RU2723805C9 - Method and computer system for control of drilling of the wells - Google Patents

Abstract

FIELD: soil or rock drilling.SUBSTANCE: invention relates to drilling wells in productive formations. In accordance with disclosed method, initial data characterizing drilling process and containing real-time data, measured during well drilling, are selected, contextual drilling data related to drilling macro-parameters, and geological data characterizing productive formation, in which drilling is performed. Collected data is computer processed and structured data arrays containing complete characteristic of drilling process at each moment of time are formed. Using structured data arrays as input data, actual conditions of drilling are determined in real time and probability of occurrence of complications during drilling is predicted. Using input data determined in real time actual conditions of drilling and probability of occurrence of complications during drilling, recommendations are made on correction of trajectory of well shaft for obtaining maximum productivity of well and / or recommendations on change of drilling modes, composition of drilling mud. Computer system for controlling well drilling in accordance with the present invention comprises a drilling data collection unit, a computer unit for collecting collected data, a real-time analytics unit and a real-time recommendation generation unit.EFFECT: technical result consists in creation of effective and high-speed control system of wells drilling, providing accident-free well cabling under varying conditions and maximum extraction of hydrocarbons.9 cl, 3 dwg

Description

The invention relates to the field of drilling wells in productive formations and, more specifically, to methods and systems for well drilling control using machine learning and mathematical optimization methods.

Well drilling is the first stage in the development of oil and gas fields. Drilling of directional and horizontal wells provides the process of hydrocarbon production from thin productive layers (intervals) of complex geometrical shape. When drilling wells, the trajectory and drilling modes are controlled by specialists who are guided by:

- data of geological and technical information (GTI, or telemetry of drilling from the surface, including weight on the hook, load on the bit, rotor speed, pressure in the injection line, fluid flow rate, etc.);

- directional survey data containing information about the borehole deviation angle and its spatial position;

- logging data while drilling, including gamma logging, neutron, density logging, imager, etc .;

- geological and technical order for well construction;

- data on the proposed section based on the results of seismic surveys and the results of drilling reference wells.

This approach to drilling control has the following disadvantages:

- non-measurement zone - the distance between the bit and the logging sensors. This zone causes a delay in information about the drilling conditions, which leads to lengthy corrections or overdrilling of a part of the wellbore;

- impossibility of visual control over a large number of telemetry curves simultaneously, which is the cause of accidents;

- inconsistency of the well construction project with the actual drilling conditions, which leads to overestimated terms and cost of well construction, failure to achieve the design flow rate.

There is a known method for controlling well drilling, described in the patent of the Russian Federation 2542026 and containing the following steps: preparation of drilling equipment having a bottom drill string layout, which includes a controlled directional drilling subsystem and a directional logging instrument while drilling with the possibility of circular viewing and proactive viewing; determining the presence of a given type of formation feature in the target formation; and navigation of the drilling trajectory in the target formation by said drilling equipment, including receiving measurement signals from said directional measuring device, obtaining, based on the received signals, measurements of formation parameters relative to said formation feature in the target formation and controlling said subsystem of directional drilling for drilling in the direction determined depending on the specified obtained parameter of the formation parameters.

The disadvantage of this method is that the properties of the formation are determined based on the results of interpretation by a specialist geologist of information from a logging tool located much higher than the bit. Thus, information about the strength properties of the rock being drilled is obtained with a delay associated with:

- The remoteness of the device;

- The time of expert interpretation.

This leads to overshoot of the target formation and lengthy trajectory adjustments.

RF patent 2588526 describes a method for controlling well drilling, including receiving a continuous stream of real-time data associated with a current downhole operation in a data warehouse. In this case, the user receives the choice of the bottomhole parameter. Then, using the computer system, the selected bottomhole parameter is optimized based on a portion of the received data stream to achieve the target value of the selected bottomhole parameter. The optimized bottomhole parameter can then be used in the current operation.

The disadvantages of this method are:

The presence of calculated parameters based on formulas leads to lengthy calculations, possible errors and the impossibility of calculating a large number of variations in the development of events;

Only one bottomhole parameter is optimized at a time, but not their combination;

Works on the principle of comparing current and project data and only allows you to set indicators for deviations from the plan;

Manual input of some of the drilling macroparameters is required.

There is also known a drilling control method described in RF patent 2600497, in accordance with which sensor data is collected with respect to adjacent wells and contextual data with respect to adjacent wells, placed in storage, and models for predicting the drilling process are created.

The disadvantages of this method are:

The method described in the invention for smoothing the initial data leads to the removal of a part of the values corresponding to the real drilling anomaly (complications, precursors of accidents);

Removal of incorrect values occurs with the formation of "voids" in their place;

Data from offset wells is required to generate a forecast;

The presence of calculated parameters based on formulas leads to lengthy calculations and possible errors;

Requires manual input of some of the drilling macroparameters;

Contains multiple models for one forecast unit, which can lead both to errors within each model, and in the process of choosing the main model at the moment;

Predicts bottomhole operation without recommending parameters to achieve the best result.

The technical result achieved during the implementation of the invention consists in creating an efficient and fast-acting well drilling control system that ensures trouble-free drilling in changing conditions and maximizes hydrocarbon production during subsequent operation with minimal investment in well construction.

The specified technical result is achieved by the fact that, in accordance with the proposed method of well drilling control, initial data is collected characterizing the drilling process and containing real-time data measured during the well drilling process, contextual drilling data related to drilling macroparameters and geological data characterizing the productive formation , in which drilling is carried out. Computer processing of the collected data is carried out and structured data arrays are formed containing a complete description of the drilling process at each moment of time. Using the generated structured data sets as input data containing the full characteristics of the drilling process at each moment of time, they determine in real time the actual drilling conditions, including the properties of the rock being drilled at the current moment, the potential productivity of the reservoirs, the rheological characteristics of the drilling fluid and predict the likelihood of complications during drilling. Using as input data actual drilling conditions determined in real time and the likelihood of complications during drilling, recommendations are made on adjusting the wellbore trajectory to obtain maximum well productivity and / or recommendations for changing the drilling modes, the composition of the drilling fluid.

Real-time data measured while drilling a well contains at least one type of data selected from the group consisting of measurements from sensors located on the surface, directional survey results, logging data while drilling, and measurements of downhole sensors.

The drilling contextual data related to the drilling macroparameters contains at least one type of data selected from the group containing the well construction project, the position of the offset wells, the parameters of the drilling fluid, the components of the drilling fluid available.

The contextual data characterizing the reservoir contains at least one type of data selected from the group containing an initial geological model, a geomechanical model, a predicted section, an initial flow rate, and production levels.

Computer processing of data for the formation of structured data sets containing a complete description of the drilling process at each point in time contains at least one of the following actions: bringing the collected data to a unified format, eliminating the fact of sensor calibration while drilling, aligning dimensions and units of measurement, identifying gaps in sensor readings using machine learning methods and missing values recovery, error estimation when recovering missing values in sensor readings using machine learning methods.

In accordance with the proposed method, at least one of the actual drilling conditions is determined in real time, selected from the group: properties of the rock being drilled according to data from sensors on the surface, lithotype according to the results of logging while drilling, the productivity of the formations being drilled while drilling, rheological characteristics of the drilling solution in real time, the likelihood of complications.

A computer-based well drilling control system in accordance with the invention comprises a drilling data collection unit designed to collect real-time data measured while drilling a well, contextual drilling data related to drilling macroparameters and geological data characterizing the reservoir in which drilling, a block for computer processing of collected data, designed to generate structured data sets containing a complete description of the drilling process at each moment of time, a real-time analytics unit using generated structured data sets containing a complete description of the drilling process at each moment of time as input , and designed to determine the actual drilling conditions and predict the likelihood of complications during drilling and a block for generating recommendations in real time, using the results of work as input data. You are a real-time analytics unit designed to develop recommendations for adjusting the wellbore trajectory while drilling to obtain maximum productivity, recommendations for changing drilling modes, recommendations for changing the composition of the drilling fluid.

The block of computer processing of the collected data, designed to form structured data sets containing a complete description of the drilling process at each moment of time, uses artificial intelligence methods to perform at least one action selected from the group: bringing the collected data to a single format, identifying anomalies in the data - outliers, incorrect values, facts of sensor calibration during drilling, alignment of dimensions and units of measurement, elimination of the facts of sensor calibration during drilling, restoration of missing values in sensor readings using machine learning methods, error estimation when recovering missing values in sensor readings using machine learning methods learning.

The real-time analytics unit, using as input data generated structured data sets containing a complete description of the drilling process at each point in time, uses artificial intelligence methods to determine at least one of the actual drilling conditions: determining the properties of the rock being drilled according to the data of sensors on the surface , determination of the lithotype based on the results of LWD interpretation, determination of the productivity of the formations being drilled while drilling, determination of the rheological characteristics of the drilling fluid in real time, as well as for predicting the likelihood of complications during drilling.

The invention is illustrated by drawings, in which FIG. 1. shows a general diagram of a computer system for well drilling control; in fig. 2 shows a block diagram of data collection and formation of structured data sets; in fig. 3 shows a block diagram of the work of the analytics unit and the unit for generating recommendations in real time.

The proposed method and system for well drilling control is a combination of software blocks and operations that ensure trouble-free well construction in changing or different from the design conditions.

FIG. 1 shows a general diagram of a well drilling control system containing a block 1 for collecting initial data, a preprocessor 2 - a block for computer processing of the collected data, block 3 for analytics in real time, block 4 for generating recommendations in real time. The collected initial data is preprocessed using preprocessor 2, after which it can be used by the real-time analytics block 3 and the real-time recommendations block to issue recommendations.

As shown in FIG. 1 and FIG. 2, through the block 1 collecting the initial data collect the initial data 1 about drilling. As the initial data 1, drilling tracking data are used, which can be divided into measured in real time (hereinafter real-time data 5 in Fig. 2) and data not measured during drilling and related to the entire current well or previously drilled wells (hereinafter contextual or non-real-time data 6 in Fig. 2). As real-time drilling data 5, at least one type of data is used, selected from the group consisting of measurement results from sensors located on the surface (sensors of a drilling rig or geological research station - GTI), inclinometry results (measurements of the position of the wellbore in space), logging data while drilling, measurement results of downhole sensors.

Measurements from surface sensors include drill string or tool weight, bottomhole depth, inlet pressure, WOB, pump strokes, tool speed, temperature, hydrocarbon concentrations (methane, ethane, propane, etc.) etc.), etc.

As contextual or non-real-time data 6 of drilling, not measured during drilling, at least one type of data selected from the group containing the project for the construction of a well, the results of measurements of drilling fluids, the position of the boreholes of adjacent wells, data characterizing the productive formation, which contain at least one type of data selected from the group containing the initial geological model, geomechanical model, initial well rates, well production rates, etc.

The initial data 1 goes to the preprocessor 2 (see Fig. 1) - a block of computer processing of the collected data, which performs computer processing of the collected data. Preprocessor 2 aggregates data on measurements of various formats into a single database using intelligent algorithms for approximate matching and error correction. Preprocessor 2 uses artificial intelligence techniques to perform one or more actions selected from the group:

- bringing the collected data to a single format, including the alignment of dimensions and units of measurement (7 in Fig. 2);

- identification of data anomalies: gaps, outliers, incorrect values, the facts of sensor calibration during drilling (8 in Fig. 2) using machine learning methods;

- intelligent processing: filling in the gaps in the readings of the sensors, eliminating part of the data containing the calibrations of the sensors and restoring the missing values using machine learning methods (9 in Fig. 2);

- error estimation when recovering missing values in sensor readings by machine learning methods (10 in Fig. 2).

Function 7 of preprocessor 2 - converting the collected data to a unified format - performs operations such as aggregating time series to separate multilateral wellbores, translating .las files into .csv format, determining units of measurement (based on statistical tests of the correspondence of distributions of values to reference samples with known units), alignment of dimensions and units of measurement (using conversion tables of known dimensions and rules based on trained decision trees in general), dividing time series into sections (direction, conductor, production casing, etc.), dividing operations into phases (drilling , descent, ascent, development, reverse development, flushing, idle), aggregation of time series by depth, extraction of information on the composition of the BHA, drill string, properties of drilling fluids, etc. from the design data.

Function 8 of preprocessor 2 "data anomaly detection" uses machine learning algorithms to determine the presence of anomalies in the sensor readings corresponding to gaps, outliers, incorrect values (for example, values many tens of times higher than physically justified values), instrument calibrations and marks them in accordance with the identified anomaly.

Function 9 of preprocessor 2 "intelligent processing" uses machine learning algorithms to recover missing values, learning at intervals with a full range of measurements, corrects outliers, removes some of the data containing spurious information from the sensor calibration process. As such algorithms, recurrent neural networks for multivariate time series are used (see, for example, Graves, Alex. "Generating sequences with recurrent neural networks." ArXiv preprint arXiv: 1308.0850 (2013).) As well as generative autoencoders (see, for example, Makhzani, A., Shlens, J., Jaitly, N., Goodfellow, I., & Frey, B. (2015). Adversarial autoencoders. ArXiv preprint arXiv: 1511.05644). Thus, preprocessor 2 restores the missing values, does not delete, but corrects the anomalous values.

The function 10 of preprocessor 2 "estimation of the error in the restoration of missing values in the sensor readings" uses information about the proportion of missing values and the size of the training sample to calculate the probability of correct restoration of missing values. The model for estimating the probability of correct recovery is trained on the basis of artificially created gaps of various nature on complete data.

As shown in FIG. 2, the result of the work of the preprocessor 2 is structured data sets containing a complete description of the drilling process at each point in time - dataframes 11, which are tables in .csv format, containing in a structured form a complete description of the drilling process at each point in time throughout the entire construction cycle wells.

Thus, preprocessor 2 - a block of computer processing of the collected data - provides the formation of structured data arrays containing a full description of the drilling process at each moment of time - dataframes. As input data, drilling tracking data from various contractors and obtained from various sensors can be used. This does not require manual input of additional parameters. On dataframes, predictive and optimization algorithms based on machine learning methods are trained, and trained models are used to work with new input data obtained during drilling in real time and generate forecasts.

FIG. 3 shows the principle of operation of those shown in FIG. 1 block 3 analytics in real time and block 4 making recommendations in real time.

Real-time analytics and advisory blocks are tools for diagnosing and managing drilling contingencies. The system determines the actual drilling conditions, analyzes the degree of their deviation from the design values, determines the presence of critical anomalies and recommends further actions based on the conditions for obtaining maximum production and safe operation.

Real-time analytics block 3 uses dataframes 11 as input and uses machine learning algorithms to determine at least one of the actual drilling conditions, selected from the group, such as the properties of the rock being drilled (module 12 in Fig. 3), lithotype based on the results LWD interpretation (module 13 in Fig. 3), productivity of the formations being drilled (module 14 in Fig. 3), rheological characteristics of the drilling fluid (module 15 in Fig. 3, the probability of complications (module 16 in Fig. 3).

The principle of determining the properties of the rock to be drilled (module 12 in Fig. 3) is based on the fact that when the strength properties of the rock change, the drilling regimes, which are recorded by the rig sensors or GTI stations on the surface, also change. To solve the problem of determining the drillable lithotype, a machine learning algorithm is used (see, for example, Klyuchnikov, N., Zaytsev, A., Gruzdev, A., Ovchinnikov, G., Antipova, K., Ismailova, L., ... & Koryabkin, V. (2019). Data-driven model for the identification of the rock type at a drilling bit. Journal of Petroleum science and Engineering, 178, 506-516), which uses attributes such as ROP, WOB, RPM rotor, torque and other recorded parameters. Historically trained algorithm based on an ensemble of decision trees (XGBoost, see T. Chen and C. Guestrin. XGBoost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pages 785- 794. ACM, 2016) analyzes patterns in parameter changes and determines the type and properties of the rock being drilled. After receiving data from the logging tools, they are automatically interpreted by module 13 in FIG. 3 with the subsequent refinement of predictive models (see, for example, Romanenkova, E., Zaytsev, A., Klyuchnikov, N., Gruzdev, A., Antipova, K., Ismailova, L., ... & Koroteev, D. (2019 ). Real-time data-driven detection of the rock type alteration during a directional drilling. ArXiv preprint arXiv: 1903.11436.

Autointerpretation of LWD data (module 13 in Fig. 3) is performed using machine learning algorithms based on decision trees and reduces the influence of the human factor on the interpretation. Similarly to module 12, the algorithm is trained on the full set of historical logging data and the results of well logging interpretation (WGIS) with the subsequent issuance of a forecast based on the measurement results.

The productivity of the formations being drilled in real time (module 14 in Fig. 3) is determined based on the analysis of measurement data of the operating parameters of drilling, information about the drilled lithotype from the algorithm for predicting the type of rock at the bottomhole, as well as interpretation data from logging tools (module 13), received from the delay associated with the remoteness of the tools from the bit. Similar to module 12, an algorithm based on gradient boosting and neural networks analyzes the information received and gives the productivity value of the interval being drilled.

The rheological characteristics of the drilling fluid (module 15 in Fig. 3) are reconstructed in the form of time series using machine learning algorithms based on the dataframes 11 of its component composition, the properties of the rock being drilled from modules 12 and 13, the productivity of formations 14 and the time of the drilling fluid in the well. The algorithm is trained on historical data from previously drilled wells, the result of its operation is calibrated in real time when the data of measurements of the parameters of the drilling fluid enters the system (part of the drilling context data 6 in Fig. 2, and real-time data 5 in Fig. 2 from downhole sensors and sensors on the surface).

Analytics unit 3 also predicts the probability of drilling complications in real time (module 16 in FIG. 3).

Determination of the likelihood of drilling complications is based on determining the presence of critical deviations from normal operations and assessing the proximity of the current drilling situation to one of the groups of possible complications (see, for example, Gurina, E., Klyuchnikov, N., Zaytsev, A., Romanenkova , E., Antipova, K., Simon, I., ... & Koroteev, D. (2019). Failures detection at directional drilling using real-time analogs search.arXiv preprint arXiv: 1906.02667).

Module 16 works only on the basis of the analysis of surface measurements data, without reference to adjacent wells, while applying a machine learning algorithm based on ensembles of decision trees. The algorithm analyzes the parameters of the sliding window drilling modes received in real time. For each measurement, the probability of a complication in the well is determined. Along with this, a comparison is made with areas from different categories of complications loaded into the training sample and the level of similarity of the current situation with anomalies of various types is determined. If the threshold value for both comparisons is exceeded, an alarm occurs about a possible anomaly with an indication of its type. Such an algorithm is able to identify both complications that have already begun and their precursors, as well as find matches with any other categories of cases loaded into the training set. Anomaly alarms can be configured with different sensitivities and signal signs of complication at different levels of risk. Having trained on a sufficient dataset, it can detect signs of complications without additional training and calibration.

Shown in FIG. 1 block 4 recommendations in real time uses the results of the work of block 3 analytics in real time as input data.

The work of module 17 (Fig. 3) on trajectory correction is characterized by the fact that it analyzes information about the potential productivity of the formations being drilled (module 14 in Fig. 3) and compares them with the dataframes of the initial geological model. The initial geological model contains information about the geological structure of layers and their characteristics (porosity, permeability, fluid saturation). The system calculates the values of the predicted starting flow rate for various options for the position of the wellbore in the productive formation, analyzes the risks of intersection with other boreholes, calculates the possibility of running completion assemblies and makes recommendations for wellbore adjustments with the condition of obtaining the maximum flow rate with minimal risks. It uses a machine learning algorithm.

The module of recommendations for changing the drilling modes (module 18 in Fig. 3) is based on an intelligent analysis of the operating parameters of drilling to optimize the ROP. For this, a surrogate model is applied (see, for example, Forrester, A., Sobester, A., & Keape, A. (2008). Engineering design via surrogate modeling: a practical guide. John Wiley & Sons) using one of the methods among Gaussian processes, nuclear regression, multidimensional spline, gradient boosting on decision trees. Constraints are imposed on the constructed model based on the safe conduct of work and trouble-free operation of the equipment. At the output, the optimizer gives the recommended values for WOB, RPM, and pump flow to achieve the best performance.

The module of recommendations for changing the composition of the drilling fluid (module 19 in Fig. 3) is based on an intelligent analysis of the rheological characteristics 15, the date of the ROP frames 11, the properties of the rock being drilled 12, 13, the likelihood of complications 16. At the output, the module gives recommendations for changing the component composition of the drilling fluid based on the conditions for achieving the maximum drilling speed while ensuring safe operation and maintaining the properties of the productive formation.

All algorithms used in the drilling control system and method are trained on a set of historical data from previously drilled wells. At the same time, the operation of the tuned algorithms can be carried out even when an incomplete data set is received, and the returned result contains information about the prediction accuracy or recommendations based on the available data set.

Thus, the proposed method and well drilling control system allow:

1.determine the type and properties of the rock being drilled out at the time of observation;

2. to recommend a drilling trajectory with the condition of obtaining maximum production;

3. Recommend optimal drilling regimes to achieve drilling goals with a minimum total cost of ownership of the well, while maintaining or increasing production;

4. to recommend a drilling fluid that provides the maximum rate of penetration with a minimum risk of complications and minimum labor costs to change its properties;

5. determine the likelihood of an accident or complication, determine the type of possible accident or complication in order to take preventive measures to prevent it or minimize its consequences.

The following is an example of the operation of the method and the computer system for well drilling control.

In the process of well construction, raw real-time data of drilling 5, which is a multidimensional time series, is measured using the rig sensors, and is stored in the user's local storage in las or csv format.

Contextual data 6 are loaded into local storage in the form of tables or sets of structured text files (for example, csv, json formats) containing design characteristics, measurement results and geological data. The contextual data 6 can be updated by the user in case of changes in drilling conditions or design decisions.

Preprocessor 2, according to a specific schedule specified by the user, loads a new batch of initial real-time drilling data 5 and context data 6. When each new portion of data arrives, preprocessor 2 brings them to a single format - it performs operations such as time series aggregation, translation csv format, defining and aligning units of measure, dividing time series into sections, dividing operations into phases, aggregating time series by depth, extracting information about the composition of the BHA, drill string, properties of drilling fluids, etc. from context data. function (7), we obtain standardized wellbore data sets suitable for further analysis, which contain all the information necessary for this from the initial data.

Then, data anomalies (8) are detected, corresponding to gaps, outliers, incorrect values, instrument calibrations. Machine learning algorithms predict the expected values of the time series, then they are compared with the actual ones, and, if there is a strong discrepancy, they are marked as anomalous. As a result of the execution of this function (8), an additional service file appears in the data arrays, indicating the presence of anomalies and their types for each trunk.

After that, intelligent processing of anomalies is carried out (9). The preprocessor algorithms restore missing values, correct outliers, and remove some of the data containing sensor calibrations. As a result, we get data arrays similar to the original, but with corrected values.

In this case, the error is estimated when restoring the missing values in the sensor readings. Algorithms (10) use information about the proportion of missing values and the size of the training sample to calculate the probability of correct restoration of missing values. As a result of the execution of this function (10), an additional column appears in the data arrays for each wellbore indicating the reliability in the restoration of missing values.

The result of the preprocessor is written into the terminal table in the form of dataframes 11, which are csv files.

The processed dataframes 11 are received by the analytics unit in real time, where tasks 12-16 in FIG. 3.

Receiving a portion of data as input, the trained algorithms of block 12 for determining the properties of the rock being drilled while drilling select from the dataframe a data interval corresponding to one drilled meter, and statistics are calculated on it from the time series, such as mean, variance, inclination angle and the difference between values on borders. The feature vector from these statistics is then passed to the input of the trained rock properties classifier. The output is the number of the class of rocks and, accordingly, their properties. The result of the algorithms is written into the terminal table of the database and displayed to users using a graphical interface.

Receiving a portion of the data as input, the trained algorithms of block 13 of automatic interpretation of logs calculate time series statistics as in block 13, as well as such statistics of log data as moving averages with different window sizes, their differences, etc. All the listed statistics are then passed to the input of a trained rock classifier. The output is the name of the rock from the log interpretation data. The result of the algorithms is written into the terminal table of the database and displayed to users using a graphical interface.

Receiving a portion of data as input, the trained algorithms of block 14 for determining the productivity of the formations being drilled analyze dataframes of drilling operating parameters, data from logging tools, geological data, and predict productivity parameters by the regression method. The output is the expected starting flow rate for a given borehole length and position in the formation. The result of the algorithms is written into the terminal table of the database and displayed to users using a graphical interface.

Receiving a portion of data at the input, the trained algorithms of block 15 for determining the rheological characteristics of the drilling fluid analyze the dataframes of the operating parameters of drilling and predict the values of the parameters of the drilling fluid. The predicted values of the parameters are calibrated based on the results of actual measurements. The output is the value of the current parameters of the drilling fluid at each moment of time. The result of the algorithms is written into the terminal table of the database and displayed to users using a graphical interface.

Receiving a portion of data as input, the trained algorithms of block 16 for determining the probability of complications select from the dataframe a data interval corresponding to two hours, and statistics are calculated on it from the time series, such as mean, variance, slope and the difference between values at the boundaries. The feature vector from these statistics is compared with the cases from the complications database to assess the similarity of the situation to the emergency. The result of the algorithms is written into the terminal table of the database and displayed to users using a graphical interface.

The result of the operation of blocks 12-16 is transmitted to the recommendation block in real time 4.

Real-time recommendation block 4 selects the best combination of drilling parameters, mud and trajectory by simulating the final well productivity and predicting the probability of complications for various combinations of combinations.

Block 4 issues recommendations for adjusting the wellbore trajectory to achieve maximum productivity 17; recommendations for changing at least one of the parameters: pump flow, bit load, rotation speed to ensure the maximum drilling speed in a drillable cast type, provided that the work is carried out safely (18); recommends changes in the composition of the drilling fluid to achieve maximum drilling efficiency, high-quality hole cleaning and safe operation conditions (19).

Claims (37)

1. Method for controlling well drilling, according to which - collecting initial data characterizing the drilling process and containing real-time data measured in the process of drilling a well, contextual drilling data related to drilling macroparameters, and geological data characterizing the productive formation in which drilling is carried out; - carry out computer processing of the collected data and form structured data arrays containing a complete description of the drilling process at each moment of time; - using as input data generated structured data sets containing a complete description of the drilling process at each moment of time, they determine in real time the actual drilling conditions, including the properties of the rock being drilled at the current moment, the potential productivity of the reservoirs, the rheological characteristics of the drilling fluid, predict the likelihood of complications while drilling; - using actual drilling conditions determined in real time and the likelihood of complications during drilling as input data, they develop recommendations for adjusting the wellbore trajectory to maximize well productivity and / or recommendations for changing drilling modes, drilling fluid composition. 2. The well drilling control method according to claim 1, in accordance with which the real-time data measured while drilling the well contains at least one type of data selected from the group consisting of measurements from sensors located on the surface, results directional survey, logging data while drilling, measurement results of downhole sensors. 3. The well drilling control method according to claim 1, in accordance with which the drilling contextual data related to the drilling macroparameters contains at least one type of data selected from the group containing the well construction project, the position of the adjacent wells, the parameters of the drilling mud , drilling mud components on hand. 4. A method for controlling well drilling according to claim 1, in accordance with which the contextual data characterizing the productive formation contains at least one type of data selected from the group containing an initial geological model, a geomechanical model, a predicted section, an initial flow rate, production levels ... 5. A method for controlling well drilling according to claim 1, in accordance with which computer processing of data to generate structured data sets containing a complete characteristic of the drilling process at each moment of time comprises at least one of the following actions: - bringing the collected data to a single format; - exclusion of the facts of calibration of sensors during drilling; - alignment of dimensions and units of measurement; - identifying gaps in sensor readings using machine learning methods and recovering missing values; - estimation of the error in the restoration of missing values in sensor readings using machine learning methods. 6. A method for controlling well drilling according to claim 1, in accordance with which at least one of the actual drilling conditions selected from the group is determined in real time: - properties of the rock being drilled according to the data of sensors on the surface; - lithotype based on the results of logging while drilling interpretation; - productivity of the formations being drilled while drilling; - rheological characteristics of the drilling fluid in real time; - the likelihood of complications. 7. Computer control system for drilling wells, containing: - a drilling data collection unit designed to collect real-time data measured while drilling a well, drilling contextual data related to drilling macroparameters, and geological data characterizing the productive formation in which drilling is performed; - a block of computer processing of the collected data, designed to form structured data sets containing a complete description of the drilling process at each moment in time; - a real-time analytics unit using the generated structured data arrays as input data containing a complete description of the drilling process at each time point, and designed to determine the actual drilling conditions and predict the likelihood of drilling complications; - a block for generating recommendations in real time, using the results of the work of the analytics block in real time as input data and designed to develop recommendations for adjusting the trajectory of the wellbore during drilling to obtain maximum productivity, recommendations for changing drilling modes, recommendations for changing the composition of the drilling fluid. 8. Computer control system for drilling wells according to claim 7, characterized in that the computer processing unit for the collected data, designed to generate structured data sets containing a complete description of the drilling process at each point in time, uses artificial intelligence methods to perform at least one action selected from the group: - bringing the collected data to a single format; - identification of anomalies in the data - outliers, incorrect values, the facts of sensor calibration during drilling; - alignment of dimensions and units of measurement; - exclusion of the facts of calibration of sensors during drilling; - restoration of missing values in sensor readings using machine learning methods; - estimation of the error in the restoration of missing values in sensor readings using machine learning methods. 9. Computer control system for drilling wells according to claim 7, characterized in that the real-time analytics unit, using the generated structured data arrays as input data, containing a complete characteristic of the drilling process at each moment in time, uses artificial intelligence methods to determine at least at least one of the actual drilling conditions: - determination of the properties of the drilled rock according to the data of sensors on the surface; - determination of lithotype based on the results of logging while drilling interpretation; - determination of the productivity of the formations being drilled in the process of drilling; - determination of the rheological characteristics of the drilling fluid in real time, as well as to predict the likelihood of complications during drilling.

Images