoising are performed on all the collected production data in order to ensure that the method of the present invention has higher accuracy in subsequent oil well detection; data integrity and format standardization can be achieved through data processing, thereby improving the accuracy of subsequent detection.
[0034] 3. In the present invention, by combining the CNN model with the SVM method, convolution operation and maximum pooling layer operation are performed on the dynamometer image, and the cross entropy function is used as the loss function during the operation to improve the prediction accuracy of the machine learning model for the true probability distribution and ensure the correlation between the analysis data and the actual oil well.
[0035] 4. The present invention obtains the corresponding production dynamic data through the classification model, thereby obtaining the oil production limit value Q o And the liquid production limit value Q l , and further obtain the fluctuation coefficient δ of the oil well, and then judge whether the oil well is an abnormal well according to the fluctuation coefficient δ. The present invention can detect oil wells in different conditions, reduce the erroneous judgment of oil well production abnormality caused by the working condition of the pumping unit, and achieve the purpose of rapid detection of abnormal oil wells. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 A schematic diagram of the method for detecting abnormal oil wells provided by the present invention;
[0037] Figure 2 Schematic diagram of convolution operation and maximum pooling operation of the present invention;
[0038] Figure 3 Accuracy plots for training and testing of the CNN+SVM classification model. DETAILED DESCRIPTION
[0039] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0040] Example
[0041] See also Figure 1 The method for detecting abnormal pumping wells based on production dynamic data and dynamometer diagram provided in this embodiment includes the following steps:
[0042] 1) Collect all dynamic production data of multiple oil wells and perform missing value filling and denoising;
[0043] 2) From the processed production dynamic data, select the load-displacement data generated by the oil well pumping unit during daily oil production within M production days in a month (M is any positive integer between 1 and 31), obtain the corresponding dynamometer diagram, and classify each dynamometer diagram according to the category; label the dynamometer diagrams of different categories to obtain the corresponding data set;
[0044] 3) Each dynamometer diagram in the data set of step 2) is input into CNN and SVM in turn, and the corresponding dynamometer diagram classification model is obtained after processing;
[0045] Among them, CNN is Convolutional Neural Networks, convolutional neural network; SVM is Support Vector Machine, support vector machine;
[0046] 4) According to the classification results of the dynamometer classification model in step 3), the critical value Q of oil production is obtained o The limit value of liquid production Q l ;
[0047] 5) The critical value Q of oil production obtained according to step 4) o And the limit value of liquid production Q l , and combined with the production dynamic data of the oil well in step 1), the fluctuation coefficient δ is obtained;
[0048] 6) Determine whether the oil well is abnormal based on the size of the fluctuation coefficient δ.
[0049] During implementation, since the oil well production data may be missing, messy, or redundant, in order to ensure that the method of the present invention has a high degree of accuracy in subsequent oil well detection, all production data need to be processed. During processing: (1) For missing data, the mean filling method is used to complete the missing data. The specific process is to fill the missing data with the average value of each column of data such as daily oil production and daily liquid production; (2) For redundant data, it is necessary to delete it, such as deleting variables such as formation coefficient that have little correlation with oil production changes. These processed data will be used as input to identify whether the oil well has production anomalies in the future. Therefore, data processing can ensure the integrity of the data and the standardization of the format, thereby improving the accuracy of the subsequent prediction model judgment.
[0050] In this embodiment, in step 2), the data (load-displacement) generated during the daily oil production of each oil well within the production days M in a month is used as the x-value and y-value, and the corresponding indicator diagram is drawn using the function in the python matplotlib library, and the size of the indicator diagram is fixed to 256*256, and then each indicator diagram is classified and labeled.
[0051] The dynamometer is a curve diagram drawn by the dynamometer to show the relationship between the suspension point load and the suspension point displacement. During the operation of the pumping unit, the suspension point reciprocates up and down. Therefore, the dynamometer diagram is a closed curve and has many forms.
[0052] During implementation, for the data collected during the production process of multiple oil wells, the dynamometer diagram is obtained according to the relationship between the suspension point load and the suspension point displacement during the operation of the pumping well; and all the dynamometer diagrams are classified according to the working conditions of the pumping unit. Specifically, they are divided into 10 categories and 10 different dynamometer diagrams are obtained.
[0053] During implementation, the 10 categories specifically include Category 1, Category 2, Category 3, Category 4, Category 5, Category 6, Category 7, Category 8, Category 9 and Category 10, and Category 1 means that the pumping well is working normally, Category 2 means that the pumping well has insufficient fluid supply, Category 3 means that the pumping well is affected by gas, Category 4 means that the pumping well is gas locked, Category 5 means that the pump hits the pump up and down, Category 6 means that the floating valve of the pumping well is slowly closed, Category 7 means that the plunger of the pumping well is out of action, Category 8 means that the floating and fixed valves of the pumping well are leaking, Category 9 means that the pumping well is affected by sand, and Category 10 means that the pumping well is affected by inertia.
[0054] Then, the 10 categories of indicator diagrams are labeled so that each indicator diagram corresponds to a category label. The labels corresponding to the 10 categories are [1,0,0,0,0,0,0,0,0,0], [0,1,0,0,0,0,0,0,0,0], [0,0,1,0,0,0,0,0,0], [0,0,0,1,0,0,0,0,0], [0,0,0,0,1,0,0,0,0], [0,0,0,0,0,1,0,0,0], [0,0,0,0,0,0,1,0,0,0], [0,0,0,0,0,0,0,1,0,0], [0,0,0,0,0,0,0,1,0], and [0,0,0,0,0,0,0,0,0,1].
[0055] The dynamometer diagrams produced every day within M days of a month are classified according to the above 10 categories, and then the corresponding categories are labeled accordingly. The dynamometer diagrams with the same labels are then put in the same set as a data set. During implementation, 10 data sets of different working conditions of the oil pump are finally obtained; each data set then enters step 3) and is output to the network model for training and testing.
[0056] See also Figure 2 In this embodiment, the specific process in step 3) is:
[0057] 3.1) Fix the size of each dynamometer diagram to 256*256; (In this step, the unit of the dynamometer diagram size is pixel, the same below)
[0058] 3.2) Input the image of the dynamometer diagram into CNN, and obtain the image features of size 256*256*32 through the convolution operation of the convolution kernel of size 5*5; and then obtain the image features of size 128*128*32 through the maximum pooling layer of 2*2;
[0059] 3.3) Repeat the convolution operation and pooling operation described in 3.2) 5 times each to obtain an image feature with a size of 4*4*1024;
[0060] 3.4) Then pass through the fully connected layer to obtain the image features of size 1*1*4096;
[0061] 3.5) Input the image features in step 3.4) into SVM for training and multi-classification operations to obtain a classification model.
[0062] During implementation, the machine learning model adopts CNN and SVM methods, and the specific processing process is: the dynamometer image data and corresponding labels obtained in step 2) are input into CNN; in the CNN model, the step size is set to 1, and an image feature with a size of 256*256*32 is obtained through a convolution operation with a convolution kernel of size 5*5, and then an image feature with a size of 128*128*32 is obtained through a 2*2 maximum pooling layer; the obtained 128*128*32 feature is convolved and pooled, using a 5*5 convolution kernel and a 2*2 maximum pooling layer, and a feature image with a size of 64*64*64 is output.
[0063] Repeat the 5*5 convolution operation and the 2*2 maximum pooling layer operation 4 times to obtain a feature image of size 4*4*1024.
[0064] The output of the previous step is passed through a fully connected layer to obtain a feature image of size 1*1*4096, which is converted into a one-dimensional vector and input into the SVM; SVM is used as a classifier, and the feature image obtained in the previous step is used as input to obtain the output result and classification model, and then the machine learning model is trained.
[0065] Through the above steps, the classification results are obtained, and the classification of the dynamometer diagram is realized. CNN is used to extract the features of the dynamometer diagram, mainly through the convolution layer and the pooling layer to extract feature information on the dynamometer diagram.
[0066] During the operation, the cross entropy function is used as the loss function, Adam is used as the optimizer, and SVM is used as the classifier to output the classification results. The cross entropy loss function is:
[0067] F=-∑p(x)logq(x) (4)
[0068] In the above formula, p represents the true label, that is, the correct category; q represents the predicted value. The smaller the cross entropy value, the closer the distribution of the two probabilities is, which can be used to determine the accuracy of the machine learning model for the true probability distribution. In this way, the correlation between the training results and the true values can be improved, and the accuracy of the final prediction results can be enhanced.
[0069] During implementation, in step 3.5), the classification model includes the first category, the second category and the third category; the first category is category 1; the second category includes category 2, category 4, category, category 8 and category 9; the third category includes category 3, category 5, category 6 and category 10.
[0070] Using the trained CNN+SVM model, the working condition of the pumping unit in each oil well during the oil production process can be predicted, and then the results can be further classified. According to the impact on the oil production of the oil well, it can be divided into three categories: the first category is normal, the second category is factors that have a greater impact on the oil production of the oil well, and the third category is factors that have a smaller impact on the oil production of the oil well.
[0071] When classifying, since Category 2 (insufficient liquid supply), Category 4 (gas lock), Category 7 (plunger disengagement), Category 8 (floating and fixed valve leakage), and Category 9 (sand influence) will cause serious consequences on oil production, they can be defined as factors with greater impact on oil production, that is, as the second largest category; Category 3 (gas influence), Category 5 (up and down pump collision), Category 6 (slow closing of floating valve), and Category 10 (inertia influence) have relatively small impact on oil production, so they can be defined as factors with less impact on oil production, as the third largest category.
[0072] In this embodiment, in step 4), when applying the dynamometer diagram, the boundary values are the average daily oil production and the average daily liquid production corresponding to the normal operation of the oil well. In the classification result of step 3), normal operation is category 1, and the sample label is [1,0,0,0,0,0,0,0,0,0]. Each dynamometer diagram belonging to the normal category will correspond to the production dynamic data of the day, such as oil production and liquid production, so the production dynamic data corresponding to the dynamometer diagram of the normal category is used to obtain the boundary value Q of the oil production. o And the limit value of liquid production Q l .
[0073] The main operation steps are as follows: find the oil wells that belong to the normal category after classification, and then find the oil production and liquid production of the normal category oil wells on that day. Suppose the daily oil production of a normal category oil well is x, the daily liquid production is y, the number of production days in a month is M, and the number of normal category oil wells in the well area is N. Then the total oil production and total liquid production of the normal category oil wells in the well area on that day are respectively and The total oil production and total liquid production for the month were and The daily oil production and daily liquid production of all normal oil wells are averaged to obtain the oil production limit value Q o And the liquid production limit value Q l The specific formula is as follows:
[0074]
[0075]
[0076] Where: i refers to the i-th day of production of an oil well, 1≤i≤M; j refers to the j-th oil well, 1≤j≤N; x ij represents the oil production of the jth oil well on the i-th day of production; y ij It indicates the liquid production of the jth oil well on the i-th production day.
[0077] In this embodiment, step 5), according to the oil production limit value Q o And the liquid production limit value Q l , and the daily oil production is x, the daily liquid production is y, and the fluctuation coefficient δ is further obtained. The specific calculation formula is:
[0078]
[0079] In this embodiment, step 6) is to determine the abnormal oil wells according to the fluctuation coefficient.
[0080] Since the daily oil production and daily liquid production of multiple oil wells within M days are used, a fluctuation coefficient δ is established to detect the fluctuation of oil production and liquid production. The oil production limit value Q determined in step 4) o And the liquid production limit value Q l , and input the daily oil production x and liquid production y, the fluctuation coefficient δ can be obtained.
[0081] During implementation, for the third category (factors with less impact on oil production) and the first category (normal operation), if the fluctuation coefficient δ of the daily oil production and liquid production of the oil well is greater than 30%, it is regarded as an oil well with abnormal oil production.
[0082] During implementation, for the second category (factors with a greater impact on oil production), if the fluctuation coefficient δ of the daily oil production and liquid production of an oil well is greater than 50%, it is considered to be an oil well with abnormal oil production.
[0083] This step can detect oil wells in different conditions, reduce the misjudgment of abnormal oil production due to pumping unit working conditions, and achieve the purpose of accurately detecting abnormal oil production wells.
[0084] The method provided by the present invention ensures the integrity of data and the correctness of corresponding dynamometer diagram annotations through data processing, dynamometer diagram classification labels and multiple training processes, thereby ensuring the effectiveness and feasibility of the method. Moreover, the method has a simple process and a small processing error. It can not only diagnose the working condition of the pumping unit, but also diagnose the abnormal oil production of the oil well, thereby providing an improvement basis for the further production of the oil well.
[0085] The advantages of the method for detecting abnormal oil pumping wells provided by the present invention are illustrated by specific comparative tests as follows.
[0086] Comparative group
[0087] A relatively simple method is to use only production performance data to identify abnormally producing wells.
[0088] Test process: First, collect the oil and liquid production of a certain oil well on a certain day, then compare them with the average daily oil and liquid production of the oil well in the previous month, and calculate the fluctuation coefficient δ of the oil well. If the fluctuation coefficient δ is greater than 30%, the oil well is defined as an abnormal oil well. There are 30 oil wells.
[0089] Specifically, this method was used to judge 30 oil wells in a certain well area on a certain day during the oil production process. The final results are shown in Table 1.
[0090] Table 1 Judgment results of abnormal oil wells in comparative group
[0091]
[0092]
[0093] As can be seen from Table 1, using a simple method, 10 oil wells were judged to be abnormal, and the rest were normal oil wells.
[0094] Experimental Group:
[0095] This experiment adopts the method provided by the present invention, using a method combining production dynamic data definition and indicator diagram analysis. First, the indicator diagram is obtained according to the displacement-load of the oil pumping unit of each oil well on that day; the indicator diagrams of the 30 oil wells that produced oil on that day are classified, labeled, trained and classified using the trained CNN+SVM model; a total of 3 major categories are obtained, the first major category is represented by 1, the second major category is represented by 2, and the third major category is represented by 3. After classification, 12 oil wells are obtained in the first major category, 8 oil wells in the second major category, and 10 oil wells in the third major category. Then, by calculating the fluctuation coefficient δ, abnormal oil wells are further judged. The specific results are shown in Table 2.
[0096] Table 2 Judgment results of abnormal oil wells in the test group
[0097]
[0098]
[0099] It can be seen from Table 2 that by using the judgment method provided by the present invention, 7 oil wells with abnormal oil production were obtained.
[0100] From the judgment results of the comparative group and the test group, it can be found that the number of abnormal oil wells judged by the method of the present invention is 3 less than that of the simple judgment method of the comparative group. By monitoring and analyzing the actual working conditions of the oil wells, it is found that the number of oil wells with abnormal oil production is actually 7, which is mainly caused by the failure of the oil pump during the oil production process, rather than by the abnormal oil production in the well.
[0101] Furthermore, by collecting and summarizing all the production data of 30 oil wells within one month, 900 dynamometer diagrams were obtained. These 900 dynamometer diagrams were set to 256*256 pixels through data preprocessing operations, and then these images were divided into training set and test set in a ratio of 8:2. The training set has 720 dynamometer diagrams and the test set has 180 dynamometer diagrams. Then the CNN+SVM model was trained and tested using the training set and test set respectively. Figure 3 The graph shown. Figure 3 The blue curve in the middle is the curve of the classification accuracy of the CNN+SVM model on the training set as the number of iterations increases; the yellow curve is the curve of the classification accuracy of the CNN+SVM model on the test set as the number of iterations increases.
[0102] Depend on Figure 3 It can be found that when the number of iterations reaches 10, the CNN+SVM model achieves a high classification accuracy on both the training set and the test set. After that, the classification results of the CNN+SVM model are combined with production dynamic data to determine abnormal oil wells.
[0103] Therefore, the judgment method provided by the present invention is used: the dynamometer diagram is obtained from the displacement-load data, and then labeling, classification, CNN+SVM model training and classification are performed to obtain the fluctuation coefficient δ. The fluctuation coefficient δ is used to judge the abnormal oil wells, which can effectively enhance the correlation of the analysis data, reduce the analysis error, improve the analysis accuracy, and ultimately improve the accuracy of the judgment results of the abnormal oil wells, thereby conveniently realizing the diagnosis of abnormal oil production wells.
Claims
1. A method for detecting abnormal pumping wells based on production dynamic data and indicator diagram, characterized in that: The following steps are involved: 1) Collect production dynamic data of multiple oil wells and perform missing value filling and denoising; 2) From the processed production dynamic data, select the load-displacement data generated by the oil well pumping unit on each production day within a month to obtain the corresponding dynamometer diagram, and classify each dynamometer diagram according to the category; label the dynamometer diagrams of different categories to obtain the corresponding data set; 3) Each dynamometer diagram in the data set of step 2) is sequentially input into the convolutional neural network CNN and the support vector machine SVM, and the corresponding dynamometer diagram classification model is obtained after processing; 4) According to the classification results of the dynamometer classification model in step 3), the critical value Q of oil production is obtained o And the limit value of liquid production Q l ; The main steps are as follows: find the oil wells that belong to the normal category after classification, and then find the oil production and liquid production of the normal category oil wells on that day; suppose the daily oil production of a normal category oil well is x, the daily liquid production is y, the number of production days in a month is M, and the number of normal category oil wells in the well area is N, then the total oil production and total liquid production of the normal category oil wells in the well area on that day are and The total oil production and total liquid production for the month were and The daily oil production and daily liquid production of all normal oil wells are averaged to obtain the oil production limit value Q o And the liquid production limit value Q l ; The specific formula is as follows: Where: i refers to the i-th day of production of an oil well, 1≤i≤M; j refers to the j-th oil well, 1≤j≤N; x ij represents the oil production of the jth oil well on the i-th day of production; y ij It represents the liquid production of the jth oil well on the i-th day of production; 5) The critical value Q of oil production obtained according to step 4) o And the limit value of liquid production Q l , and combined with the production dynamic data of the oil well in step 1), the fluctuation coefficient δ is obtained; According to the critical value Q of oil production obtained in step 4) o And the limit value of liquid production Q l , and the daily oil production is x, the daily liquid production is y, and the fluctuation coefficient δ is further obtained. The specific calculation formula is: 6) Determine whether the oil well is abnormal based on the size of the fluctuation coefficient δ.
2. The method for detecting abnormal pumping wells based on production dynamic data and dynamometer diagram according to claim 1, characterized in that: In step 1), the oil well production dynamic data includes the number of oil wells N, the daily oil production of a single oil well x, the monthly oil production of a single oil well z, the daily liquid production of a single oil well y and the total number of production days M of a single oil well in a month, where M is any positive integer between 1 and 31.
3. The method for detecting abnormal pumping wells based on production dynamic data and dynamometer diagram according to claim 2, characterized in that: In the step 1), the missing values are filled by using the mean filling method.
4. The method for detecting abnormal pumping wells based on production dynamic data and dynamometer diagram according to claim 3, characterized in that: In the step 2), the categories include category 1, category 2, category 3, category 4, category 5, category 6, category 7, category 8, category 9 and category 10; The labels of the 10 categories are [1,0,0,0,0,0,0,0,0,0], [0,1,0,0,0,0,0,0,0,0], [0,0,1,0,0,0,0,0,0,0], [0,0,0,1,0,0,0,0,0,0], [0,0,0,0,1,0,0,0,0,0], [0,0,0,0,0,1,0,0,0,0], [0,0,0,0,0,0,1,0,0,0], [0,0,0,0,0,0,1,0,0,0], [0,0,0,0,0,0,0 ,0,1,0,0], [0,0,0,0,0,0,0,0,1,0] and [0,0,0,0,0,0,0,0,0,0,1]; Category 1 indicates that the pumping well is working normally, Category 2 indicates insufficient fluid supply to the pumping well, Category 3 indicates gas impact on the pumping well, Category 4 indicates gas lock on the pumping well, Category 5 indicates up and down pump collision on the pumping well, Category 6 indicates slow closing of the traveling valve of the pumping well, Category 7 indicates plunger disengagement of the pumping well, Category 8 indicates leakage of the traveling and fixed valves of the pumping well, Category 9 indicates sand impact on the pumping well, and Category 10 indicates inertia impact on the pumping well.
5. The method for detecting abnormal pumping wells based on production dynamic data and dynamometer diagram according to claim 4, characterized in that: The specific process in step 3) is: 3.1) Fix the size of each dynamometer diagram to 256*256; 3.2) Input the dynamometer image into CNN, and obtain image features of size 256*256*32 through convolution operation with a convolution kernel of size 5*5; and then obtain image features of size 128*128*32 through a 2*2 maximum pooling layer; 3.3) Repeat the convolution operation and pooling operation described in 3.2) 5 times each to obtain an image feature with a size of 4*4*1024; 3.4) Then pass through the fully connected layer to obtain the image features of size 1*1*4096; 3.5) Input the image features in step 3.4) into SVM for training and multi-classification operations to obtain a dynamometer diagram classification model.
6. The method for detecting abnormal pumping wells based on production dynamic data and dynamometer diagram according to claim 5, characterized in that: In the step 3.5), the dynamometer diagram classification model includes a first category, a second category and a third category; the first category is category 1; the second category includes category 2, category 4, category, category 8 and category 9; the third category includes category 3, category 5, category 6 and category 10; the first category is normal, the second category is factors that have a greater impact on oil well production, and the third category is factors that have a smaller impact on oil well production.
7. The method for detecting abnormal pumping wells based on production dynamic data and dynamometer diagram according to claim 6, characterized in that: In the step 6), the dynamometer diagram classification model is the second largest category, and when the fluctuation coefficient δ is greater than 50%, it is judged to be a well with abnormal oil production.
8. The method for detecting abnormal pumping wells based on production dynamic data and dynamometer diagram according to claim 7, characterized in that: In the step 6), when the dynamometer diagram classification model is the first category and the third category and the fluctuation coefficient δ is greater than 30%, it is judged as a well with abnormal oil production.
