Well logging abnormity identification and early warning training system and use method thereof

Abstract

The invention discloses a well logging abnormity identification and early warning training system and a use method thereof, and relates to the technical field of well logging professional technical training systems and methods, and the system comprises a well logging real-time database, a well logging abnormity prediction knowledge base, a well control learning unit and an early warning learning unit. The early warning learning unit comprises an abnormal curve simulation module, an extraction module, a labeling module, a student forecast filling module, an answer forecast generation module, an abnormal forecast comparison module and a forecast result analysis module. The real-time logging abnormal data is combined online to train logging abnormal identification and report, well control knowledge and other contents, and meanwhile, the learning result of a trainee can be assessed and evaluated, so that the training efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mud logging professional technical training systems and methods, and particularly relates to a mud logging anomaly identification and early warning training system and its usage method. Background Art

[0002] Abnormal prediction is an important part of mud logging work, which is related to the control of drilling safety and drilling costs, and there is also a requirement that the abnormal prediction rate reaches 100% in the assessment of mud logging quality. However, due to the lack of experience of on-site staff and the uneven business levels, it is often impossible to accurately judge the abnormalities of mud logging parameters in a timely manner, which easily leads to a decrease in the abnormal prediction rate and accuracy; the existing mud logging prediction training is mostly limited to the PPT teaching material mode and cannot combine the dynamic change characteristics of mud logging parameter curves, making it difficult to give intuitive impressions to the trained employees. This easily causes problems such as a single training method and limited training content, and it is difficult to achieve the expected training effect. The comprehensiveness and timeliness of the existing mud logging abnormal prediction training need to be improved. The training content received by each employee is not comprehensive enough and not systematic enough, and in the face of new mud logging abnormalities generated under new technologies and new processes, it is impossible to effectively train all employees in a timely manner.

[0003] For example, a patent document with the title of Mud Logging Technology Training Simulation Client System and the publication number of CN103854530A, which was published on June 11, 2014, records a mud logging technology training simulation client system, including a teaching demonstration module, an interactive practice module, a skills assessment module, a user information module, and a database module; the teaching demonstration module is used to display the name of an object when the mouse hovers over an object in the virtual digital well site and pop up the corresponding course content when double-clicking on an object in the digital well site scene. When the user selects any course, the corresponding teaching demonstration screen will appear; the interactive connection module is used for the user to select the tools or accessories to be used in the toolbox according to the corresponding business logic and place them in the current practice scene and place them at the position where the tool is to be used or the position where the accessory is to be installed. At the same time, when the placement is correct, the corresponding action will be automatically executed, and when the placement is incorrect, an error prompt will be given; the skills assessment module is used to give corresponding scores to the user's operations after the user conducts drills according to certain assessment content. Another example is a patent document with the title of Comprehensive Mud Logging Simulation System and the publication number of CN104680878A, which was published on June 3, 2015, records a comprehensive mud logging simulation system, which includes a teacher console, a sensor detection system, a sand table model, a power control system, and a cabinet. The teacher console includes a main control machine and a graphics machine. The cabinet includes a chromatograph, CO 2Analyzer, hydrogen generator, collection computer, chromatographic computer and client group, the sensor detection system includes multiple sensors, multiple sensors are gathered in their own junction boxes, connected to the main control machine via sensor signal bus cables, the main control machine is equipped with various virtual simulation program modules; the graphic machine is equipped with a virtual presentation module and other existing technologies all have the above defects.

[0004] Based on demand survey and business analysis, it is found that it is very necessary to provide mud logging forecast training for mud logging operators, and a more efficient, fast and flexible training method is needed: establish a mud logging anomaly forecast knowledge base, develop a mud logging forecast training system, and continuously update the mud logging forecast knowledge base to achieve employee training and practice, thereby improving employees' abnormal forecast capabilities. SUMMARY OF THE INVENTION

[0005] The purpose of the present invention is to provide a mud logging anomaly identification and early warning training system and its use method, which can complete the online scenario simulation of the entire process of mud logging field anomaly reporting, so as to realize online training of mud logging anomaly identification and reporting, well control knowledge, etc. in combination with real-time mud logging anomaly data, thereby improving the training efficiency, and at the same time, it can also realize the assessment and evaluation of the learning results of the trainees.

[0006] The present invention is achieved through the following technical solutions:

[0007] The logging anomaly identification and early warning training system includes a real-time logging database, a logging anomaly prediction knowledge base, a well control learning unit and an early warning learning unit.

[0008] The early warning learning unit includes an abnormal curve simulation module, an extraction module, a marking module, a student forecast filling module, an answer forecast generation module, an abnormal forecast comparison module and a forecast result analysis module;

[0009] The abnormal curve simulation module is connected to the logging anomaly prediction knowledge base. The function of the abnormal curve simulation module is to simulate and generate corresponding logging curves according to different types of logging anomaly cases;

[0010] The well control learning unit includes a well control knowledge simulation module and a result scoring module;

[0011] The real-time logging database is used to obtain logging data at the logging operation site;

[0012] The logging anomaly prediction database is connected to the logging real-time database. The logging anomaly prediction database standardizes the logging data into various logging anomaly parameters and divides them into different types of logging anomaly cases for storage;

[0013] The extraction module is respectively connected to the mud logging anomaly prediction knowledge base and the anomaly curve simulation module. The function of the extraction module is to randomly extract mud logging anomaly cases from the mud logging anomaly prediction knowledge base and replay the mud logging curves corresponding to the mud logging anomaly cases.

[0014] The function of the annotation module is to be able to annotate the abnormal data points in the mud logging curve to generate annotation information.

[0015] The student prediction form filling module is used to generate a student prediction form.

[0016] The answer prediction generation module is connected to the extraction module. The answer prediction generation module is used to automatically generate an answer prediction form in combination with mud logging parameters.

[0017] The anomaly prediction comparison module is respectively connected to the student prediction form filling module and the answer prediction generation module. The anomaly prediction comparison module is used to compare the student prediction form with the answer prediction form and output the score of the student prediction result.

[0018] The prediction result analysis module is connected to the anomaly prediction comparison module. The prediction result analysis module is used to generate a prediction result analysis and modification suggestions for the student prediction result.

[0019] The well control knowledge simulation module is used to implement a well control knowledge simulation exam.

[0020] The result scoring module is connected to the well control knowledge simulation module. The result scoring module is used to complete the result scoring.

[0021] Furthermore, the annotation information generated in the annotation module includes the annotation curve type, annotation time, and annotation content.

[0022] Furthermore, both the answer prediction form generated by the answer prediction generation module and the student prediction form generated by the student prediction form filling module include the well number, date, well depth, anomaly type, anomaly start time, prediction time, prediction parameters, parameter change situation, analysis result prediction, and recommended treatment measures.

[0023] Furthermore, new mud logging data can be input into the mud logging anomaly prediction database.

[0024] Furthermore, the operating sites where the prediction cases are located in the mud logging anomaly prediction knowledge base include the Gaomo block, Weiyuan block, Changning block, Shuangyushi block, Luzhou block, Zigong block, Xiangguosi block, and each exploration well area; the categories of anomaly prediction cases in the mud logging anomaly prediction knowledge base include gas logging anomaly, gas invasion, overflow, drill pipe stabbing, lost circulation, water eye blockage, hydrogen sulfide anomaly, and stuck pipe.

[0025] A method for using a logging anomaly identification and warning training system, based on the aforementioned logging anomaly identification and warning training system, includes the following steps:

[0026] S1. For different logging anomaly events, collect logging data at the logging operation site;

[0027] S2. According to the parameter change characteristics in different types of logging anomaly events, standardize the logging data into various logging anomaly parameters;

[0028] S3. Establish a logging anomaly prediction database to store the logging anomaly parameters of different types of logging anomaly cases, and simulate and generate corresponding logging curves for each logging anomaly parameter;

[0029] S4. According to the logging anomaly cases selected by the trainees, automatically generate an answer prediction form, compare the trainee prediction form with the answer prediction form, and output the trainee prediction result score, prediction result analysis, and modification suggestions.

[0030] Further, in step S4, the trainee prediction result score is obtained by scoring the trainee prediction result according to the established comprehensive evaluation model. Different scoring models can be formulated for different contents on the prediction form, and the comprehensive evaluation model can be comprehensively formulated by combining each scoring model. The specific method is as follows:

[0031] S41. For the pure numerical answer content, adopt an error scoring strategy to select an error range between the trainee's answer and the standard answer. The smaller the error range, the higher the single-item score. Among them, the pure numerical answer content includes well depth, anomaly start time, prediction time, and parameter change situation;

[0032] S42. For the text answer content, the scoring of the recommended treatment measure answer item is to first train the text in the standard defense database using the Word2Vctor model, establish the mapping relationship between the text and the numerical vector. When scoring, convert the standard answer and the trainee's answer through the trained Word2Vctor model from the word vector to the numerical vector, calculate the cosine distance of the converted numerical vectors, and score the trainee's answer through the distance metric value;

[0033] S43. Conduct a comprehensive score based on the calculated scores of the numerical and text answers.

[0034] Further, in step S41, the scoring rules for the well depth, anomaly start time, prediction time, and parameter change situation are specifically as follows:

[0035] a). The specific scoring calculation rule for the well depth item satisfies the relationship of formula (1):

[0036]

[0037] In formula (1), S D represents the score of the trainee in the well depth answer item, dimensionless; D stu represents the answer filled in by the trainee in the well depth answer item, m; D sta represents the standard answer in the well depth answer item, m; D rang is the allowable error value in the well depth answer item, m;

[0038] b). For the specific scoring calculation rule of the abnormal start time item, it satisfies the following relationship of formula (2):

[0039] First, convert the time format to seconds. The calculation formula is as follows:

[0040] BT = H * 3600 + M * 60 + S (2),

[0041] In formula (2), BT represents the number of seconds of the abnormal start time, s; H is the value of the hour digit in the time format, dimensionless; M is the value of the minute digit in the time format, dimensionless; S is the value of the second digit in the time format, dimensionless;

[0042] Then, based on converting the time format to seconds, calculate the score of the abnormal start time item. The specific scoring rule satisfies the following relationship of formula (3):

[0043]

[0044] In formula (3), S BT represents the score of the trainee in the abnormal start time item, dimensionless; BT stu represents the answer filled in by the trainee in the abnormal start time answer item, s; BT sta represents the standard answer in the abnormal start time answer item, s; BT rang is the allowable error value in the abnormal start time answer item, s;

[0045] c). For the specific scoring rule of the forecast time item, it satisfies the following relationship formula (4):

[0046] First, convert the time format to seconds. The calculation formula is as follows:

[0047] WT = H * 3600 + M * 60 + S (4),

[0048] In formula (4), WT represents the number of seconds of the forecast time, s; H is the value of the hour digit in the time format; M is the value of the minute digit in the time format; S is the value of the second digit in the time format;

[0049] Then, based on converting the time format to seconds, calculate the score of the forecast time item. The specific scoring rule satisfies the following relationship of formula (5):

[0050]

[0051] In formula (5), S WT represents the score of the trainee in the forecast time item, dimensionless; WT stu represents the answer filled in by the trainee in the forecast time answer item, s; WT sta represents the standard answer in the forecast time answer item, s; WT rang is the allowable error value in the forecast time answer item, s;

[0052] d). The specific scoring rule for the parameter change item satisfies the following relationship in formula (6):

[0053]

[0054] In formula (6), represents the score of the trainee in the parameter change item, dimensionless; i represents the parameter number, and the value range is [1, M], where M represents the total number of changing parameters considered in a single forecast; P stu represents the answer filled in by the trainee in the parameter change item; P sta represents the standard answer in the parameter change item answer; P rang is the allowable error value in the parameter change item answer item.

[0055] Furthermore, in step S42, the method for obtaining the text answer content is as follows:

[0056] i). First, segment the answers of the established treatment measures in the standard answer library. When segmenting the text description type, irrelevant words and punctuation symbols with no relevant text semantics are not considered;

[0057] ii). Then, perform one-hot encoder encoding on all the words in the answer library. The encoding length is the total number of words in the answer library. On the basis of encoding, perform a word vector mapping containing semantic information, and use the word embedding vector after training with the word2vector model;

[0058] iii). The scoring rule for the proposed treatment measure answer item satisfies the following relationship in formula (7):

[0059]

[0060] In formula (7), S W is the score of the proposed treatment measure answer item, dimensionless; represents the value at the i-th position of the word embedding encoding of the answer filled in by the trainee in the proposed treatment measure answer item, dimensionless; The value at the i-th position of the word embedding encoding representing the standard answer in the recommended treatment measure answer item, dimensionless.

[0061] Furthermore, in step S43, on the basis of calculating scores for numerical and text answers, a comprehensive score is performed, and the scoring method satisfies the following relationships of formulas (8) and (9):

[0062]

[0063] w 1 +w 2 +w 3 +w 4 +w 5 =1 (9);

[0064] In formulas (8) and (9), S is the score of the trainee, dimensionless; w 1 is the score weight of the deep item answer, dimensionless; w 2 is the score weight of the abnormal start time item answer, dimensionless; w 3 is the score weight of the abnormal warning time item answer, dimensionless; w 4 is the score weight of the parameter change item answer, dimensionless; w 5 is the score weight of the recommended treatment measure item answer, dimensionless.

[0065] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0066] First, in the present invention, the establishment of the abnormal prediction knowledge base first faces the standard conversion and rule formulation of text abnormal report documents and related data. Various parameter adjustments are carried out for the diversification of data sources and case selection and warehousing. Recently, multiple manual actual comparisons with the documents have been carried out, and parameter adjustments and settings are combined with manual operation thinking to form the knowledge base case foundation. During the establishment process, an independent mud logging abnormal case recognition, learning model and process are formed. A complete result evaluation and measurement analysis model, as well as related multiple reinforcement learning strategy mechanisms, are constructed.

[0067] Second, in the present invention, the mud logging curve simulation forms an independent parameter model based on the analysis of a large number of similar cases, and realizes virtualizing a similar abnormal case according to the set type under specific working conditions. At present, there are some mud logging abnormal recognition technologies, but the simulation case learning technology is lacking. This technology will enrich the existing knowledge base and flexibly construct case parameters and trend states, which are more variable, diverse and random. Through long-term learning, the sensitivity of personnel to the recognition after the occurrence of mud logging abnormalities can be greatly improved.

[0068] III. In the present invention, the logging data playback gradually displays real or simulated case data in a form of real-time refresh. The data playback of this solution can achieve the adjustment of different rates at any time, greatly enhancing the controllability of the case, and realizing that the playback process can be manually intervened for pre-pause and start while adjusting the rate. Technically, the svg vector mode is adopted in the curve display, which can achieve a higher-definition curve form and vector compatibility. In terms of real-time data extraction, the moving scale form is adopted to extract and display the data in real time through a floating layer. In terms of interface design, multiple floating layers can coexist and share, and each layer has the operation ability to realize data communication and control.

[0069] IV. In the present invention, based on the logging anomaly recognition and early warning training system, this solution has the following advantages: First, it can realize online training of logging anomaly recognition and reporting, well control knowledge, etc., improving the training efficiency; Second, through the development of this system, the prediction and reporting level of on-site logging personnel is further improved, and the logging well control safety awareness is enhanced; Third, new cases can be continuously supplemented, avoiding uneven training received by employees, and making the training more comprehensive and systematic. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a schematic structural diagram of the logging anomaly recognition and early warning training system in the present invention.

[0071] Figure 2 It is the overall function framework of the logging anomaly recognition and early warning training system in the present invention.

[0072] Figure 3 It is a schematic diagram of normal type selection.

[0073] Figure 4 It is the abnormal curve annotation a.

[0074] Figure 5 It is a schematic diagram of the trainee's prediction filling.

[0075] Figure 6 It is a screenshot of answer viewing.

[0076] Figure 7 It is a comparison chart of abnormal predictions.

[0077] Figure 8 It is a screenshot of result analysis.

[0078] Figure 9 It is a structural diagram of the word2vector model. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0079] The present invention will be further described in detail below in conjunction with the embodiments, but the embodiments of the present invention are not limited thereto.

[0080] Embodiment 1

[0081] In order to conduct more efficient, rapid, and flexible mud logging prediction training for mud logging operators, the present invention proposes a mud logging anomaly identification and early warning training system. This training system establishes a function for replaying mud logging curves of anomaly prediction cases; establishes an output function for trainees to identify anomalies, make annotations, capture images, give early warnings, fill out prediction forms, etc.; establishes an output function for trainees to completely describe oil, gas, water, and leakage displays; and compares and analyzes the information output by trainees with a standardized anomaly prediction knowledge base to evaluate the training results of trainees.

[0082] More specifically, referring to Figure 1 , the mud logging anomaly identification and early warning training system includes a mud logging real-time database, a mud logging anomaly prediction knowledge base, a well control learning unit, and an early warning learning unit.

[0083] Among them, the early warning learning unit includes an abnormal curve simulation module, an extraction module, an annotation module, a trainee prediction filling module, an answer prediction generation module, an abnormal prediction comparison module, and a prediction result analysis module; the well control learning unit includes a well control knowledge simulation module and a result scoring module.

[0084] The overall functional framework of this mud logging anomaly identification and early warning training system is as Figure 2 shown.

[0085] In this system, the mud logging real-time database is used to obtain mud logging data at the mud logging operation site.

[0086] The mud logging anomaly prediction database and the mud logging real-time database are configured to standardize the mud logging data into various mud logging anomaly parameters and store them by dividing them into different types of mud logging anomaly cases.

[0087] It should be noted that the mud logging anomaly prediction database should have the functions of continuous improvement and expansion, and the cases should cover gas logging anomalies, gas invasion, overflow, drill pipe stabbing, well leakage, water eye blockage, hydrogen sulfide anomalies, pipe sticking, etc. The operation sites where the prediction cases are located include geographical locations such as the Gaomo block, the Weiyuan block, the Changning block, the Shuangyushi block, the Luzhou block, the Zigong block, the Xiangguosi block, and various exploration well areas, etc., which have good universality.

[0088] The categories of anomaly prediction cases include gas logging anomalies, gas invasion, overflow, drill pipe stabbing, well leakage, water eye blockage, hydrogen sulfide anomalies, pipe sticking, etc. The current collected cases are as shown in Table 1 below.

[0089] Table 1: Statistical table of abnormal categories of collected cases

[0090]

[0091]

[0092] In this embodiment, the abnormal curve simulation module is connected to the mud logging abnormal prediction knowledge base. The abnormal curve simulation module is used to simulate and generate corresponding mud logging curves according to different types of mud logging abnormal cases.

[0093] In this embodiment, the extraction module is respectively connected to the mud logging abnormal prediction knowledge base and the abnormal curve simulation module. The extraction module is used to randomly extract mud logging abnormal cases from the mud logging abnormal prediction knowledge base and replay the mud logging curves corresponding to the mud logging abnormal cases. Refer to Figure 3 as shown Figure 3 is a mud logging abnormal case extracted according to the selected abnormal type.

[0094] In this embodiment, the annotation module is configured to be able to annotate abnormal data points in the mud logging curve to generate annotation information. Among them, the annotation information includes the annotation curve type, annotation time and annotation content. As Figure 4 shown is the annotation result of the abnormal curve.

[0095] In this embodiment, the trainee prediction filling module is used to generate a trainee prediction filling form. As Figure 5 shown is the page for trainee prediction filling.

[0096] In this embodiment, the answer prediction generation module is connected to the extraction module. The answer prediction generation module is used to automatically generate an answer prediction form in combination with mud logging parameters.

[0097] In this embodiment, both the trainee prediction filling form and the answer prediction form include well number, date, well depth, abnormal type, abnormal start time, prediction time, prediction parameters, parameter change situation, analysis result prediction and recommended treatment measures. As Figure 6 shown is the generated prediction answer.

[0098] In this embodiment, the abnormal prediction comparison module is respectively connected to the trainee prediction filling module and the answer prediction generation module. The abnormal prediction comparison module is used to compare the trainee prediction filling form with the answer prediction form and output the trainee prediction result score. As Figure 7 shown is the abnormal prediction comparison result.

[0099] In this embodiment, the prediction result analysis module is connected to the abnormal prediction comparison module. The prediction result analysis module is used to generate prediction result analysis and modification suggestions for the trainee prediction result. As Figure 8 shown is the result analysis result.

[0100] In this embodiment, the well control knowledge simulation module is used to implement a well control knowledge simulation exam.

[0101] In this embodiment, the result scoring module is connected to the well control knowledge simulation module. The result scoring module is used to complete the result scoring.

[0102] In summary, the logging anomaly identification and early warning training system in this embodiment can mainly achieve the following functions:

[0103] (1). Curve playback, performing curve playback on the time database;

[0104] (2). Curve annotation, annotating curve anomaly points;

[0105] (3). Curve management, increasing the data lines for observation by adding and deleting curves;

[0106] (4). Automatic screenshot capture, forming a curve graph of data in a specific time period and storing it;

[0107] (5). Output function of oil-gas-water leakage display, automatically converting parameters into a standard output format;

[0108] (6). Filling out the logging engineering anomaly report form to form a standardized forecast form;

[0109] (7). Learning evaluation, grading the answers filled in by trainees;

[0110] (8). Simulated curve, simulating an abnormal curve according to abnormal forecast cases.

[0111] Based on the above logging anomaly identification and early warning training system, the present invention also provides a usage method of the logging anomaly identification and early warning training system. The training method includes the following steps:

[0112] S1. For different logging anomaly events, collect logging data at the logging operation site.

[0113] S2. According to the parameter change characteristics in different types of logging anomaly events, standardize the logging data into various logging anomaly parameters. The logging anomaly forecast parameter reference formed after standardization is shown in Table 2, and Table 2 exemplifies the characteristics of some logging anomaly forecast parameters formed after standardization.

[0114] Table 2: Table of Logging Anomaly Forecast Parameter Characteristics (Partial)

[0115]

[0116] S3. Establish a logging anomaly forecast database to store the logging anomaly parameters of different types of logging anomaly cases, and simulate and generate corresponding logging curves for each logging anomaly parameter.

[0117] S4. According to the logging anomaly cases selected by trainees, automatically generate an answer forecast form, compare the trainee forecast filling form with the answer forecast form, and output the trainee forecast result score, forecast result analysis, and modification suggestions.

[0118] Among them, the score of the trainee's prediction result is obtained by grading the trainee's prediction result according to the established comprehensive evaluation model. Different scoring models can be formulated for different contents on the prediction form, and the comprehensive evaluation model can be comprehensively formulated by combining each scoring model. The specific implementation steps of the trainee's prediction result scoring are as follows:

[0119] S41. For the content of pure numerical answers, an error scoring strategy is adopted to select an error range between the trainee's answer and the standard answer. The smaller the error range, the higher the score for this item. Among them, the content of pure numerical answers includes well depth, abnormal start time, prediction time, parameter change situation, etc.

[0120] a). The specific scoring calculation rules for the well depth item are as follows:

[0121]

[0122] In formula (1), S D represents the score of the trainee in the well depth answer item, dimensionless; D stu represents the answer filled in by the trainee in the well depth answer item, m; D sta represents the standard answer in the well depth answer item, m; D rang is the allowable error value in the well depth answer item, m.

[0123] b). The specific scoring calculation rules for the abnormal start time item are as follows:

[0124] First, convert the time format to seconds. The calculation formula is as follows:

[0125] BT = H * 3600 + M * 60 + S (2),

[0126] In formula (2), BT represents the number of seconds of the abnormal start time, s; H is the value of the hour digit in the time format, dimensionless; M is the value of the minute digit in the time format, dimensionless; S is the value of the second digit in the time format, dimensionless.

[0127] Then, on the basis of converting the time format to seconds, calculate the score of the abnormal start time item. The specific scoring rules are as follows:

[0128]

[0129] In formula (3), S BT represents the score of the trainee in the abnormal start time item, dimensionless; BT stu represents the answer filled in by the trainee in the abnormal start time answer item, s; BT sta represents the standard answer in the abnormal start time answer item, s; BT rang is the allowable error value in the abnormal start time answer item, s.

[0130] c). The specific scoring rules for the forecast time item are as follows:

[0131] First, convert the time format to seconds. The calculation formula is as follows:

[0132] WT = H * 3600 + M * 60 + S (4),

[0133] In formula (4), WT represents the number of seconds of the forecast time, in s; H is the value of the hour digit in the time format; M is the value of the minute digit in the time format; S is the value of the second digit in the time format.

[0134] Then, based on converting the time format to seconds, calculate the score for the forecast time item. The specific scoring rules are as follows:

[0135]

[0136] In formula (5), S WT represents the score of the trainee in the forecast time item, dimensionless; WT stu represents the answer filled in by the trainee in the forecast time answer item, in s; WT sta represents the standard answer in the forecast time answer item, in s; WT rang is the allowable error value in the forecast time answer item, in s.

[0137] d). The specific scoring rules for the parameter change situation item are as follows:

[0138]

[0139] In formula (6), represents the score of the trainee in the parameter change item, dimensionless; i represents the parameter number, and the value range is [1, M], where M represents the total number of changing parameters considered in a single forecast; P stu represents the answer filled in by the trainee in the parameter change item, and the specific unit depends on the parameter type; P sta represents the standard answer in the parameter change item answer, and the specific unit depends on the parameter type; P rang is the allowable error value in the parameter change item answer, and the specific unit depends on the parameter type.

[0140] S42. For the text answer content, such as: The scoring of the recommended treatment measure answer item is to first train the text in the standard defense library with the Word2Vctor model, establish the mapping relationship between the text and the numerical vector. When scoring, convert the standard answer and the trainee's answer through the trained Word2Vctor model from the word vector to the numerical vector, calculate the cosine distance of the converted numerical vectors, and score the trainee's answer through the distance metric value.

[0141] The specific calculation process is as follows:

[0142] i). First, segment the answers of the established treatment measures in the standard answer library. For example, the specific segmentation steps are as follows:

[0143] [“Stop the pump and observe, plug the leak”]->[“Stop the pump”, “Observe”, “Plug the leak”].

[0144] When segmenting the text description type here, irrelevant words such as modal particles and stop words that have nothing to do with the text semantics and punctuation marks are not considered.

[0145] ii). Then, perform one-hot encoder encoding on all the words in the answer library, that is, each word has a unique encoding. The specific encoding process is as follows:

[0146] “Stop the pump”->[1,0,0,0,0,0,0,...,0];

[0147] “Observe”->[0,1,0,0,0,0,0,...,0];

[0148] “Plug the leak”->[0,0,1,0,0,0,0,...,0].

[0149] The length of the encoding is the total number of words in the answer library. On the basis of encoding, word vector mapping containing semantic information can be performed. The model adopted in this patent is the word2vector model, and the model structure diagram is as Figure 9 shown. In the model, the input vector x k (k = 1,2,3,…,V) is the one-hot vector encoding of a single word; V represents the total number of words in the answer library; h i is the value of the encoding layer, N is the length of the output word vector; y j (j = 1,2,3,…,V) is the word embedding vector after model training.

[0150] iii) The scoring rule for the answer items of the recommended treatment measures is as follows:

[0151]

[0152] In formula (7), S W is the score of the answer item of the recommended treatment measure, dimensionless; represents the value at the i-th position of the word embedding encoding of the answer filled in by the student in the answer item of the recommended treatment measure, dimensionless; represents the value at the i-th position of the word embedding encoding of the standard answer in the answer item of the recommended treatment measure, dimensionless.

[0153] S43. Based on calculating the scores of numerical and text answers, a comprehensive score is given, and the scoring rules are as follows:

[0154]

[0155] w 1 +w 2 +w 3 +w 4 +w 5 =1 (9);

[0156] In formulas (8) and (9), S is the score of the trainee, dimensionless; w 1 is the score weight of the depth item answer, dimensionless; w 2 is the score weight of the abnormal start time item answer, dimensionless; w 3 is the score weight of the abnormal warning time item answer, dimensionless; w 4 is the score weight of the parameter change item answer, dimensionless; w 5 is the score weight of the recommended treatment measure item answer. The specific values of the above weights are given by experts, dimensionless.

[0157] The above is only a preferred embodiment of the present invention, and does not impose any form of limitation on the present invention. Any simple modification or equivalent change made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. Mud logging anomaly identification and early warning training system, characterized by: Including real-time logging database, logging anomaly prediction knowledge base, well control learning unit and early warning learning unit, The early warning learning unit includes an abnormal curve simulation module, an extraction module, a marking module, a student forecast filling module, an answer forecast generation module, an abnormal forecast comparison module and a forecast result analysis module; The abnormal curve simulation module is connected to the logging anomaly prediction knowledge base, and the abnormal curve simulation module is used to simulate and generate corresponding logging curves according to different types of logging anomaly cases; The well control learning unit includes a well control knowledge simulation module and a result scoring module; The real-time logging database is used to obtain logging data at the logging operation site; The logging anomaly prediction database is connected to the logging real-time database, and the logging anomaly prediction database is used to standardize the logging data into various logging anomaly parameters, and divide them into different types of logging anomaly cases for storage; The extraction module is connected to the mud logging anomaly prediction knowledge base and the abnormal curve simulation module respectively, and the extraction module is used to randomly extract mud logging anomaly cases from the mud logging anomaly prediction knowledge base and replay the mud logging curves corresponding to the mud logging anomaly cases; The marking module is used to mark abnormal data points in the logging curve to generate marking information; The student forecast filling module is used to generate a student forecast filling form; The answer forecast generation module is connected to the extraction module, and the answer forecast generation module is used to automatically generate an answer forecast list in combination with the logging parameters; The abnormal forecast comparison module is connected to the student forecast filling module and the answer forecast generation module respectively, and the abnormal forecast comparison module is used to compare the student forecast filling form with the answer forecast form and output the student forecast result score; The forecast result analysis module is connected to the abnormal forecast comparison module, and the forecast result analysis module is used to generate forecast result analysis and modification suggestions for the forecast results of the students; The well control knowledge simulation module is used to implement a well control knowledge simulation test; The result scoring module is connected to the well control knowledge simulation module, and the result scoring module is used to complete the result scoring.

2. The logging anomaly identification and early warning training system according to claim 1 is characterized by: The annotation information generated in the annotation module includes the annotation curve type, annotation time and annotation content.

3. The logging anomaly identification and early warning training system according to claim 1 is characterized by: The answer forecast form generated by the answer forecast generation module and the student forecast filling form generated by the student forecast filling module both include the well number, date, well depth, anomaly type, anomaly start time, forecast time, forecast parameters, parameter changes, analysis result forecast and suggested treatment measures.

4. The logging anomaly identification and early warning training system according to claim 1 is characterized by: The logging anomaly prediction database can input new logging data.

5. The logging anomaly identification and early warning training system according to claim 1 is characterized by: The operation sites where the prediction cases in the logging anomaly prediction knowledge base are located include Gaomo block, Weiyuan block, Changning block, Shuangyushi block, Luzhou block, Zigong block, Xiangguosi block and various exploration well areas; the categories of abnormal prediction cases in the logging anomaly prediction knowledge base include gas logging anomaly, gas invasion, overflow, drill bit thorn, well leakage, water eye blockage, hydrogen sulfide anomaly, and stuck.

6. A method for using a logging anomaly recognition and early warning training system, characterized in that: The logging anomaly recognition and early warning training system according to claim 1 comprises the following steps: S1. Collect logging data at the logging operation site for different logging abnormal events; S2. According to the parameter change characteristics in different types of abnormal logging events, the logging data is standardized into various abnormal logging parameters; S3, establishing a logging anomaly prediction database to store logging anomaly parameters of different types of logging anomaly cases, and simulating and generating corresponding logging curves for each logging anomaly parameter; S4. Automatically generate an answer forecast sheet based on the abnormal logging cases selected by the students, compare the student forecast filling sheet with the answer forecast sheet, and output the student forecast result score, forecast result analysis and modification suggestions.

7. The method for using a logging anomaly identification and early warning training system according to claim 1, characterized in that: In step S4, the student forecast result score is obtained by scoring the student forecast result according to the established comprehensive evaluation model. Different scoring models can be formulated for different contents on the forecast form. A comprehensive evaluation model can be formulated by combining various scoring models. The specific method is as follows: S41. For purely numerical answers, an error scoring strategy is used to select an error range between the student's answer and the standard answer. The smaller the error range, the higher the score for the single item. The purely numerical answers include well depth, abnormal start time, forecast time, and parameter changes. S42. For the content of the text answer, the recommended treatment measures for the scoring of the answer items are to first train the text of the standard answer library with the Word2Vctor model, establish a mapping relationship between the text and the numerical vector, and convert the standard answer and the student's answer from the word vector to the numerical vector through the trained Word2Vctor model during scoring, calculate the cosine distance of the converted numerical vector, and score the student's answer using the distance metric; S43. Comprehensive scoring is performed based on the scores calculated for numerical and textual answers.

8. The method for using the logging anomaly identification and early warning training system according to claim 7 is characterized by: In step S41, the scoring rules for the well depth, abnormal start time, forecast time, and parameter change are specifically as follows: a). The specific scoring calculation rules for the well depth item satisfy the relationship in formula (1): In formula (1), S D Denotes the student's score in the well depth answer, dimensionless; D stu represents the answer given by the student in the well depth answer item, m; D sta Denotes the standard answer in the well depth answer, m; D rang is the allowable error value in the well depth answer, m; b) The specific scoring calculation rules for the abnormal start time item satisfy the following formula (2): First convert the time format to seconds. The calculation formula is as follows: BT=H*3600+M*60+S (2), In formula (2), BT represents the number of seconds of the abnormal start time, s; H is the time value in the time format, dimensionless; M is the value of the minute in the time format, dimensionless; S is the value of the second in the time format, dimensionless; Then, based on the time format converted into seconds, the score of the abnormal start time item is calculated. The specific scoring rule satisfies the following formula (3): In formula (3), S BT It represents the student score in the abnormal start time item, dimensionless; BT stu Indicates the answer filled in by the student in the abnormal start time answer item, s; BT sta Indicates the standard answer in the abnormal start time answer item, s; BT rang is the error value allowed in the answer item of abnormal start time, s; c) The specific scoring rules for the forecast time item satisfy the following relationship (4): First convert the time format to seconds. The calculation formula is as follows: WT=H*3600+M*60+S (4), In formula (4), WT represents the number of seconds of the forecast time, s; H is the hour value in the time format; M is the minute value in the time format; S is the second value in the time format; Then, based on the conversion of the time format into seconds, the score of the forecast time item is calculated. The specific scoring rule satisfies the following formula (5): In formula (5), S WT Represents the student score in the forecast time term, dimensionless; WT stu Indicates the answer filled in by the student in the forecast time answer item, s; WT sta Indicates the standard answer in the forecast time answer item, s; WT rang is the allowable error value in the forecast time answer item, s; d) The specific scoring rules for parameter change items satisfy the following formula (6): In formula (6), It represents the student score in the parameter change item, dimensionless; i represents the parameter number, the value range is [1, M], M represents the total number of parameters considered for a single forecast; P stu Indicates the answers filled in by the students in the parameter change items; P sta Indicates the standard answer in the parameter change answer item; P rang is the error value allowed in the answer item of the parameter change item.

9. The method for using the logging anomaly recognition and early warning training system according to claim 8, characterized in that: In step S42, the method for obtaining the text answer content is: i). First, the answers to the establishment of treatment measures in the standard answer library are segmented. When segmenting the text description type, irrelevant words and punctuation marks that are irrelevant to the text semantics are not considered; ii) Then all the words in the answer library are encoded by one-hot encoder. The length of the encoding is the total number of words in the answer library. On the basis of the encoding, word vector mapping containing semantic information is performed, and the word embedding word vector after word2vector model training is used; iii). The scoring rules for the answer items of the recommended treatment measures satisfy the following relationship (7): In formula (7), S W It is the score of the answer item of the recommended treatment measures, dimensionless; The word embedding encoding of the answer filled in by the student in the recommended treatment measures answer item is the value of the i-th position, dimensionless; The word embedding encoding of the standard answer in the recommended treatment answer item is a dimensionless value at the i-th position.

10. The method for using the logging anomaly recognition and early warning training system according to claim 9, characterized in that: In step S43, a comprehensive score is given based on the numerical and text answers. The scoring method satisfies the following equations (8) and (9): w1+w2+w3+w4+w5=1 (9); In formula (8) and formula (9), S is the student score, dimensionless; w1 is the depth item answer score weight, dimensionless; w2 is the score weight of the answer to the abnormal start time item, dimensionless; w3 is the score weight of the answer to the abnormal warning time item, dimensionless; w4 is the score weight of the parameter change answer, dimensionless; w5 is the score weight of the answer to the recommended treatment measure, which is dimensionless.

Images