Data-driven downhole drilling condition recognition method and device

Abstract

The application provides a downhole drilling working condition recognition method and device based on data driving. The method comprises the following steps: before downhole drilling, the importance of different working condition characteristics is quantified by introducing contribution value, and the output of the working condition classifier of each working condition is subjected to explainability analysis; according to the contribution value of different working condition characteristics to each working condition, a judgment matrix of each working condition is constructed, thereby obtaining a weight vector of each working condition, and an objective weight distribution method driven by data is established; during downhole drilling, downhole measurement while drilling data is collected, and downhole drilling parameter data can be directly obtained; according to the weight vector of each working condition and the working condition index value of each working condition, the working condition of each window is decided through multi-classifier fusion; when a preset number of different working conditions is cumulatively output, the feature vector of each window and the working condition index threshold of each working condition are updated, and a double dynamic updating strategy is established to adapt to the complex and changeable drilling environment. The method provided by the application improves the accuracy and real-time performance of working condition recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring of drilling engineering, and particularly relates to a downhole drilling working condition recognition method and device based on data driving. BACKGROUND

[0002] Downhole drilling refers to a process of operating underground through drilling equipment in the process of resource exploitation such as oil and natural gas. The deep and ultra-deep drilling operation environment is complex, the edge computing resource is limited, and real-time and accurate recognition of the downhole drilling working condition is crucial to guarantee the drilling safety and improve the operation efficiency.

[0003] At present, the existing technology relies on the ground drilling parameters for recognition. In the recognition process, the traditional analytic hierarchy process is used to determine the weight of the drilling parameters, and the working condition is recognized according to the fixed threshold. However, the traditional analytic hierarchy process highly depends on the subjective judgment of experts in the weight determination process.

[0004] However, on the one hand, the ground drilling parameters can only provide indirect mapping of the downhole state, and are subject to signal propagation lag and insufficient timeliness. On the other hand, the traditional analytic hierarchy process highly depends on the subjective judgment of experts in the weight determination process, and cannot fully reflect the internal relationship between data, lacks quantification and interpretability of objective feature importance. In addition, the working condition is recognized according to the fixed threshold, which cannot guarantee the accuracy of the working condition recognition and adapt to the complex and changeable drilling environment. Therefore, the existing technology cannot guarantee the real-time and accuracy of the working condition recognition. SUMMARY

[0005] The downhole drilling working condition recognition method and device based on data driving are provided in the present application, so as to achieve the effect of improving the real-time and accuracy of the working condition recognition.

[0006] In a first aspect, the embodiments of the present application provide a drilling working condition recognition method based on data driving, applied to an edge computing unit of a drilling device, comprising: before drilling, performing the following steps: obtaining historical downhole measurement-while-drilling data, and extracting a plurality of working condition features from the historical downhole measurement-while-drilling data; training a deep learning model according to the plurality of working condition features to obtain a plurality of working condition classifiers; calculating contribution values of different working condition features to each working condition based on the working condition classifiers; constructing a judgment matrix of each working condition according to the contribution values of different working condition features to each working condition; obtaining a weight vector of each working condition according to the judgment matrix of each working condition; when drilling, performing the following steps: collecting downhole measurement-while-drilling data in real time; obtaining downhole measurement-while-drilling data of a plurality of windows from the downhole measurement-while-drilling data based on a sliding time window; obtaining a feature vector of each window according to the downhole measurement-while-drilling data of each window; determining working condition index values of each working condition according to the feature vector of each window and the weight vector of each working condition; determining working condition index minimum threshold values and working condition index maximum threshold values of each working condition by using a box plot method; if only one working condition index value is greater than or equal to the working condition index minimum threshold value corresponding to the current working condition and less than or equal to the working condition index maximum threshold value corresponding to the current working condition, the current working condition is determined as the working condition of each window; if a plurality of working condition index values are all greater than or equal to the working condition index minimum threshold values of each working condition and less than or equal to the working condition index maximum threshold values of each working condition, the working condition of each window is selected from the plurality of working conditions according to the working condition index values of the plurality of working conditions; outputting the working condition of each window; when a preset number of different working conditions are cumulatively output, updating the feature vector of each window by using a normalization method, and updating the working condition index threshold values of each working condition by using the box plot method and the exponential smoothing method; based on the updated feature vector of each window and the working condition index threshold values of each working condition, and the weight vector of each working condition, the working conditions of the remaining windows are continuously recognized.

[0007] In a possible implementation, the judgment matrix of each working condition includes a plurality of elements; accordingly, constructing the judgment matrix of each working condition according to the contribution values of different working condition features to each working condition comprises: calculating the ratio of the contribution values of any two working condition features to each working condition by using the analytic hierarchy process; determining the ratio of the contribution values of any two working condition features to each working condition as an element in the judgment matrix of each working condition, so as to complete the construction of the judgment matrix of each working condition.

[0008] In a possible implementation, the formula for calculating the contribution values of different working condition features to each working condition based on the working condition classifiers is:

[0009]

[0010] wherein, φ irepresents the contribution value of the i-th working condition feature to each working condition, F represents a set of multiple working condition features, |F| represents the number of multiple working condition features, |F|! represents the factorial of the number of multiple working condition features; F\{i} represents that the i-th working condition feature is not included in F, S represents a subset not containing the i-th working condition feature, |S| represents the number of working condition features in the subset not containing the i-th working condition feature, |S|! represents the factorial of the number of working condition features in the subset not containing the i-th working condition feature; f x (S∪{i}) represents the prediction value of the working condition classifier of each working condition when the i-th working condition feature is included, f x (S) represents the prediction value of the working condition classifier of each working condition when the i-th working condition feature is not included.

[0011] In a possible implementation, the working condition indicator minimum threshold value and the working condition indicator maximum threshold value of each working condition are determined by using a box plot method, and the formula is as follows:

[0012]

[0013] In the formula, Q1 represents the working condition indicator minimum threshold value when the working condition is k; represents the working condition indicator maximum threshold value when the working condition is k; Q1 k represents the lower quartile of the working condition indicator value when the working condition is k, Q3 k represents the upper quartile of the working condition indicator value when the working condition is k, and α represents an optimal width coefficient. k represents the interquartile range of the working condition indicator value when the working condition is k.

[0014] In a possible implementation, the working conditions of each window are selected from the multiple working conditions according to the working condition indicator values of the multiple working conditions, and the formula is as follows:

[0015]

[0016] In the formula, k * represents the selected working conditions of each window, V t represents the working condition indicator value of each working condition in the t-th window, and Q2 represents the median of the working condition indicator values of each working condition. represents the working condition selected from the multiple working conditions in the t-th window, in which the difference between the working condition indicator value of each working condition and the median of the working condition indicator values of each working condition is the smallest.

[0017] In a possible implementation, the feature vector of each window is obtained according to downhole measurement-while-drilling data of each window, including: obtaining feature parameters of multiple working condition features in the downhole measurement-while-drilling data of each window; obtaining parameter ranges corresponding to the feature parameters of each working condition feature in the downhole measurement-while-drilling data of each window; performing normalization on the feature parameters of each working condition feature according to the parameter ranges corresponding to the feature parameters of each working condition feature, to obtain normalized feature parameters of each working condition feature; and obtaining the feature vector of each window according to the normalized feature parameters of each working condition feature. Correspondingly, the feature vector of each window is updated by using the normalization method, including: obtaining the latest parameter ranges corresponding to the feature parameters of each working condition feature from the downhole measurement-while-drilling data of all windows corresponding to the output working condition; and performing normalization on the feature parameters of each working condition feature to complete the update of the feature vector of each window.

[0018] In a possible implementation, the working condition indicator threshold of each working condition is updated by using the box plot method and the exponential smoothing method, including: updating the working condition indicator threshold of each working condition by using the box plot method to obtain initial working condition indicator thresholds of each working condition; and performing smoothing processing on the initial working condition indicator thresholds of each working condition by using the exponential smoothing method to complete the update of the working condition indicator threshold of each working condition.

[0019] In a possible implementation, the working condition indicator value of each working condition is determined according to the feature vector of each window and the weight vector of each working condition, including: multiplying the feature vector of each window and the weight vector of each working condition to obtain the working condition indicator value of each working condition.

[0020] In a second aspect, an embodiment of the present application provides a data-driven downhole drilling working condition recognition device, applied to an edge computing unit of a drilling device, and including:

[0021] A first execution module is configured to perform the following steps before downhole drilling is performed.

[0022] The first execution module includes:

[0023] An extraction submodule is configured to obtain historical downhole measurement-while-drilling data and extract multiple working condition features from the historical downhole measurement-while-drilling data.

[0024] A training submodule is configured to train a deep learning model according to the multiple working condition features to obtain multiple working condition classifiers.

[0025] A calculation submodule is configured to calculate contribution values of different working condition features to each working condition based on the multiple working condition classifiers.

[0026] A construction submodule is configured to construct a judgment matrix of each working condition according to the contribution values of different working condition features to each working condition.

[0027] The first obtaining sub-module is configured to obtain a weight vector of each working condition according to a judgment matrix of each working condition.

[0028] The second executing module is configured to perform the following steps when drilling a well:

[0029] The second executing module includes:

[0030] The collecting sub-module is configured to collect real-time MWD data.

[0031] The second obtaining sub-module is configured to obtain MWD data of multiple windows from the MWD data based on a sliding time window.

[0032] The third obtaining sub-module is configured to obtain a feature vector of each window according to the MWD data of each window.

[0033] The first determining sub-module is configured to determine a working condition index value of each working condition according to the feature vector of each window and the weight vector of each working condition.

[0034] The second determining sub-module is configured to determine a working condition index minimum threshold value and a working condition index maximum threshold value of each working condition by using a box plot method.

[0035] The third determining sub-module is configured to determine the current working condition as the working condition of each window if only one working condition index value of the working conditions is greater than or equal to the working condition index minimum threshold value corresponding to the current working condition and less than or equal to the working condition index maximum threshold value corresponding to the current working condition.

[0036] The screening sub-module is configured to screen the working condition of each window from the multiple working conditions according to the working condition index values of the multiple working conditions if the working condition index values of the multiple working conditions are all greater than or equal to the working condition index minimum threshold value of each working condition and less than or equal to the working condition index maximum threshold value of each working condition.

[0037] The output sub-module is configured to output the working condition of each window.

[0038] The updating module is configured to update the feature vector of each window by using a normalization method and update the working condition index threshold value of each working condition by using a box plot method and an exponential smoothing method when a preset number of different working conditions are output cumulatively.

[0039] The identifying module is configured to continue to identify the working conditions of the remaining windows based on the updated feature vector of each window, the working condition index threshold value of each working condition, and the weight vector of each working condition.

[0040] Thirdly, embodiments of this application provide an edge computing unit for a drilling rig, comprising: a memory and a processor; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0041] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0042] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0043] The data-driven downhole drilling condition identification method and apparatus provided in this application extracts multiple condition features from historical downhole measurement-while-drilling data before downhole drilling. A deep learning model is trained based on these features to obtain multiple condition classifiers. Based on each classifier, the contribution value of different condition features to each condition is calculated, introducing a contribution value to quantify the importance of each condition feature. The interpretability of the outputs of the condition classifiers for each condition is analyzed. A judgment matrix for each condition is constructed based on the contribution values ​​of different condition features. A weight vector for each condition is obtained based on the judgment matrix, establishing a data-driven objective weight allocation method. The weight vectors of multiple conditions are deployed on the edge computing unit of the drilling equipment. During downhole drilling, downhole drilling parameters can be directly obtained from the downhole measurement-while-drilling data collected by the drilling equipment. Based on a sliding time window, downhole measurement-while-drilling data from multiple windows are acquired, and the condition index values ​​for each condition are determined according to the weight vectors of the multiple conditions. If only one working condition has a working condition index value greater than or equal to the minimum threshold and less than or equal to the maximum threshold, then the current working condition is identified as the working condition for each window. If multiple working conditions have working condition index values ​​all greater than or equal to the minimum threshold and less than or equal to the maximum threshold, then the working conditions for each window are selected from the multiple working conditions based on their index values, and a multi-classifier is used to fuse and determine the working condition for each window. Every time a preset number of different working conditions are output, the feature vectors of each window and the working condition index thresholds for each working condition are updated. Based on the updated feature vectors of each window, the working condition index thresholds for each working condition, and the weight vectors of each working condition, the working conditions for the remaining windows are identified. This dual dynamic update strategy adapts to the complex and ever-changing drilling environment. In summary, this improves the accuracy and real-time performance of working condition identification. BRIEF DESCRIPTION OF DRAWINGS

[0044] The accompanying drawings, which are incorporated herein and constitute part of this specification, illustrate embodiments consistent with the application and, together with the description, further serve to explain the principles of the application.

[0045] Figure 1 An architectural flow diagram of a data-driven downhole drilling working condition recognition method provided for embodiments of the application is shown in

[0046] Figure 2 A flow diagram of a data-driven downhole drilling working condition recognition method provided for embodiments of the application is shown in

[0047] Figure 3 A hierarchical structure diagram of working condition recognition provided for embodiments of the application is shown in

[0048] Figure 4 A division diagram of a sliding time window data provided for embodiments of the application is shown in

[0049] Figure 5 A structural diagram of a data-driven downhole drilling working condition recognition device provided for embodiments of the application is shown in

[0050] Figure 6 A structural diagram of an edge computing unit of a drilling equipment provided for embodiments of the application is shown in

[0051] The specific embodiments of the application have been shown by way of example in the above-described drawings and will be described in greater detail below. These drawings and description are not meant to restrict the scope of the inventive concept in any way but serve to explain the inventive concept to those skilled in the art by way of reference to specific embodiments. DETAILED DESCRIPTION

[0052] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The description of the exemplary embodiments is intended to apply to all alternative embodiments, as would be understood by one skilled in the art. The following exemplary embodiments are described herein with reference to the figures. Wherever possible, the same reference numbers are used in the drawings and the following description to illustrate the same or similar components. It should be noted that the functionality related to a process, engine, module, mechanism or component described herein can be split into multiple processes, engines, modules, mechanisms or components, or can be implemented by multiple processes, engines, modules, mechanisms or components.

[0053] Figure 1 An architectural flow diagram of a data-driven downhole drilling working condition recognition method provided for embodiments of the application is shown in Figure 1As shown, it comprises three main modules, namely data-driven working condition classifier training, weight vector construction of multiple working conditions, and downhole working condition identification. Among them, in the data-driven working condition classifier training, historical downhole measurement-while-drilling data is collected and preprocessed, the historical downhole measurement-while-drilling data after preprocessing is used to train the working condition classifier, and multiple working condition classifiers are obtained, each of which is a binary classifier; the weight vector of multiple working conditions is constructed, based on the multiple working condition classifiers, the contribution value of each working condition feature to each working condition is calculated, and then the judgment matrix of each working condition is constructed by combining the analytic hierarchy process, so as to construct the weight vector of each working condition; in the downhole working condition identification, downhole measurement-while-drilling data is collected in real time, combined with the constructed weight vector of multiple working conditions, working condition discrimination is performed, and the working condition identification result is output and uploaded to the ground. In this process, when dynamic updating is triggered, dynamic updating is performed to adapt to the changes of downhole formation conditions or well depth.

[0054] At present, the existing technology relies on the ground drilling parameters for identification. In the identification process, the traditional analytic hierarchy process is used to determine the weight of the drilling parameters, and the working condition is identified according to the fixed threshold. However, the traditional analytic hierarchy process highly depends on the subjective judgment of experts in the weight determination process. On the one hand, the ground drilling parameters can only provide indirect mapping of the downhole state, and are subject to signal propagation lag and lack of timeliness. On the other hand, the traditional analytic hierarchy process highly depends on the subjective judgment of experts in the weight determination process, and cannot fully reflect the internal relationship between data, lacks quantification and interpretability of objective feature importance; in addition, identifying the working condition according to the fixed threshold cannot guarantee the accuracy of working condition identification, and cannot adapt to the complex and changeable drilling environment. Therefore, the existing technology cannot guarantee the real-time and accuracy of working condition identification.

[0055] To solve the above technical problems, the present application proposes the following technical concept: considering that the ground drilling parameters can only provide indirect mapping of the downhole state and are subject to signal propagation lag and lack of timeliness, the inventors thought of collecting downhole measurement-while-drilling data in real time through the edge computing unit of the drilling equipment during downhole drilling, which can directly obtain downhole drilling parameter data; in view of the low identification accuracy of the existing technology, the inventors improve it from four aspects to improve the identification accuracy.

[0056] 1) Introducing contribution value to quantify the importance of different working condition features: extracting multiple working condition features from historical downhole measurement-while-drilling data; training a deep learning model according to the multiple working condition features to obtain multiple working condition classifiers; based on the working condition classifiers, calculating the contribution value of different working condition features to each working condition, and performing interpretability analysis on the output of the working condition classifier of each working condition;

[0057] 2) Establish a data-driven objective weight allocation method: according to the contribution value of different working conditions to each working condition, the judgment matrix of each working condition is constructed; according to the judgment matrix of each working condition, the weight vector of each working condition is obtained;

[0058] 3) Determine the working condition of each window through multi-classifier fusion decision: if there is only one working condition index value of the working condition, which is greater than or equal to the minimum threshold value of the working condition index corresponding to the current working condition, and less than or equal to the maximum threshold value of the working condition index corresponding to the current working condition, the current working condition is determined as the working condition of each window; if there are multiple working condition index values of the working condition, which are all greater than or equal to the minimum threshold value of the working condition index of each working condition, and less than or equal to the maximum threshold value of the working condition index of each working condition, the working condition of each window is selected from the multiple working conditions according to the working condition index value of the multiple working conditions;

[0059] 4) Establish a double dynamic updating strategy to adapt to the complex and changeable drilling environment: when a preset number of different working conditions are cumulatively output, update the feature vectors of each window and the working condition index thresholds of each working condition, and continue to identify the working conditions of the remaining windows based on the updated feature vectors of each window and the working condition index thresholds of each working condition, and the weight vectors of each working condition.

[0060] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific examples. The following specific examples can be combined with each other, and the same or similar concepts or processes may not be described again in some examples. The embodiments of the present application will be described below with reference to the accompanying drawings.

[0061] Figure 2 The flowchart of the downhole drilling working condition recognition method based on data driving provided by the embodiments of the present application is shown in FIG. 1, which includes the following steps. Figure 2

[0062] S201: Before downhole drilling, the following steps are performed.

[0063] Specifically, step S201 includes S2011-S2015:

[0064] S2011: Obtain historical downhole drilling measurement data, and extract a plurality of working condition features from the historical downhole drilling measurement data.

[0065] S2012: Train a deep learning model according to the plurality of working condition features to obtain a plurality of working condition classifiers.

[0066] Optionally, the deep model includes a random forest model, a support vector machine, and a convolutional neural network, etc.

[0067] ​In the embodiment, a deep learning model such as a random forest model, a support vector machine or a convolutional neural network is selected as a basic model for training to obtain the working condition classifiers of multiple working conditions.

[0068] In the embodiment, seven working conditions including static, tripping, pump starting and stopping, reaming, circulation, sliding drilling and compound drilling are included. The static working condition classifier, the tripping working condition classifier, the pump starting and stopping working condition classifier, the reaming working condition classifier, the circulation working condition classifier, the sliding drilling working condition classifier and the compound drilling working condition classifier are obtained by training.

[0069] In the embodiment, the working condition classifiers of each working condition are binary classification. For example, the static working condition classifier only outputs the static working condition and other working conditions.

[0070] S2013: Based on the working condition classifiers, the contribution values of different working condition features to each working condition are calculated.

[0071] In the embodiment, the contribution values of different working condition features to each working condition are calculated, and the formula is:

[0072]

[0073] In the formula, φ i represents the contribution value of the i th working condition feature to each working condition, F represents a set of multiple working condition features, |F| represents the number of multiple working condition features, |F|! represents the factorial of the number of multiple working condition features; F\{i} represents that the i th working condition feature is not included in F, S represents a subset not containing the i th working condition feature, |S| represents the number of working condition features in the subset not containing the i th working condition feature, |S|! represents the factorial of the number of working condition features in the subset not containing the i th working condition feature; f x (S∪{i}) represents the prediction value of the working condition classifier of each working condition when the i th working condition feature is included, f x (S) represents the prediction value of the working condition classifier of each working condition when the i th working condition feature is not included.

[0074] In the embodiment, the core of the calculation principle of the contribution value is to fairly allocate the contribution of each working condition feature to the output of the working condition classifier, to quantify the influence of each working condition feature on the output result, to quantify the importance of each working condition feature to the working condition recognition, to make the output result of the working condition classifier interpretable, and to provide a data-driven objective basis for weight allocation.

[0075] S2014: According to the contribution values of different working condition features to each working condition, a judgment matrix of each working condition is constructed.

[0076] Specifically, the analytic hierarchy process is adopted to calculate a ratio of contribution values of any two working condition characteristics to each working condition; the ratio of the contribution values of any two working condition characteristics to each working condition is determined as an element in a judgment matrix of each working condition, so as to complete construction of the judgment matrix of each working condition.

[0077] In the embodiment, the judgment matrix of each working condition is used as an objective basis for analytic hierarchy weight distribution. Figure 3 A hierarchical structure diagram of working condition recognition provided in the embodiment is shown in FIG. 1. Figure 3 As shown in FIG. 1, the hierarchical structure includes a target layer, a criterion layer and a scheme layer. The target layer performs working condition recognition; the criterion layer is downhole measurement-while-drilling data, which is a basis for working condition recognition; and the scheme layer is a working condition object set of a whole drilling process, i.e., working conditions.

[0078] In the embodiment, a composed element of the judgment matrix of each working condition is a ratio of contribution values of any two working condition characteristics to each working condition, i.e., a relative importance between any two working condition characteristics, which reflects the relative importance between the working condition characteristics. If there are 16 working condition characteristics, the judgment matrix of each working condition is a 16x16 matrix.

[0079] Specifically, a formula for obtaining any element in the judgment matrix of each working condition is as follows.

[0080]

[0081] In the formula, a ij is a ratio of a contribution value of an i th working condition characteristic to each working condition to a contribution value of a j th working condition characteristic to each working condition, indicates a relative importance of the i th working condition characteristic to the j th working condition characteristic, Φ i indicates the contribution value of the i th working condition characteristic to each working condition, and Φ j indicates the contribution value of the j th working condition characteristic to each working condition.

[0082] For example, when judging which working condition characteristic of the drilling pressure and the torque is more important, the traditional analytic hierarchy method depends on expert experience for determination, and there is a weight confusion caused by different expert experience differences and no unified standard. In the present application, the contribution value of the drilling pressure to each working condition and the contribution value of the torque to each working condition are calculated, and the ratio of the contribution value of the drilling pressure to each working condition to the contribution value of the torque to each working condition is used to judge which one of the drilling pressure and the torque is more important. The contribution value is introduced into the weight vector to quantify the importance of the working condition characteristic, and a data-driven objective weight distribution method is established.

[0083] S2015: Obtain a weight vector of each working condition according to a judgment matrix of each working condition.

[0084] In the embodiment, the weight vector of each working condition is obtained according to the judgment matrix of each working condition based on a characteristic equation and a normalization formula.

[0085] In the embodiment, the weight vector of each working condition is obtained based on the characteristic equation. At this time, the weight vector of each working condition is a weight vector which is not normalized and needs to be normalized to eliminate the dimensional difference between different working condition characteristics.

[0086] Specifically, the characteristic equation is:

[0087] A x W = λ max x W

[0088] In the formula, A represents the judgment matrix of each working condition, W represents the weight vector of each working condition, and λ max represents the maximum eigenvalue of the judgment matrix of each working condition.

[0089] Specifically, the formula for normalizing each weight in the weight vector of each working condition is:

[0090]

[0091] In the formula, represents the normalized weight of the i th working condition characteristic, W i represents the original data value of the weight of the i th working condition characteristic, and m represents the number of weights in the weight vector of each working condition.

[0092] Exemplarily, downhole measurement-while-drilling data of 11 wells is collected, and a total of 1361 hours of data is accumulated, covering seven typical working conditions, including static, tripping, pump starting and stopping, reaming, circulation, sliding drilling and compound drilling. The data preprocessing process uses a filtering algorithm for noise reduction, while coordinating the sampling rate of each sensor, identifying and processing abnormal values. The working condition feature extraction section extracts 16 working condition feature parameters in a 30-second sliding time window, providing sufficient working condition feature information for subsequent model training. When the base model is a random forest model, the data of the first 10 wells is divided into a training set and a validation set in a 7:3 ratio, and the 11th well is used as an independent test set for final performance evaluation. The random forest model is constructed in the base model training stage, and one-versus-all binary classification learning is performed for the seven working conditions. The model is explained using the Shapley additive interpretation method, the contribution value of each working condition feature to each working condition is calculated, and a judgment matrix based on the analytic hierarchy process is constructed based on the contribution values. Finally, the weight vector of each working condition is calculated to provide a weight vector for working condition recognition. In the field test of 11 wells for 38 hours, excellent performance is achieved. The normal working condition recognition accuracy is 95.5%, the F1 score of abnormal detection is 0.931, the single inference time is only 42 milliseconds, and the memory occupation is controlled within 48KB, fully meeting the performance requirements of real-time working condition recognition in the downhole environment. Further, when the base model is a random forest and the Shapley additive interpretation method is not used for explanatory analysis of the model, and the judgment matrix based on the analytic hierarchy process is not constructed based on the contribution value, the working condition recognition accuracy is also high, but due to high memory occupation and long calculation time, it cannot meet the requirements of the downhole environment. Therefore, the lightweight algorithm combining the contribution value and the analytic hierarchy process can reduce memory occupation and calculation time while meeting the working condition recognition accuracy.

[0093] In this embodiment, the process of constructing the weight vector of each working condition is regarded as the construction process of the weight model, and the weight model is deployed in the edge computing unit of the drilling equipment.

[0094] In this embodiment, the weight vector of each working condition is a one-dimensional matrix, representing the weight of each working condition feature.

[0095] S202: When drilling downhole, the following steps are performed;

[0096] Specifically, step S202 includes S2021-S2028:

[0097] S2021: Real-time acquisition of downhole measurement-while-drilling data.

[0098] The downhole measurement-while-drilling data is measurement data collected in real time by the measurement-while-drilling system during the drilling process.

[0099] In this embodiment, the downhole measurement-while-drilling data includes fluid pressure parameters, drill bit load parameters, drill bit motion parameters, and tubing vibration parameters.

[0100] S2022: Based on a sliding time window, obtain downhole measurement data from multiple windows of downhole measurement data.

[0101] In this embodiment, Figure 4 This is a schematic diagram illustrating the division of sliding time window data provided in an embodiment of this application. For example... Figure 4 As shown, an example of downhole measurement-while-drilling data with three windows is given. The first window acquires data from T0 to T... n Downhole measurement-while-drilling data at time T1; the second window acquired data from T1 to T2. n+1 Downhole measurement-while-drilling data at time T; the third window acquired data from T2 to T... n+2 Downhole measurement-while-drilling data at any given time.

[0102] S2023: Obtain the feature vector of each window based on the downhole measurement-while-drilling data of each window.

[0103] Specifically, step S2023 includes steps Sa to Sd:

[0104] Sa: Obtain characteristic parameters of multiple working conditions from the downhole measurement-while-drilling data in each window.

[0105] In this embodiment, feature parameters of multiple operating conditions are obtained from the downhole measurement-while-drilling data of each window, including pressure difference, internal pressure energy, external pressure energy, drilling pressure statistics, torque statistics, rotational speed statistics, and triaxial vibration characteristics. Optionally, the number of extracted operating condition features can be adjusted according to actual conditions.

[0106] Sb: Obtain the parameter range corresponding to the characteristic parameters of each working condition from the downhole measurement-while-drilling data in each window.

[0107] In this embodiment, the downhole measurement-while-drilling data of each window contains the parameter range of the characteristic parameters of each working condition, that is, the maximum and minimum values ​​of the characteristic parameters of each working condition.

[0108] Sc: Based on the parameter range corresponding to the characteristic parameters of each working condition, the characteristic parameters of each working condition are normalized to obtain the normalized characteristic parameters of each working condition.

[0109] For example, in the downhole measurement-while-drilling data in the first window, the maximum and minimum values ​​of the pressure difference are obtained; and the current pressure difference is normalized based on the maximum and minimum values ​​of the pressure difference.

[0110] In the embodiment, the purpose of normalization is to eliminate the dimensional difference of different working condition characteristics.

[0111] Sd: According to the characteristic parameters of the normalized working condition characteristics, the feature vectors of the windows are obtained.

[0112] S2024: According to the feature vectors of the windows and the weight vectors of the working conditions, the working condition index values of the working conditions are determined.

[0113] Specifically, the feature vectors of the windows are multiplied by the weight vectors of the working conditions to obtain the working condition index values of the working conditions.

[0114] In the embodiment, the formula for obtaining the working condition index values of the working conditions is:

[0115] V = W × X

[0116] In the formula, V represents the working condition index values of the working conditions, W represents the weight vectors of the working conditions, and X represents the feature vectors of the windows.

[0117] In the embodiment, after step S2051, the working condition index values of the seven types of working conditions are obtained.

[0118] S2025: The box plot method is used to determine the working condition index minimum threshold and the working condition index maximum threshold of each working condition.

[0119] In the embodiment, the formula for determining the working condition index minimum threshold and the working condition index maximum threshold of each working condition by using the box plot method is:

[0120]

[0121] In the formula, represents the working condition index minimum threshold when the working condition is k; represents the working condition index maximum threshold when the working condition is k; Q1 k represents the lower quartile of the working condition index values when the working condition is k, Q3 k represents the upper quartile of the working condition index values when the working condition is k, α represents the optimal width coefficient, IQR k represents the interquartile range of the working condition index values when the working condition is k.

[0122] In the embodiment, IQR k = Q3 k - Q1 k .

[0123] S2026: If only one working condition index value is greater than or equal to the working condition index minimum threshold corresponding to the current working condition and less than or equal to the working condition index maximum threshold corresponding to the current working condition, the current working condition is determined as the working condition of the windows.

[0124] Exemplarily, if there is only the working condition index value of the working condition k, the following conditions are met:

[0125]

[0126] In the formula, indicates the working condition index value when the tth window working condition is k, indicates the working condition index minimum threshold value when the working condition is k, indicates the working condition index maximum threshold value when the working condition is k. If there is only the working condition index value of the working condition k, which is greater than or equal to the working condition index minimum threshold value corresponding to the working condition k, and less than or equal to the working condition index maximum threshold value corresponding to the working condition k, the working condition k is determined as the working condition of the downhole measurement while drilling data of the tth window.

[0127] S2027: If there are working condition index values of multiple working conditions, which are greater than or equal to the working condition index minimum threshold value of each working condition, and less than or equal to the working condition index maximum threshold value of each working condition, the working conditions of each window are selected from the multiple working conditions according to the working condition index values of the multiple working conditions.

[0128] Exemplarily, the working condition index value of the circulating working condition is greater than or equal to the working condition index minimum threshold value of the circulating working condition, and less than or equal to the working condition index maximum threshold value of the circulating working condition; the working condition index value of the sliding drilling working condition is greater than or equal to the working condition index minimum threshold value of the sliding drilling working condition, and less than or equal to the working condition index maximum threshold value of the sliding drilling working condition; the working condition index value of the compound drilling working condition is greater than or equal to the working condition index minimum threshold value of the compound drilling working condition, and less than or equal to the working condition index maximum threshold value of the compound drilling working condition. At this time, the circulating, sliding drilling and compound drilling three working conditions are greater than or equal to the working condition index minimum threshold value of each working condition, and less than or equal to the working condition index maximum threshold value of each working condition, so it is necessary to select a working condition from the three working conditions according to the working condition index values of the three working conditions, and determine the working condition as the downhole measurement while drilling data of the current window.

[0129] In the embodiment, the formula for selecting the working condition of each window from the multiple working conditions according to the working condition index values of the multiple working conditions is:

[0130]

[0131] In the formula, k * indicates the selected working condition of each window, V t indicates the working condition index value of each working condition of the tth window, Q2 indicates the median of the working condition index values of each working condition, indicates the working condition selected from the multiple working conditions in the tth window, whose difference between the working condition index value of each working condition and the median of the working condition index values of each working condition is the smallest.

[0132] Exemplarily, the difference between the condition index value of each of the three working conditions of the cyclic drilling, the sliding drilling and the compound drilling and the median of the condition index values of the three working conditions is calculated, and one working condition with the smallest difference is selected from the three differences.

[0133] Optionally, if the condition index values of the seven working conditions are all abnormal, that is, all less than the lowest threshold of the condition index of each working condition or all greater than the highest threshold of the condition index of each working condition, the abnormal working condition is determined, and the abnormal working condition and the abnormal data are immediately uploaded.

[0134] S2028: Output the working condition of each window.

[0135] In the embodiment, the downhole drilling working condition recognition method based on data driving is used for testing, seven working conditions and an abnormal working condition can be recognized, the average recognition accuracy of 95.5% and the F1 score of 0.954 are achieved on the seven normal working conditions, the abnormal detection F1 value reaches 0.931, the number of parameters of the downhole measurement while drilling data is only 160, the peak memory is 48 KB, the average time consumption of single recognition is 42 ms, the requirements of the downhole resource limited environment are met, and the efficiency of the downhole safety and decision is improved.

[0136] In the embodiment, the working condition recognition result visualization is supported, the working condition of the downhole measurement while drilling data of each window is output to the display terminal on the ground, and the seven working condition recognition results are dynamically displayed, for example, the current working condition is the compound drilling. When it is an abnormal working condition, the abnormal data is synchronously displayed.

[0137] S203: When a preset number of different working conditions are cumulatively output, the feature vectors of the windows are updated by using a normalization method, and the working condition index thresholds of the working conditions are updated by using a box plot method and an exponential smoothing method.

[0138] In the embodiment, in order to adapt to the changes of the formation conditions or the well depth, when a preset number of different working conditions are cumulatively output, the feature vectors of the windows and the working condition index thresholds of the working conditions are triggered to be updated. Meanwhile, a minimum update interval of 30 minutes is set to control the calculation cost.

[0139] In the embodiment, the preset number can be 4.

[0140] In the embodiment, the step of updating the feature vectors of the windows by using the normalization method is that the latest parameter range corresponding to the feature parameters of each working condition feature is obtained from the downhole measurement while drilling data of all the windows corresponding to the output working conditions, the feature parameters of each working condition feature are normalized according to the latest parameter range corresponding to the feature parameters of each working condition feature, and the updating of the feature vectors of the windows is completed.

[0141] In this embodiment, the steps of updating the threshold values ​​of the operating conditions for each operating condition using the box plot method and the exponential smoothing method are as follows: the box plot method is used to update the threshold values ​​of the operating conditions for each operating condition to obtain the initial threshold values ​​of the operating conditions for each operating condition; the exponential smoothing method is used to smooth the initial threshold values ​​of the operating conditions for each operating condition to complete the update of the threshold values ​​of the operating conditions for each operating condition.

[0142] For example, after outputting the operating conditions of the first 7 windows, 4 different operating conditions were identified, triggering the update of the feature vectors of each window and the threshold values ​​of the operating condition indicators for each operating condition. From the downhole measurement-while-drilling data of the first 7 windows, the latest parameter range corresponding to the feature parameters of each operating condition is obtained, that is, the latest maximum and minimum values ​​corresponding to the feature parameters of each operating condition. Because the operating conditions change with the formation conditions or well depth, the feature parameters of different operating conditions also change. It is necessary to normalize the feature parameters of each operating condition based on the latest maximum and minimum values ​​corresponding to the feature parameters of each operating condition to complete the update of the feature vectors of each window and avoid feature mapping distortion caused by the drift of operating condition feature parameters.

[0143] Specifically, the normalization formula is as follows:

[0144]

[0145] In the formula, x represents the normalized value of the feature parameter of the i-th working condition feature. i This represents the original data value of the feature parameter for the i-th operating condition. This represents the latest minimum value corresponding to the feature parameter of the i-th working condition feature. This represents the latest maximum value corresponding to the feature parameter of the i-th working condition feature.

[0146] In this embodiment, the process of updating the threshold values ​​of the operating conditions for each operating condition is as follows: based on the box plot method, the threshold values ​​of the operating conditions for each operating condition are updated to obtain the initial threshold values ​​of the operating conditions for each operating condition; the exponential smoothing method is used to smooth the initial threshold values ​​of the operating conditions for each operating condition to complete the update of the threshold values ​​of the operating conditions for each operating condition.

[0147] In this embodiment, based on the statistical principles of box plots, the threshold values ​​of the operating conditions for each operating condition are dynamically updated.

[0148] In this embodiment, Q1, Q2, and Q3 are updated using the following formula:

[0149] Q1 new =Q1 old +γ·(q sample -Q1 old )

[0150] Q2 new = Q2 old + γ·(q sample - Q2 old )

[0151] Q3 new = Q3 old + γ·(q sample - Q3 old )

[0152] In the formula, γ represents a learning rate coefficient, q sample represents a sample quartile of the measured historical data, q sample includes the upper quartile, the median and the lower quartile of the sample.

[0153] In the formula, Q1 new represents an updated lower quartile, Q1 old represents a lower quartile before updating, at this time, q sample represents the lower quartile of the sample.

[0154] In this embodiment, the updating of Q2 and Q3 is the same, and will not be repeated here. When updating Q2, q sample represents the median of the sample; when updating Q3, q sample represents the upper quartile of the sample. According to the updated Q1, Q2 and Q3, the formula for obtaining the working condition index threshold of each working condition is obtained according to the above embodiment, the working condition index threshold of each working condition, and η lower new and η upper new .

[0155] In this embodiment, in order to avoid the recognition fluctuation caused by the sudden change of the working condition index threshold, an exponential smoothing method is introduced to obtain the final updated working condition index threshold of each working condition. The formula is as follows:

[0156] η lower updated = λη lower new + (1-λ)η lower old

[0157] η upper updated = λη upper new + (1-λ)η upper old

[0158] In the formula, λ represents a smoothing factor, η lower oldη represents the lowest threshold of the working condition indicator of each working condition before updating lower new η represents the lowest threshold of the working condition indicator of each working condition after updating Q1, Q2 and Q3 lower updated η represents the lowest threshold of the working condition indicator of each working condition after updating by introducing the exponential smoothing method.

[0159] In the formula, η upper old η represents the highest threshold of the working condition indicator of each working condition before updating upper new η represents the highest threshold of the working condition indicator of each working condition after updating Q1, Q2 and Q3 upper updated η represents the highest threshold of the working condition indicator of each working condition after updating by introducing the exponential smoothing method.

[0160] In the embodiment, to ensure the reliability of updating, whether to update is determined by comparing the consistency rate of working condition recognition results before and after updating.

[0161] Specifically, the formula for obtaining the consistency rate of working condition recognition results before and after updating is:

[0162]

[0163] In the formula, C r represents the consistency rate, T valid represents the number of windows, t represents the tth window, represents the working condition before updating of the tth window, represents the working condition after updating of the tth window. The essence of the consistency rate is the proportion of the number of windows with the same working condition recognition results before and after updating.

[0164] In the embodiment, if C r > 85%, updating is performed, if 70% < C r ≤ 85%, partial updating is performed with a smaller smoothing factor, and if C r ≤ 70%, updating is abandoned.

[0165] Optionally, the influence brought by updating is monitored by calculating the working condition conversion frequency change rate within 30 minutes before and after updating, and the formula is as follows:

[0166]

[0167] In the formula, Δ freq represents the working condition conversion frequency change rate, f after represents the number of working condition switching times per unit time after updating, f before represents the number of working condition switching times per unit time before updating.

[0168] S204: Based on the updated feature vectors of each window and the working condition indicator threshold values of each working condition, and the weight vectors of each working condition, the working conditions of the remaining windows are continuously identified.

[0169] In the embodiment, based on the updated feature vectors of each window and the weight vectors of multiple working conditions, the working condition indicator values of each working condition are determined, and according to the working condition indicator values of each working condition and the updated working condition indicator threshold values of each working condition, the working conditions of the remaining windows of the downhole measurement while drilling data are continuously identified.

[0170] In summary, before drilling downhole, multiple working condition features are extracted from historical downhole measurement while drilling data; a deep learning model is trained according to the multiple working condition features to obtain multiple working condition classifiers; based on the working condition classifiers, the contribution values of different working condition features to each working condition are calculated, the importance of different working condition features is quantified by the contribution values, and the output of the working condition classifier of each working condition is analyzed for interpretability; according to the contribution values of different working condition features to each working condition, a judgment matrix of each working condition is constructed; according to the judgment matrix of each working condition, a weight vector of each working condition is obtained, and a data-driven objective weight allocation method is established, and the weight vector of each working condition is applied to downhole working condition identification. When drilling downhole, downhole measurement while drilling data is collected in real time, and downhole drilling parameter data can be directly obtained; based on a sliding time window, downhole measurement while drilling data of multiple windows is obtained from the downhole measurement while drilling data; according to the feature vectors of each window and the weight vectors of each working condition, the working condition indicator values of each working condition are determined. If there is only one working condition indicator value of a working condition, which is greater than or equal to the minimum threshold value of the working condition indicator corresponding to the current working condition and less than or equal to the maximum threshold value of the working condition indicator corresponding to the current working condition, the current working condition is determined as the working condition of each window; if there are multiple working condition indicator values, which are all greater than or equal to the minimum threshold value of the working condition indicator of each working condition and less than or equal to the maximum threshold value of the working condition indicator of each working condition, the working condition of each window is selected from the multiple working conditions according to the working condition indicator values of the multiple working conditions, and the working condition of each window is determined by multi-classifier fusion decision. When a preset number of different working conditions are output, the feature vectors of each window and the working condition indicator threshold values of each working condition are updated, based on the updated feature vectors of each window and the working condition indicator threshold values of each working condition, and the weight vectors of each working condition, the working conditions of the remaining windows are continuously identified, and a double dynamic update strategy is established to adapt to complex and variable drilling environments. Therefore, the real-time performance and accuracy of working condition identification are improved.

[0171] On the basis of the above-mentioned embodiments, in this embodiment, the case when the basic model is a convolutional neural network is introduced, which is described in detail as follows:

[0172] In this embodiment, the convolutional neural network is used for deep working condition feature extraction according to the strong correlation between the working condition characteristics in the complex formation environment, which can better capture the nonlinear relationship and time sequence dependence between the working condition characteristics.

[0173] In this embodiment, the network structure design adopts a classic convolutional neural network architecture, and the input layer receives a 16-dimensional feature vector. The first convolutional layer uses a 3x1 convolutional kernel, and the output channel number is 32. The second convolutional layer also uses a 3x1 convolutional kernel, and the output channel number is expanded to 64. The fully connected layer adopts a decreasing structure of 64→32→7, and the final output layer generates a probability distribution of 7 working conditions. In the training parameter setting, the learning rate is set to 0.001, the batch size is 64, the training rounds are 100 rounds, and the loss function adopts cross-entropy loss to adapt to the requirements of the multi-classification task.

[0174] In this embodiment, the DeepExplainer method is used for explainability analysis of the convolutional neural network. The explainer is initialized by calling shap.DeepExplainer(model, background_data), and then the contribution value of each working condition feature to each working condition is calculated using explainer.shap_values(test_data). Based on the contribution value, a judgment matrix based on analytic hierarchy process is constructed. The weight distribution shows obvious characteristics: the vibration-related features have a higher weight of 0.15-0.22, the pressure feature weight is at a medium level of 0.08-0.15, and the torque feature weight is moderate at 0.06-0.12. The recognition accuracy reaches 92.6%, and the F1 score of anomaly detection is 0.921. Although the computational complexity is comparable to that of the random forest model, the convolutional neural network shows better adaptability and robustness in the working condition environment with high noise and strong coupling.

[0175] In this embodiment, in order to evaluate the performance of the working condition classifier based on the introduction of the contribution value to quantify the importance of different working condition features, a plurality of weight working condition classifiers are constructed,

[0176] On the basis of the above embodiments, the adaptive time window mechanism is introduced, which is described in detail as follows:

[0177] In this embodiment, the working condition changes frequently and complex during the drilling of ultra-deep wells, and the traditional fixed time window is difficult to adapt to this change characteristic, so an adaptive time window mechanism is needed.

[0178] In this embodiment, the working condition stability is quantified by the stability index, and the calculation formula is:

[0179]

[0180] In the embodiment, the stability index is obtained by calculating the average of the absolute value change rate of the feature parameters of adjacent operating condition features in the window.

[0181] In the embodiment, a dynamic window adjustment strategy is established based on the stability index: for stable operating conditions such as static and circulation, the window length is set to 10-20 seconds; for transition operating conditions such as tripping in and starting and stopping the pump, the window length is 30-40 seconds; and for complex operating conditions such as drilling and reaming, a window length of 40-60 seconds is needed to fully capture the operating condition feature changes.

[0182] Among them, the short window feature, the window length is 10 seconds, is mainly used to capture instantaneous changes, including vibration peak value, pressure pulse and speed mutation and other rapid changes. The medium window feature, the window length is 30 seconds, focuses on capturing the features of the operation mode, covering statistical features, energy features and frequency domain features and other parameters describing the operation state. The long window feature, the window length is 60 seconds, focuses on capturing the trend features, including trend slope, periodicity analysis and steady state features and other parameters reflecting the long-term change law.

[0183] Optionally, the weight of each operating condition feature can be dynamically adjusted according to the depth and formation conditions. This adjustment mechanism can compensate for the influence of depth changes on the performance of operating condition features, and the adjustment formula is:

[0184]

[0185] In the formula, a is the depth correction coefficient, the value range is 0.1-0.3; depth represents the current depth, depth max represents the maximum depth, W i base represents the weight of the i th operating condition feature before adjustment, W i depth represents the weight of the i th operating condition feature after adjustment.

[0186] In the embodiment, in the test of the ultra-deep well section greater than 5000m, the operating condition recognition accuracy reaches 94.2%, and the response time of adaptive adjustment is controlled within 2 seconds. Compared with the fixed time window, the generalization performance in adapting to different depths is improved by 8.5%.

[0187] On the basis of the above embodiment, in the embodiment, the frequent switching between sliding drilling operating conditions and composite drilling operating conditions in the drilling process of the horizontal well section is introduced, and the details are as follows:

[0188] In the embodiment, the sliding drilling operating condition and the composite drilling operating condition frequently switch in the drilling process of the horizontal well section, and the two drilling modes have similarities in some operating condition features, which are easy to confuse.

[0189] In the embodiment, two levels of weights are used for the working condition recognition. The first level of weights is used to distinguish drilling and non-drilling states, wherein the mean weight of the drilling pressure is 0.35, the mean weight of the torque is 0.28, the mean weight of the rotating speed is 0.20, and the total weight of other working condition characteristics is 0.17. The second level of weights is used to distinguish sliding and compound drilling, wherein the rotating speed ratio feature obtains the highest weight of 0.45, the torque fluctuation feature weight is 0.30, and the drilling pressure response feature weight is 0.25.

[0190] Specifically, the formula for calculating the rotating speed ratio feature is as follows:

[0191] RPM = (Q x 60) / V m

[0192] R = RPM actual / RPM

[0193] In the formula, RPM represents the screw rotating speed threshold, Q represents the pump displacement, V represents the screw motor displacement. m Based on the screw rotating speed threshold, a special drilling mode feature is constructed. In the formula, RPM represents the actually collected screw rotating speed, and R represents the rotating speed ratio feature. The sliding and compound drilling states can be effectively distinguished. actual

[0194] Specifically, the formula for calculating the torque fluctuation feature is as follows:

[0195]

[0196] In the formula, T represents the torque fluctuation feature, σ represents the torque standard deviation, and μ represents the torque mean value. f t t The torque fluctuation feature reflects the torque change characteristics in different drilling modes.

[0197] Specifically, the formula for calculating the drilling pressure response feature is as follows:

[0198]

[0199] In the formula, W represents the drilling pressure response feature, and WOB represents the drilling pressure. r

[0200] In the embodiment, a recognition confidence evaluation mechanism is introduced, and the formula for calculating the confidence is as follows:

[0201] Confidence = 1 - |V slide -V rotary | / (V slide +V rotary )

[0202] In the formula, Confidence represents the confidence, V represents the actual working condition feature, and V represents the ideal working condition feature. slide ​​​​An index value representing the sliding drilling mode, V rotary An index value representing the compound drilling mode. When the confidence is lower than 0.7, the current state is marked as mode transition, avoiding misjudgment during the transition process.

[0203] In this embodiment, in the test of the horizontal well section, the sliding drilling mode recognition accuracy reaches 91.3%, the compound drilling mode recognition accuracy is 96.1%, the mode transition detection success rate is 88.7%, and the misjudgment rate is reduced by 15.2%.

[0204] On the basis of the above embodiment, in this embodiment, the geological conditions of different oil wells are significantly different, and it is necessary to update the weight vector of each mode through multi-well collaborative learning.

[0205] In this embodiment, the geological similarity evaluation is calculated by using a Gaussian kernel function:

[0206]

[0207] In the formula, G1 and G2 represent the geological parameter vectors of two wells respectively, and σ represents a bandwidth parameter.

[0208] In this embodiment, the similar well screening strategy requires that the candidate well meets the following conditions: the difference in formation dip angle is less than 15 degrees, the difference in formation pressure coefficient is less than 0.1, and the difference in drilling fluid density is less than 0.05 g / cm 3 Only wells that meet these conditions can participate in the collaborative learning process.

[0209] Specifically, the local contribution value of each mode feature in each well to each mode is calculated, the global weight fusion process adopts a weighted average method, the global contribution value of each mode feature to each mode is obtained, and thus the global weight vector of each mode is obtained.

[0210] In this embodiment, in order to protect data privacy, only the contribution value and weight vector information between wells are shared, and no original drilling data is transmitted.

[0211] Optionally, when there is new downhole measurement-while-drilling data of a well, an incremental learning mechanism is adopted to gradually integrate the new downhole measurement-while-drilling data. The formula for updating the weight vector of each mode is:

[0212] W new = βW global + (1-β)W local

[0213] In the formula, W new represents the updated weight vector of each mode, W global represents the global weight vector of each mode before the new downhole measurement-while-drilling data is integrated, and W localThe weight vector of each working condition in the new well is represented by β, and the global weight retention coefficient is represented by β, which is in the range of 0.6-0.8. This mechanism not only maintains the global stability, but also adapts to the special conditions of the new well.

[0214] In this embodiment, the multi-well collaborative learning test achieves an accuracy of 87.3% in the first identification of the new well, which is significantly higher than the single-well test of 73.1%. The cross-well generalization performance is improved by 23.4%, and the weight model convergence time is reduced by 65%. The collaborative learning supports 3-15 wells and has good scalability.

[0215] On the basis of the above embodiment, in this embodiment, some old measurement-while-drilling systems have extremely limited computing power, and a more lightweight identification scheme needs to be developed to minimize computing and storage requirements while ensuring identification accuracy. Details are as follows:

[0216] In this embodiment, the importance of the contribution value of each working condition is sorted according to the characteristics of each working condition, and the top 8 working condition characteristics are retained, including the mean and variance of the drilling pressure, the mean and variance of the torque, the mean and variance of the rotation speed, and the X-axis and Y-axis vibration variance. To further reduce memory usage, the system converts floating-point working condition characteristics to 8-bit integer representation, significantly reducing storage requirements while maintaining sufficient accuracy.

[0217] In this embodiment, the weight vector of each working condition is calculated and stored as a lookup table in advance to avoid complex calculations during real-time inference and identification. The entire calculation process is converted to integer operations, eliminating the overhead of floating-point multiplication operations and significantly improving computational efficiency.

[0218] In this embodiment, a fixed threshold is used instead of a dynamic update mechanism, and the fixed upper and lower threshold values for each type of working condition are determined based on historical data. The decision logic uses simple comparison operations to avoid complex statistical calculations, further reducing computational complexity.

[0219] In this embodiment, it is suitable for ARM Cortex-M0 level processor platforms and can run stably in a random access memory system of less than 128KB, meeting the requirements of highly real-time application scenarios.

[0220] In this embodiment, the number of parameters is reduced to 64, which is 60% less than the complete version. The memory usage is controlled within 20KB, which is reduced by 58%. The single calculation time is shortened to 10 milliseconds, which improves the efficiency by 64%. Under these optimization conditions, the identification accuracy can still reach 89.2%, which is only 6.3% lower than the complete version.

[0221] Figure 5 The structure diagram of the downhole drilling working condition identification device based on data driving provided by the embodiments of the present application is as follows:Figure 5 As shown, the data-driven downhole drilling working condition recognition device provided by the embodiment comprises a first execution module 501, a second execution module 502, an update module 503, and a recognition module 504. The first execution module 501 comprises an extraction submodule 5011, a training submodule 5012, a calculation submodule 5013, a construction submodule 5014, and a first acquisition submodule 5015. The second execution module 502 comprises an acquisition submodule 5021, a second acquisition submodule 5022, a third acquisition submodule 5023, a first determination submodule 5024, a second determination submodule 5025, a third determination submodule 5026, a screening submodule 5027, and an output submodule 5028.

[0222] The first execution module 501 is configured to perform the following steps before downhole drilling:

[0223] The first execution module 501 comprises:

[0224] The extraction submodule 5011 is configured to acquire historical downhole measurement-while-drilling data and extract a plurality of working condition features from the historical downhole measurement-while-drilling data.

[0225] The training submodule 5012 is configured to train a deep learning model according to the plurality of working condition features to obtain a plurality of working condition classifiers.

[0226] The calculation submodule 5013 is configured to calculate a contribution value of each working condition feature to each working condition based on each working condition classifier.

[0227] The construction submodule 5014 is configured to construct a judgment matrix of each working condition according to the contribution value of each working condition feature to each working condition.

[0228] The first acquisition submodule 5015 is configured to acquire a weight vector of each working condition according to the judgment matrix of each working condition.

[0229] The second execution module 502 is configured to perform the following steps during downhole drilling:

[0230] The second execution module 502 comprises:

[0231] The acquisition submodule 5021 is configured to acquire downhole measurement-while-drilling data in real time.

[0232] The second acquisition submodule 5022 is configured to acquire downhole measurement-while-drilling data of a plurality of windows from the downhole measurement-while-drilling data based on a sliding time window.

[0233] The third acquisition submodule 5023 is configured to acquire a feature vector of each window according to the downhole measurement-while-drilling data of each window.

[0234] The first determining sub-module 5024 is configured to determine the working condition index value of each working condition according to the feature vector of each window and the weight vector of each working condition.

[0235] The second determining sub-module 5025 is configured to determine the working condition index minimum threshold value and the working condition index maximum threshold value of each working condition by using the box plot method.

[0236] The third determining sub-module 5026 is configured to, if only one working condition index value of the working condition is greater than or equal to the working condition index minimum threshold value corresponding to the current working condition and less than or equal to the working condition index maximum threshold value corresponding to the current working condition, determine the current working condition as the working condition of each window.

[0237] The screening sub-module 5027 is configured to, if there are multiple working condition index values of the working condition, all of which are greater than or equal to the working condition index minimum threshold value of each working condition and less than or equal to the working condition index maximum threshold value of each working condition, screen the working condition of each window from the multiple working conditions according to the working condition index values of the multiple working conditions.

[0238] The output sub-module 5028 is configured to output the working condition of each window.

[0239] The updating module 503 is configured to, when a preset number of different working conditions are output cumulatively, update the feature vector of each window by using the normalization method, and update the working condition index threshold value of each working condition by using the box plot method and the exponential smoothing method.

[0240] The identification module 504 is configured to continue to identify the working condition of each window based on the updated feature vector of each window and the working condition index threshold value of each working condition, and the weight vector of each working condition.

[0241] In a possible implementation, the judgment matrix of each working condition includes a plurality of elements. Correspondingly, the constructing sub-module 5014 is specifically configured to: calculate the ratio of the contribution values of any two working condition features to each working condition by using the analytic hierarchy process; and determine the ratio of the contribution values of any two working condition features to each working condition as an element in the judgment matrix of each working condition, so as to complete the construction of the judgment matrix of each working condition.

[0242] In a possible implementation, the formula of the calculating sub-module 5013 is:

[0243]

[0244] In the formula, Φ irepresents the contribution value of the i-th working condition feature to each working condition, F represents a set of multiple working condition features, |F| represents the number of multiple working condition features, |F|! represents the factorial of the number of multiple working condition features; F\{i} represents that the i-th working condition feature is not included in F, S represents a subset not containing the i-th working condition feature, |S| represents the number of working condition features in the subset not containing the i-th working condition feature, |S|! represents the factorial of the number of working condition features in the subset not containing the i-th working condition feature; f x (S∪{i}) represents the prediction value of the working condition classifier of each working condition when the i-th working condition feature is included, f x (S) represents the prediction value of the working condition classifier of each working condition when the i-th working condition feature is not included.

[0245] In a possible implementation, the formula of the second determination sub-module 5025 is:

[0246]

[0247] In the formula, k represents the minimum threshold of the working condition index when the working condition is k; represents the maximum threshold of the working condition index when the working condition is k; Q1 k represents the lower quartile of the working condition index value when the working condition is k, Q3 k represents the upper quartile of the working condition index value when the working condition is k, and α represents an optimal width coefficient; IQR k represents the interquartile range of the working condition index value when the working condition is k.

[0248] In a possible implementation, the formula of the screening sub-module 5027 is:

[0249]

[0250] In the formula, k * represents the working condition of each window screened out, V t represents the working condition index value of each working condition of the t-th window, and Q2 represents the median of the working condition index value of each working condition, represents the working condition with the minimum difference between the working condition index value of each working condition and the median of the working condition index value of each working condition in the t-th window.

[0251] In a possible implementation, the third obtaining sub-module 5023 is specifically configured to: obtain the feature parameters of the multiple working condition features in the downhole measurement-while-drilling data of each window; obtain the parameter range corresponding to the feature parameters of each working condition feature in the downhole measurement-while-drilling data of each window; normalize the feature parameters of each working condition feature according to the parameter range corresponding to the feature parameters of each working condition feature, to obtain the normalized feature parameters of each working condition feature; and obtain the feature vector of each window according to the normalized feature parameters of each working condition feature. Correspondingly, the updating module 503 is specifically configured to: obtain the latest parameter range corresponding to the feature parameters of each working condition feature from the downhole measurement-while-drilling data of all the windows corresponding to the working condition in the output; and normalize the feature parameters of each working condition feature according to the latest parameter range corresponding to the feature parameters of each working condition feature, to complete the updating of the feature vector of each window.

[0252] In a possible implementation, the updating module 503 is further configured to: update the working condition index threshold of each working condition by using a box plot method, to obtain the initial working condition index threshold of each working condition; and perform smoothing processing on the initial working condition index threshold of each working condition by using an exponential smoothing method, to complete the updating of the working condition index threshold of each working condition.

[0253] In a possible implementation, the first determining sub-module 5024 is specifically configured to: multiply the feature vector of each window by the weight vector of each working condition, to obtain the working condition index value of each working condition.

[0254] Figure 6 A structural diagram of an edge computing unit of a drilling equipment is provided in the embodiments of the present application. As shown in the figure, the edge computing unit of the drilling equipment provided in the embodiments of the present application comprises at least one processor 601 and a memory 602. Optionally, the edge computing unit of the drilling equipment further comprises a communication component 603. The processor 601, the memory 602 and the communication component 603 are connected through a bus.

[0255] In the implementation process, the at least one processor 601 executes the computer execution instructions stored in the memory 602, so that the at least one processor 601 executes the method described above.

[0256] The specific implementation process of the processor 601 can refer to the method embodiments described above, which has similar implementation principles and technical effects, and will not be described here in detail.

[0257] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU) and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or can also be any conventional processor. The steps of the method disclosed in combination with the application can be directly embodied as hardware processor execution or combined with hardware and software modules in the processor for execution.

[0258] The memory can include a random access memory (RAM) and can also include a non-volatile memory (NVM), such as at least one disk memory.

[0259] The bus can be an industry standard architecture (ISA) bus, a peripheral component (PCI) bus, an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application does not limit to only one bus or one type of bus.

[0260] The present application also provides a computer program product comprising a computer program which, when executed by a processor, implements the above method.

[0261] The present application also provides a computer readable storage medium having computer execution instructions stored therein, wherein when a processor executes the computer execution instructions, the above method is implemented.

[0262] The above readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0263] An example readable storage medium is coupled to the processor such that the processor can read information from the readable storage medium and can write information to the readable storage medium. Of course, the readable storage medium can also be a part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in the device.

[0264] The division of units is only a logical functional division, and in actual implementation, there can be another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0265] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0266] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit.

[0267] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the present application that essentially contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0268] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The foregoing program can be stored in a computer readable storage medium. The program executes to perform the steps of the above-mentioned method embodiments; and the foregoing storage medium includes various media capable of storing program codes, such as ROM, RAM, magnetic disk, or optical disk.

[0269] Finally, it should be noted that other embodiments of the present application will readily occur to those skilled in the art upon consideration of the specification and practice of the present application disclosed herein. The present application is intended to include all such variations as fall within the general scope of the application, and includes the generic principles disclosed and the best mode known to the inventors to be currently practiced as well as variations thereof, without departing from the scope of the present application as defined by the claims. The specification and examples give the best application of the present application as currently known, and together with the description of the application serve to best illustrate the principles of the application. The scope of the application is expressly set forth in the claims.

Claims

1. A data-driven downhole drilling condition identification method, characterized in that, Edge computing units applied to drilling equipment include: Before conducting downhole drilling, perform the following steps: Acquire historical downhole measurement-while-drilling data and extract multiple operating condition features from the historical downhole measurement-while-drilling data; The deep learning model is trained based on the multiple working condition features to obtain multiple working condition classifiers; Based on the classifiers for each working condition, the contribution value of different working condition characteristics to each working condition is calculated. Based on the contribution values ​​of the different working condition characteristics to each working condition, a judgment matrix for each working condition is constructed. Based on the judgment matrix of each working condition, obtain the weight vector of each working condition; When performing downhole drilling, perform the following steps: Real-time acquisition of downhole measurement data; Based on the sliding time window, downhole drilling measurement data for multiple windows is obtained from the downhole drilling measurement data; Based on the downhole measurement-while-drilling data of each window, obtain the feature vector of each window; Based on the feature vectors of each window and the weight vectors of each working condition, the working condition index values ​​for each working condition are determined. The minimum and maximum threshold values ​​of the operating conditions for each working condition are determined using the box plot method. If there is only one working condition whose working condition index value is greater than or equal to the minimum threshold of the working condition index corresponding to the current working condition, and less than or equal to the maximum threshold of the working condition index corresponding to the current working condition, then the current working condition is determined as the working condition of each window. If there are multiple working conditions whose working condition index values ​​are all greater than or equal to the minimum threshold of each working condition and less than or equal to the maximum threshold of each working condition, then the working conditions of each window are selected from the multiple working conditions based on the working condition index values ​​of the multiple working conditions. Output the operating status of each window; When a preset number of different working conditions are output, the feature vector of each window is updated using a normalization method, and the threshold values ​​of the working conditions are updated using a box plot method and an exponential smoothing method. Based on the updated feature vectors of each window and the threshold values ​​of each working condition, as well as the weight vectors of each working condition, the working conditions of the remaining windows are further identified.

2. The method according to claim 1, characterized in that, The judgment matrix for each working condition includes multiple elements; Accordingly, the step of constructing a judgment matrix for each working condition based on the contribution values ​​of the different working condition characteristics to each working condition includes: The Analytic Hierarchy Process (AHP) is used to calculate the ratio of the contribution values ​​of any two working condition characteristics to each working condition. The ratio of the contribution values ​​of any two working condition features to each working condition is determined as an element in the judgment matrix of each working condition, so as to complete the construction of the judgment matrix of each working condition.

3. The method according to claim 1, characterized in that, The formula for calculating the contribution value of different working condition features to each working condition based on each working condition classifier is as follows: In the formula, Φ i Let f represent the contribution of the i-th working condition feature to all working conditions; F represents the set of multiple working condition features; |F| represents the number of multiple working condition features; |F|! represents the factorial of the number of multiple working condition features; F{i} represents the subset of F that does not contain the i-th working condition feature; S represents the subset that does not contain the i-th working condition feature; |S| represents the number of working condition features in the subset that does not contain the i-th working condition feature; |S|! represents the factorial of the number of working condition features in the subset that does not contain the i-th working condition feature; f x (S∪{i}) represents the predicted value of the work condition classifier for each work condition when the i-th work condition feature is included, f x (S) represents the predicted value of the condition classifier for each condition when the i-th condition feature is not included.

4. The method according to claim 1, characterized in that, The formula for determining the minimum and maximum threshold values ​​of the operating conditions for each working condition using the box plot method is as follows: In the formula, This represents the minimum threshold value of the operating condition index when the operating condition is k. This represents the highest threshold value for the operating condition index when the operating condition is k; Q1 k Q3 represents the lower quartile of the operating condition index value when the operating condition is k. k This represents the upper quartile of the performance index value when the performance condition is k, α represents the optimal width coefficient, and IQR. k This represents the interquartile difference of the operating condition index values ​​when the operating condition is k.

5. The method according to claim 1, characterized in that, The formula for selecting the working conditions for each window from the multiple working conditions based on the working condition index values ​​is as follows: In the formula, k * V represents the operating conditions of each of the selected windows. t Let Q1 represent the operating condition index values ​​for each operating condition in the t-th window, and Q2 represent the median of the operating condition index values ​​for each operating condition. This means that in the t-th window, the working condition with the smallest difference between the working condition index value and the median of the working condition index values ​​is selected from multiple working conditions.

6. The method according to claim 1, characterized in that, The step of obtaining the feature vector of each window based on the downhole measurement-while-drilling data of each window includes: From the downhole measurement-while-drilling data in each window, characteristic parameters of multiple operating conditions are obtained; In the downhole measurement-while-drilling data of each window, obtain the parameter range corresponding to the characteristic parameters of each working condition. Based on the parameter range corresponding to the feature parameters of each working condition, the feature parameters of each working condition are normalized to obtain the normalized feature parameters of each working condition. Based on the feature parameters of each normalized working condition, the feature vector of each window is obtained; Accordingly, updating the feature vectors of each window using a normalization method includes: From the downhole measurement data of all windows corresponding to the output working conditions, obtain the latest parameter range corresponding to the characteristic parameters of each working condition. Based on the latest parameter range corresponding to the feature parameters of each working condition, the feature parameters of each working condition are normalized to complete the update of the feature vector of each window.

7. The method according to claim 1, characterized in that, The step of updating the threshold values ​​of the operating conditions for each operating condition using the box plot method and the exponential smoothing method includes: The box plot method is used to update the threshold values ​​of the operating conditions for each operating condition, thereby obtaining the initial threshold values ​​of the operating conditions for each operating condition. An exponential smoothing method is used to smooth the initial operating condition index thresholds for each operating condition, thereby updating the operating condition index thresholds for each operating condition.

8. The method according to claim 1, characterized in that, The step of determining the operating condition index value for each operating condition based on the feature vector of each window and the weight vector of each operating condition includes: The feature vector of each window is multiplied by the weight vector of each working condition to obtain the working condition index value of each working condition.

9. A data-driven downhole drilling condition identification device, characterized in that, Edge computing units applied to drilling equipment include: The first execution module is used to perform the following steps before downhole drilling; The first execution module includes: The extraction submodule is used to acquire historical downhole measurement-while-drilling data and extract multiple operating condition features from the historical downhole measurement-while-drilling data. The training submodule is used to train the deep learning model based on the multiple working condition features to obtain multiple working condition classifiers. The calculation submodule is used to calculate the contribution value of different working condition features to each working condition based on the classifiers for each working condition. A submodule is constructed to build a judgment matrix for each working condition based on the contribution values ​​of the different working condition characteristics to each working condition. The first acquisition submodule is used to acquire the weight vector of each working condition based on the judgment matrix of each working condition; The second execution module is used to perform the following steps during downhole drilling; The second execution module includes: The acquisition submodule is used to acquire downhole measurement-while-drilling data in real time; The second acquisition submodule is used to acquire downhole measurement data from the downhole measurement data of multiple windows based on the sliding time window; The third acquisition submodule is used to acquire the feature vector of each window based on the downhole measurement while drilling data of each window; The first determining submodule is used to determine the working condition index value of each working condition based on the feature vector of each window and the weight vector of each working condition. The second determination submodule is used to determine the minimum threshold and maximum threshold of the operating condition index for each operating condition using the box plot method. The third determination submodule is used to determine the current working condition as the working condition of each window if there is only one working condition index value that is greater than or equal to the minimum threshold of the working condition index corresponding to the current working condition and less than or equal to the maximum threshold of the working condition index corresponding to the current working condition. The filtering submodule is used to filter out the working conditions of each window from the multiple working conditions if there are multiple working condition index values ​​that are all greater than or equal to the minimum threshold of the working condition index of each working condition and less than or equal to the maximum threshold of the working condition index of each working condition. The output submodule is used to output the operating status of each window; The update module is used to update the feature vector of each window using a normalization method and update the threshold of the working condition index of each working condition using a box plot method and an exponential smoothing method when a preset number of different working conditions are output. The identification module is used to continue identifying the working conditions of the remaining windows based on the updated feature vectors of each window, the threshold values ​​of each working condition index, and the weight vectors of each working condition.

10. An edge computing unit for drilling equipment, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-8.

Images