A method and device for early warning of well drilling overflow risk

Abstract

This application discloses a method and apparatus for early warning of drilling spill risks, relating to the field of oil and gas drilling exploration technology. The proposed solution involves collecting logging data at different time intervals according to a preset time interval, and integrating these logging data based on their temporal order to obtain logging time-series data. Since logging time-series data is essentially a time-series signal, this application further decomposes the logging time-series data into multi-scale values ​​to extract key target features associated with spills and well leakage. These target features include time-domain features, frequency-domain features, and time-frequency-domain features. Then, a risk prediction model trained using a self-attention mechanism is used to process the logging time-series data containing these target features, aiming to obtain risk prediction results for spills and well leakage. Based on these risk prediction results, a risk warning is then triggered.

Description

Technical Field

[0001] This application relates to the field of oil and gas drilling exploration technology, and in particular to a method and device for early warning of drilling spill risk. Background Technology

[0002] During drilling operations, blowouts and lost circulation are two common but extremely dangerous risk events. A blowout occurs when formation fluids rush into the wellbore due to pressure imbalance, while lost circulation occurs when drilling fluid leaks into the formation through fractures or pores in the wellbore. Both can severely impact the safety and economics of drilling operations. If not handled properly, a blowout can escalate into a blowout, while lost circulation can lead to increased mud costs, reduced wellbore stability, and even well collapse or another blowout. Therefore, effectively monitoring and predicting the risks of blowouts and lost circulation is crucial for ensuring the safety of oil and gas field operations.

[0003] Traditional methods for monitoring overflows and lost circulation rely primarily on the experience and judgment of field operators and simple pressure monitoring equipment. However, due to the complex and variable formation conditions downhole, overflows and lost circulation are often accompanied by subtle changes in multiple parameters, making accurate early warnings difficult to achieve using only a single parameter or based on experience. Furthermore, early signs of overflows and lost circulation can be very weak and easily masked by noise, making it difficult for existing monitoring methods to capture these initial signals, resulting in the failure to detect and address risk events in a timely manner.

[0004] With the advancement of logging technology, there is an urgent need to research an intelligent early warning method to improve the accuracy of risk warnings and to utilize real-time online monitoring to provide more reliable technical support for the safety management of drilling operations, thereby reducing the possibility of accidents. Summary of the Invention

[0005] This application provides a method and apparatus for early warning of drilling spill risk. The main purpose is to construct an intelligent early warning solution for dealing with spills and well leakage in drilling operations by using multi-scale time-series data decomposition and a machine learning model trained using a self-attention mechanism.

[0006] To achieve the above objectives, this application mainly provides the following technical solutions:

[0007] The first aspect of this application provides a method for early warning of drilling leakage risk, the method comprising:

[0008] According to a preset time interval, logging data is collected at each unit time point, wherein the time length between any two adjacent unit time points is one preset time interval, and each unit time point corresponds to multiple data collection points;

[0009] Well logging time series data is obtained by integrating the logging data collected at different time units according to the chronological order.

[0010] By performing multi-scale decomposition on the logging time series data, target features associated with overflow and well leakage are extracted. These target features include time domain features, frequency domain features, and time-frequency domain features.

[0011] The well logging time series data containing the target features is processed using a pre-trained risk prediction model to output risk prediction results. The risk prediction model is a machine learning model trained using a self-attention mechanism, which is used to predict the risk of leakage at future times by capturing long-range dependencies in the well logging time series data.

[0012] Based on the risk prediction results, determine whether a risk warning should be triggered.

[0013] A second aspect of this application provides an early warning device for drilling leakage risk, the device comprising:

[0014] The acquisition unit is used to acquire logging data at each unit time according to a preset time interval, wherein the time length between any two adjacent unit times is one preset time interval, and each unit time corresponds to multiple data acquisition points;

[0015] The first processing unit is used to integrate and process the logging data collected at different time units according to the chronological order to obtain logging time series data.

[0016] The second processing unit is used to extract target features associated with overflow and well leakage by performing multi-scale decomposition on the logging time series data. The target features include time domain features, frequency domain features, and time-frequency domain features.

[0017] The third processing unit is used to process the logging time series data containing the target features using a pre-trained risk prediction model and output a risk prediction result. The risk prediction model is a machine learning model trained using a self-attention mechanism, which is used to predict the risk of leakage at future times by capturing the long-range dependencies in the logging time series data.

[0018] The first execution unit is used to determine whether to trigger a risk warning based on the risk prediction results.

[0019] A third aspect of this application provides a computer device including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the drilling spill risk warning method described above.

[0020] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the drilling spill risk warning method as described.

[0021] The fifth aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the drilling spill risk warning method as described.

[0022] By employing the above-described technical solution, the technical solution provided in this application has at least the following advantages:

[0023] This application provides a method and apparatus for early warning of drilling leakage risk. The method involves collecting logging data at preset time intervals and integrating this data according to its temporal order to obtain logging time-series data. Since logging time-series data is essentially a time-series signal, this application further decomposes the logging time-series data into multi-scale values ​​to extract key target features associated with leakage and wellbore leakage. These target features include time-domain features, frequency-domain features, and time-frequency-domain features. Then, a risk prediction model trained using a self-attention mechanism is used to process the logging time-series data containing these target features to obtain a risk prediction result for wellbore leakage. Based on this risk prediction result, a risk warning is triggered.

[0024] Compared to existing intelligent early warning requirements for potential overflows and wellbore leakage during drilling operations, this application achieves multi-scale decomposition of logging time-series data. This allows for the extraction of hidden, deep-level information from the data, specifically extracting key target features with time, frequency, and time-frequency domain characteristics that are associated with overflows and wellbore leakage. This high-quality data is then input into a risk prediction model. Utilizing the model's self-attention mechanism to capture long-range dependencies in the logging time-series data, a more accurate risk prediction of future overflows is obtained. If the risk prediction meets the warning criteria, a risk warning will be triggered rapidly. Therefore, this application, through multi-scale time-series data decomposition and a machine learning model trained using a self-attention mechanism, provides an intelligent early warning solution for overflows and wellbore leakage during drilling operations. This offers more reliable technical support for the safety management of drilling operations, thereby reducing the likelihood of accidents.

[0025] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0026] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0027] Figure 1 A flowchart illustrating a method for early warning of drilling leakage risk provided in an embodiment of this application;

[0028] Figure 2 A flowchart illustrating another early warning method for drilling leakage risk provided in an embodiment of this application;

[0029] Figure 3 A block diagram illustrating the composition of an early warning device for drilling leakage risk provided in this application embodiment;

[0030] Figure 4 A block diagram illustrating the composition of another early warning device for drilling spill risk provided in this application embodiment. Detailed Implementation

[0031] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0032] This application provides a method for early warning of drilling leakage risk, such as... Figure 1 As shown, the following specific steps are provided in this embodiment of the application:

[0033] Step 101: Collect logging data at each unit time according to the preset time interval. The time length between any two adjacent unit times is a preset time interval, and each unit time corresponds to multiple data collection points.

[0034] The embodiments of this application collect logging data at each unit of time according to a preset time interval. The purpose is to achieve real-time collection of logging data over an equal period of time, and to not limit the number of data points collected at each unit of time.

[0035] Furthermore, the logging data collected in this application embodiment during drilling operations includes various types, as exemplified below:

[0036] (1) Drilling engineering parameters, such as: drilling pressure, the pressure applied by the drill bit to the formation, reflecting the interaction between the drill bit and the formation; rotation speed, the speed at which the drill bit rotates, affecting drilling efficiency and drill bit wear; displacement, the volumetric flow rate of the drilling fluid circulation, used to carry cuttings and cool the drill bit; pump pressure, the pressure of the drilling fluid circulation system, reflecting the resistance to the flow of drilling fluid; torque, the torque required to rotate the drill string, which is related to the hardness of the formation and the condition of the drill bit.

[0037] (2) Drilling fluid parameters, such as: density, the unit volume mass of drilling fluid, used to balance formation pressure; viscosity, the flow resistance of drilling fluid, affecting cuttings carrying capacity and circulation efficiency; water loss, the amount of water in drilling fluid that permeates into the formation, reflecting the sealing performance of drilling fluid; sand content, the content of solid particles in drilling fluid, which may lead to equipment wear if too high; hydrogen ion concentration index (pH value), the acidity or alkalinity of drilling fluid, affecting drilling fluid performance and equipment corrosion.

[0038] (3) Formation information, such as: rock cuttings, rock fragments returned to the surface after the drill bit breaks the formation, used to analyze the formation lithology; lithological description, describing the mineral composition, color, structure, etc. of the rock cuttings; gas logging data, the content of dissolved gases (such as methane, carbon dioxide, etc.) in the drilling fluid, reflecting the formation fluid properties; formation pressure, the pressure of formation pore fluid, used to assess well control risks.

[0039] (4) Drilling equipment status, such as: drill bit wear, the degree of wear of drill bit teeth and bearings, which affects drilling efficiency; drill string vibration, the vibration generated by the drill string during rotation and drilling, which may cause equipment damage; downhole tool status, such as the operating status of measurement while drilling tools and logging while drilling tools.

[0040] (5) Other parameters, such as: well depth, the current drilling depth, used to determine the formation location; well inclination, the angle at which the wellbore deviates from the vertical direction, which affects the control of the drilling trajectory; azimuth, the direction of the horizontal projection of the wellbore, used to determine the position of the wellbore on the horizontal plane; temperature, the temperature of the drilling fluid, formation or downhole tools, which affects the performance of the drilling fluid and the life of the equipment.

[0041] Step 102: According to the chronological order, the logging data collected at different time units are integrated and processed to obtain logging time series data.

[0042] This application embodiment may, but is not limited to, adding timestamps to logging data collected at different unit times, thereby sorting the logging data in time according to different timestamps to obtain logging time series data composed of logging data at different unit times.

[0043] The timestamps (i.e., time information) carried in well logging time series data can reveal the acquisition time corresponding to each data point. Time information is the foundation of time series data, ensuring the orderliness and traceability of well logging data at different time units.

[0044] Step 103: By performing multi-scale decomposition on the logging time series data, target features associated with overflow and well leakage are extracted. The target features include time domain features, frequency domain features, and time-frequency domain features.

[0045] As exemplified in step 101, the logging data includes a variety of data types, which are diverse and complex. The embodiments of this application decompose the logging time series data at multiple scales in order to extract key target features associated with overflow and well leakage. The data information represented by these target features not only includes changes in the downhole physicochemical state, but also potentially reflects early signs of overflow and well leakage, thereby obtaining high-quality data for predicting overflow risk.

[0046] It should be noted that well logging time series data is essentially considered a time series signal, which gives it the following characteristics, as listed below:

[0047] (1) Temporal sequence:

[0048] The core characteristic of time-series signals is that the data is arranged in chronological order. Well logging time-series data strictly follows this rule, with each data point having a clear timestamp to ensure the order and continuity of the data.

[0049] (2) Dynamic monitoring and feedback:

[0050] Time-series signals are used to describe changes in a system over time. Well logging time-series data records the changes in various parameters during the drilling process in real time, forming a dynamic monitoring record. This data can promptly reflect changes in drilling conditions and formation characteristics, providing drilling engineers with a basis for decision-making.

[0051] (3) The unity of continuity and discreteness:

[0052] Time-series signals can be continuous (such as analog signals) or discrete (such as digital signals). Well logging time-series data is usually stored in discrete form, but continuous monitoring is achieved through high-frequency sampling. For example, gas detection data may be recorded once per minute, forming a continuous time series.

[0053] (4) Time series analysis and forecasting:

[0054] Analysis methods for time-series signals (such as trend analysis, periodic detection, and anomaly detection) can be directly applied to logging time-series data. By analyzing historical data, formation change trends can be predicted, drilling parameters can be optimized, and even potential risks (such as formation pressure anomalies) can be detected in advance.

[0055] (5) Temporal dependencies in engineering applications:

[0056] Drilling operations are highly time-dependent processes, and subsequent operations (such as casing and cementing) require time-series analysis of previous data. Logging time-series data provides temporal support for these decisions.

[0057] Therefore, the multi-scale decomposition of logging time-series data provided in this application relies on time-series signal processing technology to perform comprehensive analysis from three dimensions: time domain, frequency domain, and time-frequency domain. For example, time-domain features directly reflect the changing patterns of the signal on the time axis and are suitable for capturing transient changes in the signal; frequency-domain features convert the time-domain signal into a frequency-domain signal through Fourier transform, revealing the frequency components of the signal; time-frequency domain features combine time and frequency information and are suitable for analyzing non-stationary signals (such as frequency components that change over time).

[0058] This enables the multi-scale decomposition of logging time-series data (such as using Empirical Mode Decomposition (EMD), Variational Mode Decomposition (VMD), wavelet transform, etc., to decompose complex time-series signals into multi-scale, multi-frequency components, and extract time-domain, frequency-domain, and time-frequency-domain features), achieving the extraction of key target features with time-domain, frequency-domain, and time-frequency-domain characteristics that are associated with overflow and well leakage. These features provide rich information input for the subsequent risk prediction model in step 104, laying the foundation for accurate prediction.

[0059] The following are examples of key target characteristics related to overflow and well leakage, combining time domain and frequency domain analysis:

[0060] Example 1: Taking time-domain features as an example; time-domain features directly reflect the changing pattern of a signal along the time axis and are suitable for capturing transient changes in signals. Therefore, after decomposing the logging time-series data to extract key target features related to overflow and well leakage, the following are included:

[0061] The mean and variance are used to reflect the fluctuation range of parameters such as drilling fluid pool volume, flow rate, and density.

[0062] Use slope and rate of change: such as the rate of increase / decrease of drilling fluid pool volume, and the magnitude of sudden changes in pump pressure.

[0063] Peak and trough values ​​are used to record the extreme values ​​of parameters, such as the peak value of total hydrocarbon content in the gas.

[0064] Cumulative values, such as the total change in drilling fluid volume over a certain period of time, are used to determine the severity of overflow or well leakage.

[0065] Therefore, the target characteristics obtained are as follows: related to overflow: the drilling fluid pool volume increases rapidly, the density decreases, and the total hydrocarbon content of the gas increases sharply; related to well leakage: the drilling fluid pool volume decreases, the pump pressure decreases, and the return volume decreases.

[0066] Example 2: Overflow: Rapid increase in drilling fluid pool volume (time domain characteristics) + increase in low-frequency energy (frequency domain characteristics) + concentration of low-frequency energy within a specific time period (time-frequency domain characteristics).

[0067] Well leakage: Pump pressure drop (time domain characteristics) + increase in high-frequency noise (frequency domain characteristics) + abnormal distribution of high-frequency noise on the time-frequency plot (time-frequency domain characteristics).

[0068] Step 104: Process the logging time series data containing target features using a pre-trained risk prediction model and output the risk prediction results.

[0069] The risk prediction model is a machine learning model trained using a self-attention mechanism, used to predict the risk of leakage at future time points by capturing long-range dependencies in well logging time-series data. For example, a Transformer model can be used to construct the risk prediction model. The advantages of using a machine learning model trained using a self-attention mechanism in this application, enabling the model to predict leakage risk, include the following:

[0070] Drilling data typically contains time-series information spanning long periods (such as continuous changes in parameters like drilling pressure, torque, and mud density). Self-attention mechanisms, by directly calculating the correlation between any two time steps, can capture long-distance dependencies and avoid information decay. For example, a model can simultaneously focus on drilling parameters at the current moment and anomalous events from several hours ago, thereby more accurately predicting leakage risks.

[0071] Drilling risk prediction requires the integration of multi-source data (such as geological data, sensor data, and operation logs), and the feature dimensions and temporal characteristics of different modalities vary significantly. Self-attention mechanisms, by introducing multi-head attention (MHA), can process information from different subspaces in parallel, achieving the fusion of multi-modal data. For example, the model can simultaneously focus on the spatial distribution characteristics of geological parameters and the temporal dynamic characteristics of sensor data, improving the robustness of predictions.

[0072] Step 105: Based on the risk prediction results, determine whether a risk warning has been triggered.

[0073] For example, real-time logging data is input into a trained Transformer model, which outputs the probability distribution of leakage risk at the next time step. For instance, the model outputs a probability vector [0.1, 0.3, 0.5, 0.1], corresponding to low risk, low-medium risk, medium-high risk, and high risk, respectively. It can be seen that the prediction result is "medium-high risk". Then, based on whether the "medium-high risk" reaches the preset risk level, if so, a risk warning is immediately triggered.

[0074] The present application provides a method for early warning of drilling spill risks. By performing multi-scale decomposition on logging time-series data, hidden deep-level information is extracted from the data. Specifically, key target features with time, frequency, and time-frequency domains that are associated with spills and well leakage are extracted. This high-quality data information is then input into a risk prediction model for processing. The model's self-attention mechanism captures long-range dependencies in the logging time-series data, resulting in a more accurate risk prediction of future spills. If the risk prediction meets the warning criteria, a risk warning will be triggered rapidly. Therefore, the present application provides an intelligent early warning solution for dealing with spills and well leakage in drilling operations by utilizing multi-scale time-series data decomposition and a machine learning model trained with a self-attention mechanism. This provides more reliable technical support for the safety management of drilling operations, thereby reducing the likelihood of accidents.

[0075] To provide a more detailed explanation, this application also provides another method for early warning of drilling leakage risk, such as... Figure 2 As shown, the following specific steps are provided in this embodiment of the application:

[0076] Step 201: Collect logging data at each unit time according to the preset time interval. The time length between any two adjacent unit times is a preset time interval, and each unit time corresponds to multiple data collection points.

[0077] Step 202: According to the chronological order, the logging data collected at different time units are integrated and processed to obtain logging time series data.

[0078] For an explanation of steps 201-202 in the embodiments of this application, please refer to steps 101-102, which will not be repeated here.

[0079] Step 203: By performing multi-scale decomposition on the logging time series data, target features associated with overflow and well leakage are extracted. The target features include time domain features, frequency domain features, and time-frequency domain features.

[0080] In this embodiment, logging time-series data is essentially considered a time-series signal. Therefore, the multi-scale decomposition of logging time-series data provided in this embodiment relies on time-series signal processing technology to perform comprehensive analysis from three dimensions: time domain, frequency domain, and time-frequency domain. Exemplary steps include:

[0081] Step A1: Preprocess the logging time series data to eliminate differences in different dimensions and obtain the preprocessed target logging time series data.

[0082] This application's embodiments address missing and outlier values ​​in well logging time-series data by employing interpolation to fill in missing data and smoothing extreme outliers to ensure data continuity and reliability. Subsequently, filters are applied to remove noise signals, and data is processed using standardization or normalization methods to eliminate differences between different units, laying the foundation for subsequent analysis.

[0083] Step A2 involves performing empirical mode decomposition on the target logging time series data to obtain multiple intrinsic mode functions (IMFs). Each IMF represents a different frequency component, reflecting the multi-scale characteristics of the signal, and is used to identify potential overflow and well leakage signals.

[0084] The preprocessed time series data is subjected to EMD processing, which decomposes the data into several Intrinsic Mode Functions (IMFs). Each IMF represents a different frequency component, reflecting the multi-scale characteristics of the signal and helping to identify potential overflow and well leakage signals.

[0085] Example: In well logging data analysis, EMD can separate low-frequency trend terms and high-frequency noise terms related to overflow and well leakage. For example, overflow signals may manifest as abrupt amplitude changes in low-frequency IMF, while well leakage may be accompanied by energy decay of high-frequency IMF.

[0086] Step A3: For multimodal signals containing frequency overlap features in the logging time series data, variational mode decomposition is used to separate the various components of the multimodal signal and obtain multiple sub-signals.

[0087] When processing multimodal signals with frequency overlap, VMD technology is applied to decompose the data into multiple sub-signals. By adaptively selecting frequency bands, VMD effectively separates the various components of the multimodal signal and ultimately outputs multiple sub-signals with clear physical meanings (such as formation response, drill string vibration, drilling fluid disturbance, etc.), thereby enhancing the accuracy of feature extraction.

[0088] Example: For complex signals with overlapping frequency components in well logging data (such as the superposition of drilling fluid density fluctuations and formation fluid invasion signals), VMD can adaptively select frequency bands to separate the independent components of each mode. For example, a blowout signal may correspond to a mode in a specific frequency band, while a leak signal may manifest as energy changes in a mode in another frequency band.

[0089] Step A4: Wavelet transform is used to perform multi-scale decomposition on the logging time series data to obtain the frequency variation characteristics at different time scales.

[0090] To capture local characteristics in the data, wavelet transform is used to perform multi-scale decomposition of the time-series signal. By selecting appropriate wavelet basis functions, the frequency variation characteristics at different time scales can be revealed, extracting important information in the time and frequency domains.

[0091] Example: In well logging data analysis, selecting appropriate wavelet basis functions (such as Daubechies wavelet or Symlet wavelet) can effectively extract the time-frequency characteristics of overflow and well leakage events. For example, an overflow signal may show a continuous enhancement of low-frequency components at a certain time scale, while a well leakage signal may correspond to a transient decay of high-frequency components.

[0092] Step A5: The decomposition results obtained from empirical mode decomposition, variational mode decomposition, and wavelet transform are fused to obtain a multidimensional feature vector, which includes time-domain features, frequency-domain features, and time-frequency-domain features.

[0093] In actual well logging data analysis, the advantages of EMD, VMD, and wavelet transform can be combined. For example, the signal can be initially decomposed using EMD or VMD, and then wavelet transform can be performed on key IMFs or modes to further refine the time-frequency features, thereby obtaining multidimensional feature vectors, which include time-domain features, frequency-domain features, and time-frequency-domain features.

[0094] Step 204: Process the logging time series data containing target features using a pre-trained risk prediction model and output the risk prediction result. The risk prediction model is a machine learning model trained using a self-attention mechanism, which is used to predict the risk of leakage at future times by capturing long-range dependencies in the logging time series data.

[0095] In this embodiment of the application, this step can be further refined to include the following:

[0096] Step B1 involves inputting the logging time-series data containing the target features into the encoder of a pre-trained risk prediction model.

[0097] Step B2 uses the encoder's input embedding layer to convert the logging time series data containing target features into a high-dimensional vector representation.

[0098] In the encoder's input embedding layer, logging data (such as drilling pressure, torque, pump pressure, outlet flow rate, etc.) are converted into a high-dimensional vector representation.

[0099] Step B3: Add location codes to the converted logging time series data to preserve the time series information.

[0100] Since Transformer itself does not have the ability to process sequence order, positional encoding is required to preserve timing information.

[0101] Step B4: Based on the time series information, a self-attention mechanism is used to capture the dependencies between different positions in the transformed logging time series data and extract global features.

[0102] Step B5 involves using a feedforward neural network to perform a nonlinear transformation on the global features output by the autonomous force mechanism, thereby enhancing the feature representation capability.

[0103] Step B6: Use a classifier to predict the risk of well overflow based on the global features after nonlinear transformation, and output the risk level or probability.

[0104] As shown in B1-B6 above, in order to simplify the model structure, the embodiments of this application provide that the encoder part of the Transformer model is used to process logging time series data and predict drilling leakage risk, without the need for a decoder.

[0105] This application's embodiments are only applicable to the Transformer model where the encoder is responsible for mapping the input sequence (such as well logging time series data) to a high-dimensional feature space, capturing long-term dependencies and complex patterns in the sequence; multi-head self-attention mechanism: capable of processing all time steps in the sequence in parallel, capturing the correlation between different time steps; feedforward neural network: performs nonlinear transformations on the features of each time step, enhancing feature representation capabilities; position encoding: by adding position information, the model can perceive the order of the sequence; encoder output: the encoder outputs the hidden state (feature representation) of each time step, and these features can be directly used for classification or regression tasks.

[0106] Based on B1-B6 above, embodiments of this application can employ a risk prediction model to output the probability distribution of leakage risk at the next moment. For example, the model outputs a probability vector [0.1, 0.3, 0.5, 0.1], corresponding to low risk, low-medium risk, medium-high risk, and high risk, respectively, as the risk prediction result for the "next moment," to further determine whether a risk warning is triggered.

[0107] Step 205: If the risk prediction result output by the risk prediction model is the probability distribution of risk level at multiple future unit times, determine whether to trigger a risk warning based on the probability distribution of risk level at the next future unit time that is adjacent to the current time.

[0108] In some modified embodiments, if the model outputs a risk prediction result that represents the probability distribution of risk levels at multiple future time units, the encoder and decoder in the Transformer model are used. The advantage of this structure is that the encoder processes historical time-series data, and the decoder gradually generates future sequences, making it suitable for tasks that require generating sequence outputs. Specific explanations include the following:

[0109] Multi-step prediction is achieved: If it is necessary to predict the risk value of multiple future time steps at the same time (such as predicting the leakage risk of the next 12 hours), and the prediction of each time step depends on the output of the previous time step, then the encoder-decoder structure is a better choice.

[0110] Generative tasks are implemented: when the prediction results need to generate continuous time series outputs (such as generating time series of risks), the decoder can gradually generate future sequences through an autoregressive approach.

[0111] Therefore, to meet the requirement that "the risk prediction model outputs a probability distribution of risk levels over multiple future time units," the Transformer model training needs to ensure the implementation of the following functionalities:

[0112] Multi-step prediction: Transformer models can use a sequence-to-sequence (Seq2Seq) architecture to take historical time-series data as input and output predictions for multiple future time steps. For example, inputting logging data from the past 24 hours can predict the risk of well leakage in the next 12 hours or longer.

[0113] Timestamp encoding: To better handle time-series data, timestamp encoding (such as sine and cosine position encoding) can be added to the input data to help the model understand the time sequence.

[0114] Multi-task learning: If it is necessary to predict multiple risk indicators (such as spillover risk, leakage risk, etc.) at the same time, a multi-task learning framework can be designed to allow the model to learn multiple targets simultaneously.

[0115] Furthermore, regarding the statement that "the risk prediction result output by the risk prediction model is the probability distribution of risk level at multiple future time units," embodiments of this application can provide diverse judgment mechanisms for triggering risk warnings, such as, but not limited to, the following examples:

[0116] Among multiple future time units, the probability distribution of the risk level corresponding to the next future time unit adjacent to the current time is used to determine whether a risk warning should be triggered. In other words, for the current time, the probability distribution of the risk level at the nearest future time unit is used to determine whether there are any early signs of risk that may have been missed at the current time.

[0117] The probability distribution of the risk level corresponding to the next future unit time adjacent to the current time is a probability vector [0.1, 0.3, 0.5, 0.1], which corresponds to low risk, low-medium risk, medium-high risk, and high risk, respectively. It can be seen that the prediction result is "medium-high risk". Then, based on whether "medium-high risk" reaches the risk level of the preset warning, if so, the risk warning is triggered immediately.

[0118] In some modified embodiments, if the probability distribution of the risk level corresponding to the next future unit time adjacent to the current time is a probability vector [0.1, 0.5, 0.3, 0.1], corresponding to low risk, low-medium risk, medium-high risk, and high risk respectively, it can be seen that the prediction result is "low-medium risk". Based on the risk level of the preset warning, it is determined that no risk warning will be triggered. In this case, the embodiments of this application also provide the following steps 206-209 for optimization judgment, so as to make further auxiliary judgment based on the weight of the influence of multiple future unit times on "the next future unit time adjacent to the current time", thereby giving a more comprehensive risk warning result in this case.

[0119] Step 206: If it is determined that no risk warning will be triggered based on the probability distribution of the risk level corresponding to the next future unit time adjacent to the current time, then obtain the probability distribution of the risk level corresponding to a preset number of future unit times from the multiple future unit times in the risk prediction results, arranged from the closest to the farthest time from the current time.

[0120] Step 207: Assign different weights to a preset number of future unit moments based on their time distance from the current moment, from closest to furthest.

[0121] In this embodiment of the application, taking the probability distribution of risk levels corresponding to a limited number of future unit moments as an example, it should be noted that the limited number is used to limit some future moments of operation that are highly correlated with the drilling operation situation at the current moment. For example, drilling operations in the next hour will be affected by the operation effect at the current moment (i.e., in units of 10 minutes).

[0122] Therefore, regarding drilling operations at the "current moment," if the probability distribution of the risk level corresponding to the next future unit moment closest to the "current moment" determines that no risk level is triggered, this situation does not necessarily rule out the existence of early signs of risk and the degree of risk at the "current moment." Therefore, the embodiments of this application can further utilize the probability distribution of the risk level corresponding to a preset number of future unit moments predicted by the risk prediction model for a comprehensive analysis.

[0123] For example, assuming a 10-minute unit of time, regarding the "current moment," besides the "next future unit of time" closest to the "current moment," there are 5 more "future unit of time" within the next hour. For simplicity, these are labeled C1, C2, C3, C4, and C5 in chronological order. The probability distribution of the risk level (low risk, low-to-medium risk, medium-high risk, high risk) for each "future unit of time" is as follows: C1 corresponds to [c11, c12, c13, c14], C2 to [c21, c22, c23, c24], C3 to [c31, c32, c33, c34], C4 to [c41, c42, c43, c44], and C5 to [c51, c52, c53, c54]. c53, c54], and taking C1 as an example, it is explained as follows: In terms of the "current moment", apart from the "next future unit moment" which is closest to the "current moment", the probability distribution of the risk level corresponding to the first "future unit moment" is a probability vector [c11, c12, c13, c14], which correspond to low risk, medium-low risk, medium-high risk and high risk respectively.

[0124] Accordingly, in this embodiment, C1, C2, C3, C4, and C5 are assigned different weights according to their time sequence. The principle is that the closer to the "current moment", the higher the weight. That is, it is assumed that the weights of C1, C2, C3, C4, and C5 are d1, d2, d3, d4, and d5 respectively (and d1>d2>d3>d4>d5, and d1+d2+d3+d4+d5=1).

[0125] Step 208: Based on the weight allocation and the probability distribution of the risk level corresponding to the preset number of future unit time moments, re-integrate and process to obtain a new probability distribution corresponding to the risk level.

[0126] Step 209: Based on the new probability distribution corresponding to the risk level, re-determine whether to trigger a risk warning.

[0127] Continuing from the example in step 207 above, the embodiment of this application re-integrates and processes the following: based on the weights of C1, C2, C3, C4, and C5 respectively as d1, d2, d3, d4, and d5, and the corresponding values ​​for C1 ([c11, c12, c13, c14]), C2 ([c21, c22, c23, c24]), C3 ([c31, c32, c33, c34]), C4 ([c41, c42, c43, c44]), and C5 ([c51, c52, c53, c54]), for example, but not limited to, calculating a weighted average across the four risk levels to obtain a new probability distribution for the four risk levels, and then weighting this new probability distribution with the previously calculated "next future unit time closest to the 'current time'". The above is equivalent to adding the degree of risk prediction of C1, C2, C3, C4, and C5 to the "next future unit moment closest to the "current moment", and using this reintegration to obtain a new probability distribution corresponding to the risk level.

[0128] For example, in the case of "the probability distribution of the risk level corresponding to the next future unit time adjacent to the current time, which is a probability vector [0.1, 0.5, 0.3, 0.1]" as cited in example 205, if the new probability distribution obtained by integrating the risk probability distributions of multiple future units of time is [0.2, 0.45, 0.25, 0.1], it is still selected as "medium-low risk". This shows that even if the prediction is made not to trigger a risk warning based on the above "probability distribution of the risk level corresponding to the next future unit time", there may indeed be no early signs of leakage or even if there are, the degree is not high. The risk of leakage occurring over multiple future units of time is very low. Therefore, after integrating and analyzing multiple future times, it can be seen that there is indeed no need to trigger a risk warning at the current time, thereby improving the accuracy of risk warning identification.

[0129] In addition to steps 201-209, in some modified embodiments, the risk warning model of this application can also be periodically optimized and updated according to changes in the drilling operation environment. Specific implementation steps include:

[0130] Each time a risk warning is triggered, early leakage characteristic data corresponding to the leakage event is acquired; based on the early leakage characteristic data, the accuracy of decision-making by the risk prediction model is analyzed to obtain the analysis results of the risk prediction model; based on the analysis results, the risk prediction model is optimized at specified time intervals. For example, this application embodiment periodically reviews, dynamically adjusts, and continuously monitors and learns the "risk prediction model," as specifically explained below:

[0131] Establish a fixed review cycle (e.g., quarterly or semi-annually) to conduct a comprehensive review and update of the risk prediction model. This includes incorporating the latest scientific discoveries, technological advancements, and practical experience to achieve periodic reviews.

[0132] Within a specified time period, the risk prediction model is optimized based on the results of previous analysis. The optimization may involve algorithm improvements, parameter adjustments, and the integration of new datasets to achieve dynamic adjustment.

[0133] Implement a continuous monitoring system to track the performance of the optimized model and continuously learn from new data through machine learning techniques to achieve self-evolution and adaptive enhancement, thereby realizing continuous monitoring and learning.

[0134] As a response to the above Figure 1 , Figure 2 To implement the method shown, this application embodiment provides an early warning device for drilling leakage risk, serving as a virtual device for the execution subject of the aforementioned method. This device embodiment corresponds to the aforementioned method embodiment. For ease of reading, this device embodiment will not repeat the details of the aforementioned method embodiment, but it should be clear that the device in this embodiment is used to provide intelligent early warning for possible overflows and well leakage during drilling operations, such as... Figure 3 The device includes:

[0135] Acquisition unit 31 is used to acquire logging data at each unit time according to a preset time interval, wherein the time length between any two adjacent unit times is one preset time interval, and each unit time corresponds to multiple data acquisition points;

[0136] The first processing unit 32 is used to integrate and process the logging data collected at different time units according to the chronological order to obtain logging time series data.

[0137] The second processing unit 33 is used to extract target features associated with overflow and well leakage by performing multi-scale decomposition on the logging time series data. The target features include time domain features, frequency domain features and time-frequency domain features.

[0138] The third processing unit 34 is used to process the logging time series data containing the target features using a pre-trained risk prediction model and output a risk prediction result. The risk prediction model is a machine learning model trained using a self-attention mechanism, which is used to predict the risk of leakage at future times by capturing the long-range dependencies in the logging time series data.

[0139] The first execution unit 35 is used to determine whether to trigger a risk warning based on the risk prediction result.

[0140] Furthermore, such as Figure 4 As shown, the second processing unit 33 includes:

[0141] The preprocessing module 331 is used to preprocess the logging time series data to eliminate differences in different dimensions and obtain the preprocessed target logging time series data.

[0142] The first decomposition module 332 is used to perform empirical mode decomposition on the target logging time series data to obtain multiple intrinsic mode functions, wherein each intrinsic mode function represents a different frequency component, reflects the multi-scale characteristics of the signal, and is used to identify potential overflow and well leakage signals.

[0143] The second decomposition module 333 is used to perform variational mode decomposition on the multimodal signal containing frequency overlap features in the logging time series data to separate the various components of the multimodal signal and obtain multiple sub-signals.

[0144] The third decomposition module 334 is used to perform multi-scale decomposition on the logging time series data using wavelet transform to obtain frequency variation characteristics at different time scales.

[0145] The acquisition module 335 is used to fuse the decomposition results obtained by the empirical mode decomposition, the variational mode decomposition and the wavelet transform respectively to obtain a multidimensional feature vector, which includes time-domain features, frequency-domain features and time-frequency-domain features.

[0146] Furthermore, such as Figure 4 As shown, the third processing unit 34 includes:

[0147] Input module 341 is used to input the logging time series data containing the target features into the encoder of a pre-trained risk prediction model;

[0148] The conversion module 342 is used to convert the logging time series data containing the target features into a high-dimensional vector representation using the input embedding layer of the encoder;

[0149] Add module 343, used to add position encoding to the converted logging time series data to retain time series information;

[0150] Extraction module 344 is used to capture the dependency relationship between different positions in the transformed logging time series data according to the time series information using a self-attention mechanism, and extract global features;

[0151] Transformation module 345 is used to perform nonlinear transformation on the global features output by the autonomous force mechanism using a feedforward neural network to enhance the feature representation capability;

[0152] The prediction module 346 is used to predict the risk of well overflow based on the global features after nonlinear transformation using a classifier, and outputs the risk level or probability.

[0153] Furthermore, such as Figure 4 As shown, if the risk prediction result output by the risk prediction model is a probability distribution of risk levels at multiple future time units, the first execution unit 35 is specifically used for:

[0154] In multiple future time units, based on the probability distribution of the risk level corresponding to the next future time unit adjacent to the current time, it is determined whether to trigger a risk warning;

[0155] If it is determined that no risk warning will be triggered, then according to the time distance from the current time, the probability distribution of the risk level corresponding to a preset number of future unit times is obtained from multiple future unit times in the risk prediction results.

[0156] Based on the time distance from the current time, different weights are assigned to the preset number of future unit times, from closest to furthest.

[0157] Based on the probability distribution of risk levels corresponding to the weights and the preset number of future unit moments, a new probability distribution corresponding to the risk levels is obtained by re-integration and processing.

[0158] Based on the new probability distribution corresponding to the risk level, reassess whether to trigger a risk warning.

[0159] Furthermore, such as Figure 4 As shown, the device further includes:

[0160] The acquisition unit 36 ​​is used to acquire early leakage characteristic data corresponding to the leakage event each time a risk warning is triggered;

[0161] Analysis unit 37 is used to obtain the analysis results of the risk prediction model by analyzing the accuracy of the decision-making based on the early leakage characteristic data.

[0162] The second execution unit 38 is used to optimize the risk prediction model within a specified time period based on the analysis results.

[0163] This application provides an early warning device for drilling spill risk, including a processor and a memory. The acquisition unit, the first processing unit, the second processing unit, the third processing unit and the first execution unit are all stored in the memory as program units. The processor executes the program units stored in the memory to realize the corresponding functions.

[0164] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured. By adjusting kernel parameters, a smart early warning solution for overflows and wellbore leakage in drilling operations is constructed using multi-scale time-series data decomposition and a machine learning model trained with a self-attention mechanism.

[0165] In summary, this application provides a method and apparatus for early warning of drilling spill risks. By performing multi-scale decomposition on logging time-series data, this application extracts hidden, deep-level information from the data, specifically extracting key target features associated with spills and well leakage in the time, frequency, and time-frequency domains. This high-quality data is then input into a risk prediction model for processing. Utilizing the model's self-attention mechanism to capture long-range dependencies in the logging time-series data, a more accurate risk prediction result for future spills is obtained. Furthermore, a comprehensive analysis of the risk prediction results considers the probability distribution of risk levels in the next adjacent time unit or multiple future time units, thereby more comprehensively predicting whether a timely risk warning needs to be triggered. This application, through multi-scale time-series data decomposition and a machine learning model trained using a self-attention mechanism, provides an intelligent early warning solution for dealing with spills and well leakage in drilling operations, offering more reliable technical support for the safety management of drilling operations and reducing the likelihood of accidents.

[0166] This application provides a computer device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement steps such as a method for early warning of drilling spill risks.

[0167] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements steps such as a method for early warning of drilling spill risk.

[0168] This application also provides a computer program product, including a computer program that, when executed by a processor, implements steps such as a method for early warning of drilling spill risk.

[0169] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0170] In a typical configuration, the device includes one or more processors (CPUs), memory, and a bus. The device may also include input / output interfaces, network interfaces, etc.

[0171] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM, and memory includes at least one memory chip. Memory is an example of computer-readable media.

[0172] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0173] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0174] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0175] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method of alerting of a risk of blowout in a well, characterized in that, The method comprises: According to a preset time interval, collecting logging data at each unit time, wherein the time length between any two adjacent unit times is a preset time interval, and each unit time corresponds to a plurality of data collection points; According to the time sequence, the logging data collected at different unit times is integrated to obtain logging time sequence data; By multi-scale decomposition of the logging time sequence data, target features associated with overflow and lost circulation are extracted, including time domain features, frequency domain features and time-frequency domain features; The logging time sequence data containing the target features is processed using a pre-trained risk prediction model to output a risk prediction result; the risk prediction model is a machine learning model trained using a self-attention mechanism, which is used to predict the risk of overflow and lost circulation at future times by capturing long-range dependencies in the logging time sequence data; According to the risk prediction result, it is determined whether to trigger a risk warning; If the risk prediction result output by the risk prediction model is a probability distribution of risk levels at multiple future unit times, the determination of whether to trigger a risk warning according to the risk prediction result comprises: according to the probability distribution of the risk level corresponding to the next future unit time adjacent to the current time, it is determined whether to trigger a risk warning; if it is determined not to trigger a risk warning according to the probability distribution of the risk level corresponding to the next future unit time adjacent to the current time, then according to the time distance from the current time, the probability distribution of the risk level corresponding to a preset number of future unit times is obtained from the multiple future unit times in the risk prediction result; according to the time distance from the current time, different weights are assigned to the preset number of future unit times; according to the weights and the probability distribution of the risk level corresponding to the preset number of future unit times, a new probability distribution corresponding to the risk level is obtained by re-integrating; according to the new probability distribution corresponding to the risk level, it is determined whether to trigger a risk warning.

2. The method of claim 1, wherein, The target features associated with overflow and lost circulation are extracted by multi-scale decomposition of the logging time sequence data, comprising: Pretreatment of the logging time sequence data is performed to eliminate differences in different dimensions to obtain pretreated target logging time sequence data; The target logging time sequence data is subjected to empirical mode decomposition to obtain a plurality of intrinsic mode functions, each intrinsic mode function representing a different frequency component and reflecting the multi-scale characteristics of the signal, which is used to identify potential overflow and lost circulation signals; The signals with overlapping frequency components in the logging time sequence data are subjected to variational mode decomposition to separate the individual components of the multi-modal signals to obtain a plurality of sub-signals; Wavelet transform is used to perform multi-scale decomposition of the logging time sequence data to obtain frequency variation characteristics at different time scales; The decomposition results obtained by the empirical mode decomposition, the variational mode decomposition, and the wavelet transform are fused to obtain a multidimensional feature vector, which includes time-domain features, frequency-domain features, and time-frequency-domain features.

3. The method of claim 1, wherein, The process of using a pre-trained risk prediction model to process the logging time-series data containing the target features and outputting risk prediction results includes: The logging time series data containing the target features is input into the encoder of a pre-trained risk prediction model; The well logging time series data containing the target features is converted into a high-dimensional vector representation using the input embedding layer of the encoder; Location codes are added to the converted logging time series data to preserve time series information; Based on the time series information, a self-attention mechanism is used to capture the dependencies between different positions in the transformed logging time series data and extract global features. The global features output by the autonomous force mechanism are nonlinearly transformed using a feedforward neural network to enhance the feature representation capability. The classifier is used to predict the risk of well overflow based on the extracted global features, and the risk level or probability is output.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Each time a risk warning is triggered, acquire the early leakage characteristic data corresponding to the leakage event; Based on the early leakage characteristic data, analyze the accuracy of the risk prediction model in making decisions; Based on the analyzed data, the risk prediction model is optimized over a specified time period.

5. A device for early warning of well overflow risk, characterized in that, The device includes: The acquisition unit is used to acquire logging data at each unit time according to a preset time interval, wherein the time length between any two adjacent unit times is one preset time interval, and each unit time corresponds to multiple data acquisition points; The first processing unit is used to integrate and process the logging data collected at different time units according to the chronological order to obtain logging time series data. The second processing unit is used to extract target features associated with overflow and well leakage by performing multi-scale decomposition on the logging time series data. The target features include time domain features, frequency domain features, and time-frequency domain features. The third processing unit is used to process the logging time series data containing the target features using a pre-trained risk prediction model and output the risk prediction result; the risk prediction model is a machine learning model trained using a self-attention mechanism, which is used to predict the overflow risk at future times by capturing the long-range dependencies in the logging time series data. The first execution unit is used to determine whether to trigger a risk warning based on the risk prediction result; The first execution unit is further configured to: if the risk prediction result output by the risk prediction model is a probability distribution of risk levels at multiple future time units, determine whether to trigger a risk warning according to a probability distribution of a risk level corresponding to a next future time unit adjacent to the current time unit in the multiple future time units; if it is determined not to trigger the risk warning according to the probability distribution of the risk level corresponding to the next future time unit adjacent to the current time unit, obtain, from the risk prediction result, a probability distribution of a risk level corresponding to a preset number of future time units in the multiple future time units in a time distance from the current time unit; assign different weights to the preset number of future time units in a time distance from the current time unit; re-integrate the weights and the probability distributions of the risk levels corresponding to the preset number of future time units to obtain a new probability distribution corresponding to a risk level; and determine whether to trigger the risk warning according to the new probability distribution corresponding to the risk level.

6. The apparatus of claim 5, wherein, The second processing unit comprises: a preprocessing module configured to preprocess the logging time series data to eliminate differences in different dimensions and obtain target logging time series data after preprocessing; a first decomposition module configured to perform empirical mode decomposition on the target logging time series data to obtain a plurality of intrinsic mode functions, each of which represents a different frequency component and reflects a multi-scale feature of a signal, and is used to identify potential overflow and lost circulation signals; a second decomposition module configured to perform variational mode decomposition on signals with overlapping frequency components in the logging time series data to separate each component of a multi-modal signal and obtain a plurality of sub-signals; a third decomposition module configured to perform multi-scale decomposition on the logging time series data using wavelet transform to obtain frequency variation characteristics at different time scales; an acquisition module configured to fuse the decomposition results obtained by the empirical mode decomposition, the variational mode decomposition and the wavelet transform to obtain a multi-dimensional feature vector, wherein the multi-dimensional feature vector comprises time domain features, frequency domain features and time-frequency domain features.

7. A computer device comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program comprises instructions that, when executed by the processor, cause the processor to perform the method of any one of claims 1-6. The processor executes the computer program to implement the steps of the drilling overflow and lost circulation risk warning method of any one of claims 1-4.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the drilling overflow and lost circulation risk warning method of any one of claims 1-4.

9. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the drilling overflow and lost circulation risk warning method of any one of claims 1-4.

Images