A method and related apparatus for predicting drilling risk based on geophysical data

Abstract

The application discloses a drilling risk prediction method based on geophysical data and related equipment, and the method comprises the following steps: collecting geophysical parameters in a seabed drilling process in real time by using a logging while drilling device; preprocessing the geophysical parameters to generate target logging data; constructing the target logging data into a spatial sequence with depth as an axis, training a dynamic extrapolation model based on the spatial sequence of the drilled stratum by using a preset time sequence prediction algorithm; inputting the real-time target logging data into the dynamic extrapolation model to predict stratum parameters of the un-drilled stratum within a preset distance in front of the drill bit; and coupling the stratum parameters with a preset geomechanical model to generate a dynamic safe drilling window containing a safe range of drilling parameters. The application can effectively predict the geological anomaly in front, significantly improves the safety and operation efficiency of seabed oil and gas drilling, and can be widely applied to the technical field of seabed oil and gas resource exploration and development.

Description

Technical Field

[0001] This application relates to the field of subsea oil and gas resource exploration and development technology, and in particular to a drilling risk prediction method and related equipment based on geophysical data. Background Technology

[0002] In the field of oil and gas exploration and development, subsea drilling has become an important direction due to its abundant oil and gas resources. Logging while drilling (LWD) technology, as a core means of acquiring real-time information about the seabed formation, has been widely applied in subsea drilling operations. This technology can simultaneously collect geophysical parameters such as gamma ray, resistivity, and spontaneous potential during drilling, providing a basis for determining the lithology, porosity, and stress state of the drilled formations. It assists engineers in adjusting drilling parameters and is of great significance for ensuring the safety of subsea drilling.

[0003] However, the subsea drilling environment is unique and complex. Risks such as wellbore collapse and leakage seriously threaten the safety and efficiency of operations. Existing logging-while-drilling technology has significant limitations: the data it collects can only reflect formation information where the drill bit has "already" or "just passed through." Predicting geological anomalies in un-drilled or un-drilled areas relies heavily on the human experience of technical personnel, which is highly dependent on their expertise and experience and may involve subjective judgment, resulting in poor prediction accuracy. Summary of the Invention

[0004] The main objective of this application is to propose a drilling risk prediction method, system, electronic device, storage medium, and program product based on geophysical data, aiming to solve at least one problem of the prior art.

[0005] To achieve the above objectives, one aspect of this application proposes a drilling risk prediction method based on geophysical data, the method comprising:

[0006] Geophysical parameters during the seabed drilling process are collected in real time using logging-while-drilling equipment. These geophysical parameters correspond to depth information and include logging curve data.

[0007] Geophysical parameters are preprocessed to generate target well logging data;

[0008] The target logging data is constructed as a spatial sequence with depth as the axis. Based on the spatial sequence of drilled formations, a dynamic extrapolation model is trained and generated using a preset time series prediction algorithm.

[0009] Real-time target logging data is input into a dynamic extrapolation model to predict formation parameters of un-drilled formations within a preset distance in front of the drill bit.

[0010] The formation parameters are coupled with a preset geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters.

[0011] In some embodiments, the geophysical parameters further include gamma and resistivity. Preprocessing the geophysical parameters includes the following steps:

[0012] Data cleaning is performed on the logging curve data to remove outliers caused by equipment failure, thus obtaining the first logging data;

[0013] The wavelet transform denoising method was used to denoise the first logging data to remove random noise from the data, thus obtaining the second logging data.

[0014] The gamma, resistivity, and second logging data of different units and dimensions are uniformly transformed to a preset numerical range to obtain the target logging data.

[0015] In some embodiments, a dynamic extrapolation model is generated by training a preset time series prediction algorithm based on the spatial sequence of drilled strata, including the following steps:

[0016] Configure the time series forecasting algorithm as a long short-term memory network;

[0017] The spatial sequence at the nearest first preset depth is obtained from the drilled strata as a training sample. The training sample includes the first spatial sequence of the first segment and the second spatial sequence of the second segment at the first preset depth. The length of the first spatial sequence is greater than the length of the second spatial sequence. The first spatial sequence is used as input data, and the second spatial sequence and its corresponding strata parameters are used as output labels.

[0018] The Long Short-Term Memory (LSTM) network is trained using training samples. By adjusting the number of hidden layers, the number of neurons, and the learning rate of the LSM network, a dynamic extrapolation model is obtained until the prediction error of the LSM network is less than a preset error threshold.

[0019] In some embodiments, the method further includes the following steps:

[0020] Use the end depth of the first preset depth as the target depth;

[0021] To obtain the drilling depth at the target depth in seabed drilling;

[0022] When the drilling depth reaches the target depth, the spatial sequence corresponding to the target distance is used as a new training sample; historical data of spatial sequences that exceed the target distance are discarded through a forgetting mechanism.

[0023] The dynamic extrapolation model is fine-tuned using new training samples, and then updated based on the results of the fine-tuning.

[0024] Using the final depth of the target distance as the target depth, return to the step of obtaining the drilling depth of the seabed drilling distance to the target depth, and continuously update and adjust the dynamic extrapolation model.

[0025] In some embodiments, real-time target logging data is input into a dynamic extrapolation model to predict formation parameters of undrilled formations within a preset distance in front of the drill bit, including the following steps:

[0026] The input data is obtained from the target logging data at the nearest second preset depth in the drilled formation.

[0027] Input data is fed into a dynamic extrapolation model to predict formation parameters of un-drilled formations within a preset distance in front of the drill bit.

[0028] The training samples for generating the dynamic extrapolation model include a first spatial sequence and a second spatial sequence at the first preset depth of the nearest drilled stratum. The first spatial sequence is used as the training input data, and the second spatial sequence and its corresponding stratum parameters are used as the output labels. The preset distance is the same as the depth range corresponding to the second spatial sequence.

[0029] In some embodiments, coupling formation parameters with a preset geomechanical model to generate a dynamic safe drilling window that includes a safe range for drilling parameters includes the following steps:

[0030] The formation parameters are input into the geomechanical model, and the safe range of drilling fluid density and drilling speed corresponding to the formation parameters is determined by the analysis of the geomechanical model.

[0031] Among them, the stratigraphic parameters include stratigraphic lithology, porosity values ​​and stress state distribution, and the geomechanical model is pre-constructed based on the geostress and rock mechanical properties of the submarine strata;

[0032] The safe range of drilling fluid density and drilling speed is combined to form a dynamic safe drilling window.

[0033] In some embodiments, the method further includes the following steps:

[0034] By pre-calibrating the compressive strength and fracture pressure threshold of different lithological strata through rock mechanics experiments, establishing the mapping relationship between porosity and stratum permeability, and integrating the geostress field distribution data of the seabed area, a geomechanical model was constructed.

[0035] In the geomechanical model, formation lithology is mapped to corresponding rock strength parameters, porosity values ​​are converted into permeability indices, and stress state distribution is analyzed as the direction and magnitude of principal stresses within the formation. The safe range of drilling fluid density is determined by comparing the difference between predicted formation pore pressure and fracture pressure, and the safe range of drilling speed is determined by the dynamic balance between rock compressive strength and drill bit cutting force.

[0036] To achieve the above objectives, another aspect of this application proposes a drilling risk prediction system based on geophysical data, the system comprising:

[0037] The data acquisition module is used to collect geophysical parameters in real time during the seabed drilling process using logging-while-drilling equipment. The geophysical parameters correspond to depth information and include logging curve data.

[0038] The data preprocessing module is used to preprocess geophysical parameters and generate target well logging data;

[0039] The model training module is used to construct the target logging data into a spatial sequence with depth as the axis, and to train and generate a dynamic extrapolation model based on the spatial sequence of drilled formations using a preset time series prediction algorithm.

[0040] The prediction module is used to input real-time target logging data into a dynamic extrapolation model to predict formation parameters of un-drilled formations within a preset distance in front of the drill bit.

[0041] The coupling analysis module is used to couple formation parameters with a preset geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters.

[0042] In some embodiments, the system further includes a model update module for performing the following operations:

[0043] Use the end depth of the first preset depth as the target depth;

[0044] To obtain the drilling depth at the target depth in seabed drilling;

[0045] When the drilling depth reaches the target depth, the spatial sequence corresponding to the target distance is used as a new training sample; historical data of spatial sequences that exceed the target distance are discarded through a forgetting mechanism.

[0046] The dynamic extrapolation model is fine-tuned using new training samples, and then updated based on the results of the fine-tuning.

[0047] Using the final depth of the target distance as the target depth, return to the step of obtaining the drilling depth of the seabed drilling distance to the target depth, and continuously update and adjust the dynamic extrapolation model.

[0048] In some embodiments, the system further includes a model building module for performing the following operations:

[0049] By pre-calibrating the compressive strength and fracture pressure threshold of different lithological strata through rock mechanics experiments, establishing the mapping relationship between porosity and stratum permeability, and integrating the geostress field distribution data of the seabed area, a geomechanical model was constructed.

[0050] In the geomechanical model, formation lithology is mapped to corresponding rock strength parameters, porosity values ​​are converted into permeability indices, and stress state distribution is analyzed as the direction and magnitude of principal stresses within the formation. The safe range of drilling fluid density is determined by comparing the difference between predicted formation pore pressure and fracture pressure, and the safe range of drilling speed is determined by the dynamic balance between rock compressive strength and drill bit cutting force.

[0051] To achieve the above objectives, another aspect of the embodiments of this application proposes an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned method.

[0052] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method.

[0053] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method.

[0054] The embodiments of this application include at least the following beneficial effects: This application provides a drilling risk prediction method, system, electronic device, storage medium, and program product based on geophysical data. This solution utilizes logging-while-drilling equipment to collect geophysical parameters in real time during the subsea drilling process. The geophysical parameters correspond to depth information and include logging curve data. The geophysical parameters are preprocessed to generate target logging data. The target logging data is constructed into a spatial sequence with depth as the axis. Based on the spatial sequence of drilled formations, a dynamic extrapolation model is trained using a preset time series prediction algorithm. The real-time target logging data is input into the dynamic extrapolation model to predict the formation parameters of undrilled formations within a preset distance in front of the drill bit. The formation parameters are coupled with a preset geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters. This application embodiment acquires geophysical parameters in real time, constructs spatial sequences through preprocessing, and then trains a dynamic extrapolation model. Based on the data association mapping relationship learned by the model, it can accurately predict the formation parameters of the un-drilled formation ahead of the drill bit. In addition, it combines the geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters, which can provide a basis for parameter adjustment and thus avoid risks such as wellbore collapse and leakage in advance. It effectively solves the problem that the existing logging-while-drilling technology can only reflect the formation information that the drill bit "has been" or "just passed", and has insufficient ability to predict geological anomalies ahead, which significantly improves the safety and operational efficiency of subsea oil and gas drilling. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of an implementation environment for a drilling risk prediction method based on geophysical data provided in this application embodiment;

[0056] Figure 2 This is a flowchart illustrating a drilling risk prediction method based on geophysical data provided in an embodiment of this application.

[0057] Figure 3 This is a schematic diagram illustrating the unfolded process of step S200 provided in the embodiments of this application;

[0058] Figure 4 This is a schematic diagram of the unfolding process for generating a dynamic extrapolation model provided in an embodiment of this application;

[0059] Figure 5 This is a schematic diagram of the extended process of the updated dynamic extrapolation model provided in the embodiments of this application;

[0060] Figure 6 This is a schematic diagram illustrating the unfolded process of step S400 provided in the embodiments of this application;

[0061] Figure 7 This is a schematic diagram illustrating the unfolded process of step S500 provided in the embodiments of this application;

[0062] Figure 8 This is a schematic diagram illustrating a structural example of a drilling risk prediction system based on geophysical data provided in an embodiment of this application;

[0063] Figure 9 This is a schematic diagram of another structural example of the drilling risk prediction system based on geophysical data provided in the embodiments of this application;

[0064] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of systems and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0066] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0067] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0069] Among related technologies, existing logging-while-drilling (LOD) technology has significant limitations: the data it collects can only reflect formation information where the drill bit has "already" or "just passed through." Predicting geological anomalies in un-drilled or un-drilled areas relies heavily on the human experience of technical personnel, heavily depending on their expertise and experience, and potentially involving subjective judgment, resulting in poor prediction accuracy.

[0070] In view of this, this application provides a drilling risk prediction method and related equipment based on geophysical data. This method utilizes logging-while-drilling equipment to collect geophysical parameters in real time during the subsea drilling process. These geophysical parameters correspond to depth information and include logging curve data. The geophysical parameters are preprocessed to generate target logging data. The target logging data is constructed into a spatial sequence with depth as the axis. Based on the spatial sequence of drilled formations, a dynamic extrapolation model is trained using a preset time series prediction algorithm. The real-time target logging data is input into the dynamic extrapolation model to predict the formation parameters of undrilled formations within a preset distance ahead of the drill bit. The formation parameters are coupled with a preset geomechanical model to generate a dynamic safe drilling window that includes a safe range for drilling parameters. This application embodiment acquires geophysical parameters in real time, constructs spatial sequences through preprocessing, and then trains a dynamic extrapolation model. Based on the data association mapping relationship learned by the model, it can accurately predict the formation parameters of the un-drilled formation ahead of the drill bit. In addition, it combines the geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters, which can provide a basis for parameter adjustment and thus avoid risks such as wellbore collapse and leakage in advance. It effectively solves the problem that the existing logging-while-drilling technology can only reflect the formation information that the drill bit "has been" or "just passed", and has insufficient ability to predict geological anomalies ahead, which significantly improves the safety and operational efficiency of subsea oil and gas drilling.

[0071] It is understood that the drilling risk prediction method based on geophysical data provided in this application can be applied to any computer device with data processing and computing capabilities, and this computer device can be various types of terminals or servers. When the computer device in the embodiment is a server, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the terminal can be a smartphone, tablet, laptop, or desktop computer, but it is not limited to these.

[0072] like Figure 1The diagram shown is a schematic representation of an implementation environment provided in an embodiment of this application. (Refer to...) Figure 1 The implementation environment includes at least one terminal 102 and a server 101. The terminal 102 and the server 101 can be connected via a network, either wirelessly or via a wired connection, to complete data transmission and exchange.

[0073] Server 101 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0074] Additionally, server 101 can also be a node server in a blockchain network. Blockchain is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms.

[0075] Terminal 102 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. Terminal 102 and server 101 can be directly or indirectly connected via wired or wireless communication, and this embodiment of the application does not impose any limitations.

[0076] For example, based on Figure 1 The implementation environment shown in this application embodiment provides a drilling risk prediction method based on geophysical data. The following description uses the application of the drilling risk prediction method based on geophysical data in server 101 as an example. It can be understood that the drilling risk prediction method based on geophysical data can also be applied in terminal 102.

[0077] Reference Figure 2 , Figure 2 This is an optional flowchart of the drilling risk prediction method based on geophysical data provided in the embodiments of this application. The subject executing the drilling risk prediction method based on geophysical data can be any of the aforementioned computer devices (including servers or terminals). Figure 2 The method may include, but is not limited to, steps S100 to S500.

[0078] Step S100: Use logging-while-drilling equipment to collect geophysical parameters in real time during the seabed drilling process;

[0079] Among them, geophysical parameters correspond to depth information, and geophysical parameters include well logging curve data;

[0080] For example, in some specific implementations, geophysical parameters, including gamma ray, resistivity, and spontaneous potential logging curves, are first acquired in real time during the subsea drilling process using logging-while-drilling equipment. These data correspond to depth information and can reflect the physical characteristics of the formation. Specifically, the real-time acquisition of geophysical parameters corresponding to depth information during subsea drilling by logging-while-drilling equipment refers to the simultaneous measurement and recording of gamma ray, resistivity, and spontaneous potential data related to formation depth using logging-while-drilling tools. This can be achieved using a gamma ray detector, resistivity sensor, and spontaneous potential electrode mounted on the logging-while-drilling tool, thereby acquiring real-time formation information around the drill bit to support subsequent predictive analysis.

[0081] Step S200: Preprocess the geophysical parameters to generate target well logging data;

[0082] It should be noted that geophysical parameters also include gamma and resistivity, as in some embodiments, such as Figure 3 As shown, step S200 may include the following steps: S210, cleaning the logging curve data to remove outliers caused by equipment failure, and obtaining the first logging data; S220, using wavelet transform denoising method to denoise the first logging data to remove random noise in the data, and obtaining the second logging data; S230, uniformly transforming the gamma, resistivity and second logging data of different units and dimensions to a preset numerical range to obtain the target logging data.

[0083] For example, in some specific implementations, the acquired well logging curve data is preprocessed. The preprocessing process includes data cleaning, denoising, and standardization, with the aim of generating high-quality data that meets modeling requirements. Preprocessed data is more reliable and beneficial for subsequent model training and prediction. Specifically, generating data that meets modeling requirements through preprocessing involves outlier removal, noise removal, and standardization transformation of the original well logging curves. This can be achieved using data cleaning algorithms, wavelet transform denoising methods, and standardization formulas to eliminate equipment interference and dimensional differences, thereby improving data quality.

[0084] Step S300: Construct the target logging data into a spatial sequence with depth as the axis, and use a preset time series prediction algorithm to train and generate a dynamic extrapolation model based on the spatial sequence of the drilled formation.

[0085] Among them, the spatial sequence with depth as the axis refers to arranging well logging curve data in order of formation depth to form a continuous sequence. Specifically, it can be constructed using depth alignment and sliding window methods, which facilitates the analysis of the variation of formation parameters with depth using time series algorithms.

[0086] It should be noted that in some embodiments, such as Figure 4As shown, a dynamic extrapolation model is generated by training a preset time series prediction algorithm based on the spatial sequence of drilled strata. This can include the following steps: S310, configuring the time series prediction algorithm as a long short-term memory (LSTM) network; S320, obtaining the spatial sequence at the nearest first preset depth from the drilled strata as training samples. The training samples include a first spatial sequence at the beginning of the first preset depth and a second spatial sequence at the end, with the length of the first spatial sequence being greater than the length of the second spatial sequence. The first spatial sequence serves as input data, and the second spatial sequence and its corresponding strata parameters serve as output labels; S330, training the LTM network using the training samples by adjusting the number of hidden layers, the number of neurons, and the learning rate until the prediction error of the LTM network is less than a preset error threshold, thus obtaining the dynamic extrapolation model.

[0087] It should be noted that, in some embodiments, when training a Long Short-Term Memory (LSTM) network using training samples, the network's data processing logic may include the following steps: initializing the hidden states and cell states of the LSM network before training; calculating the target hidden state corresponding to the input data using the initialized LSM network and the trained model parameters; and determining the output data based on the target hidden state.

[0088] For example, in some specific implementations, the time series prediction algorithm training to generate a dynamic extrapolation model refers to using algorithms such as long short-term memory networks to model drilled strata data. Specifically, the model can be optimized by adjusting the number of network layers, the number of neurons, and the learning rate parameters to achieve the ability to dynamically update the parameters of the strata ahead based on historical data.

[0089] Specifically, training a Long Short-Term Memory (LSTM) network can be achieved by constructing a spatial sequence with depth as the axis after preprocessing the well logging data (i.e., the target well logging data). In this constructed spatial sequence with depth as the axis, each depth point corresponds to a three-dimensional vector containing gamma values, resistivity values, and spontaneous potential values ​​(in API, Ω•m, and mV, respectively). The 45-meter sequence input to the dynamic extrapolation model is a 45-meter time step, with each time step corresponding to a three-dimensional vector (i.e., dimension 45×3), where 45 represents the number of depth points (interval 0.1 meters), and 3 represents the three geophysical parameters. This step correlates formation parameter variations with depth, giving the data spatial continuity. Based on these spatial sequences, a pre-defined time series prediction algorithm is used to train the drilled formation data, generating a dynamic extrapolation model. This model can capture the patterns of formation parameter variations with depth. For example, a spatial sequence of drilled strata from the past 50 meters can be selected as a training sample; wherein, the spatial sequence of the first 45 meters is set as the historical data sequence, and the spatial sequence of the last 5 meters and the corresponding stratum parameters are set as the target data sequence.

[0090] It should also be noted that in some embodiments, such as Figure 5 As shown, the method may further include the following steps: T100, taking the end depth of the first preset depth as the target depth; T200, obtaining the drilling depth of the seabed drilling from the target depth; T300, when the drilling depth reaches the target distance, taking the spatial sequence corresponding to the target distance as a new training sample; wherein, historical data of spatial sequences exceeding the target distance are discarded through a forgetting mechanism; T400, using the new training sample to fine-tune the dynamic extrapolation model, and updating the dynamic extrapolation model based on the fine-tuning result; T500, taking the end depth of the target distance as the target depth, returning to execute the step of obtaining the drilling depth of the seabed drilling from the target depth, and continuously updating and adjusting the dynamic extrapolation model.

[0091] For example, in some specific implementations, the dynamic extrapolation model is continuously updated during drilling, rather than being trained once and used permanently. The update is triggered automatically every 5 meters of drilling depth or when a new 5-meter logging data segment is received. A sliding window mechanism is used to determine the training data range during updates, meaning only the most recently drilled 50 meters of formation data (covering the 50 meters before the current drill bit position) is used, and historical data exceeding this range is discarded through a forgetting mechanism to ensure data timeliness. The update operation employs incremental learning, fine-tuning the existing model weights using new data (reducing the learning rate to 1 / 10 of the initial value) rather than completely retraining, to reduce computational time and maintain model stability.

[0092] Step S400: Input the real-time target logging data into the dynamic extrapolation model to predict the formation parameters of the un-drilled formation within a preset distance in front of the drill bit.

[0093] It should be noted that in some embodiments, such as Figure 6 As shown, step S400 may include the following steps: S410, obtaining the target logging data at the nearest second preset depth from the drilled formation as input data; S420, inputting the input data into a dynamic extrapolation model to predict the formation parameters of the un-drilled formation within a preset distance in front of the drill bit; wherein, the training samples for training the dynamic extrapolation model include a first spatial sequence of the first segment and a second spatial sequence of the latter segment of the nearest first preset depth in the drilled formation, the first spatial sequence as training input data, the second spatial sequence and its corresponding formation parameters as output labels, and the preset distance is the same as the depth range corresponding to the second spatial sequence.

[0094] For example, in some specific implementations, during drilling, real-time acquired and preprocessed logging curve data are input into a dynamic extrapolation model. Based on the input data, the model predicts the formation parameters of un-drilled formations within a preset distance ahead of the drill bit. This step enables the prediction of unknown formation characteristics, providing a basis for subsequent decision-making. Predicting the formation parameters of un-drilled formations within a preset distance ahead of the drill bit refers to using the dynamic extrapolation model to output the lithology, porosity values, and stress state distribution of the formation within a range of one to five meters ahead. Specifically, this can be achieved by inputting real-time logging data to trigger forward calculations in the model, identifying geological anomalies that may lead to wellbore collapse or leakage in advance.

[0095] Step S500: Couple the formation parameters with the preset geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters;

[0096] It should be noted that in some embodiments, such as Figure 7 As shown, step S500 may include the following steps: S510, inputting formation parameters into the geomechanical model, and determining the safe range of drilling fluid density and drilling speed corresponding to the formation parameters through analysis of the geomechanical model; wherein, the formation parameters include formation lithology, porosity values ​​and stress state distribution, and the geomechanical model is pre-constructed based on the geostress and rock mechanical properties of the seafloor strata; S520, combining the safe range of drilling fluid density and drilling speed to form a dynamic safe drilling window.

[0097] For example, in some specific implementations, the predicted formation lithology, porosity values, and stress state distribution can be input into a geomechanical model. This geomechanical model is pre-constructed and takes into account the geostress and rock mechanical properties of the submarine strata. Through calculation and analysis using the geomechanical model, the safe ranges for drilling fluid density and drilling speed corresponding to the predicted formation parameters are determined. These safe ranges for drilling fluid density and drilling speed are then combined to form a dynamic safe drilling window. Specifically, the dynamic safe drilling window generated by coupling the geomechanical model refers to inputting the predicted parameters into a pre-established mechanical model to calculate the safe ranges for drilling fluid density and drilling speed. This can be achieved using finite element analysis or rock mechanics equations, and the quantified safety boundaries guide real-time adjustments to drilling parameters.

[0098] It should be noted that, in some embodiments, the method may further include the following steps: pre-calibrating the compressive strength and fracturing pressure thresholds of different lithological strata through rock mechanics experiments, establishing a mapping relationship between porosity and formation permeability, and integrating the geostress field distribution data of the seabed area to construct a geomechanical model; wherein, in the geomechanical model, the lithology of the strata is mapped to the corresponding rock strength parameters, the porosity value is converted into a permeability index, and the stress state distribution is analyzed as the direction and magnitude of the principal stress within the strata; the safe range of drilling fluid density is determined by comparing the difference between the predicted formation pore pressure and the fracturing pressure, and the safe range of drilling speed is determined by the dynamic balance relationship between the rock compressive strength and the drill bit cutting force.

[0099] Specifically, this involves coupling the predicted formation parameters with a pre-defined geomechanical model. Through this coupling analysis, a dynamic safe drilling window, encompassing safe ranges for drilling parameters, is calculated and generated. This safe window provides real-time guidance for parameter adjustments during drilling operations, helping to avoid potential drilling risks.

[0100] For example, in some specific implementations, specifically when the formation lithology is shale, the rock strength parameter is set to the shear strength value under low permeability conditions, the porosity value is converted to the permeability coefficient using Darcy's formula, and the stress state distribution is decomposed into vertical stress and horizontal stress components. The geomechanical model superimposes the vertical stress and pore pressure to calculate the lower limit of formation collapse pressure, while comparing the horizontal stress with the rock tensile strength to determine the upper limit of formation fracturing pressure. The drilling fluid density is limited between the collapse pressure and the fracturing pressure to avoid wellbore instability. The drilling rate is dynamically adjusted based on real-time feedback of the shale compressive strength and drill bit torque to ensure that the drilling pressure does not exceed the rock bearing capacity limit. By synchronously inputting formation lithology, porosity data, and stress state distribution parameters into the geomechanical model, the linkage control of drilling fluid density and drilling rate is achieved, forming a real-time updated safe operating range.

[0101] To explain in detail the principles of the technical solution of this application, the overall process of this application will be described below with reference to some specific embodiments. It is easy to understand that the following is an explanation of the technical principles of this application and should not be regarded as a limitation of this application.

[0102] First, it should be noted that, in response to the shortcomings of existing technologies, this application provides a method and system for predicting drilling risks using geophysical data. This solves the problem that existing technologies, compared to logging-while-drilling data which typically reflects formation information where the drill bit is "already" or "just passed", are insufficient in predicting sudden geological anomalies within a few meters ahead.

[0103] In some specific embodiments, this application provides a method for predicting drilling risks using geophysical data. The method includes: using logging-while-drilling equipment to collect geophysical parameters corresponding to depth information during subsea drilling in real time, including gamma, resistivity, and spontaneous potential logging curve data; preprocessing the collected logging curve data to generate data that meets modeling requirements; constructing a spatial sequence with depth as the axis from the preprocessed logging curve data, and using a pre-set time series prediction algorithm to train and generate a dynamic extrapolation model based on the spatial sequence of drilled formations; inputting the real-time collected and preprocessed logging curve data into the dynamic extrapolation model to predict formation parameters of undrilled formations within a preset distance in front of the drill bit; and coupling the predicted formation parameters with a pre-set geomechanical model to calculate and generate a dynamic safe drilling window that includes a safe range of drilling parameters.

[0104] The real-time acquisition of geophysical parameters corresponding to depth information during subsea drilling by logging-while-drilling equipment refers to the simultaneous measurement and recording of gamma, resistivity, and spontaneous potential data related to formation depth using logging-while-drilling tools. This can be achieved using gamma-ray detectors, resistivity sensors, and spontaneous potential electrodes mounted on the logging-while-drilling tool, acquiring real-time formation information around the drill bit to support subsequent predictive analysis. Preprocessing to generate data that meets modeling requirements involves outlier removal, noise reduction, and standardization of the original logging curves. This can be achieved using data cleaning algorithms, wavelet transform denoising methods, and standardization formulas to eliminate equipment interference and dimensional differences, thereby improving data quality. The spatial sequence based on depth refers to arranging logging curve data in order of formation depth to form a continuous sequence. This can be constructed using depth alignment and sliding window methods, facilitating the analysis of formation parameter variations with depth using time-series algorithms. The time-series prediction algorithm training to generate a dynamic extrapolation model refers to using algorithms such as Long Short-Term Memory (LSTM) networks to model drilled formation data. Specifically, the model can be optimized by adjusting the number of network layers, neurons, and learning rate parameters to achieve dynamic updates of formation parameters based on historical data. Predicting formation parameters of un-drilled formations within a preset distance ahead of the drill bit involves using the dynamic extrapolation model to output the lithology, porosity, and stress distribution of formations within a range of one to five meters ahead. This can be achieved by inputting real-time logging data to trigger forward calculations, identifying potential geological anomalies that could lead to wellbore collapse or leakage. The geomechanical model coupling to generate a dynamic safe drilling window involves inputting predicted parameters into a pre-established mechanical model to calculate the safe range for drilling fluid density and drilling speed. This can be achieved using finite element analysis or rock mechanics equations, quantifying safety boundaries to guide real-time drilling parameter adjustments.

[0105] The core innovation of this application lies in combining real-time logging-while-drilling data with a geomechanical model through a dynamic extrapolation model to form a closed-loop system that couples depth sequence prediction with mechanical parameters. This breaks through the limitation of existing technologies that can only analyze drilled formations, enabling accurate prediction of formation parameters at short distances in front of the drill bit, and generating a dynamic safe drilling window to proactively avoid risks.

[0106] The working process and principle of this application are as follows: First, geophysical parameters, including gamma ray, resistivity, and spontaneous potential logging curves, are collected in real time during the subsea drilling process using logging-while-drilling equipment. These data correspond to depth information and can reflect the physical characteristics of the formation.

[0107] Next, the collected well logging data is preprocessed. The preprocessing process includes data cleaning, denoising, and standardization, with the aim of generating high-quality data that meets the modeling requirements. Preprocessed data is more reliable and beneficial for subsequent model training and prediction.

[0108] Then, the preprocessed well logging data is constructed into a spatial sequence with depth as the axis. In the constructed 'spatial sequence with depth as the axis,' each depth point corresponds to a three-dimensional vector containing gamma value, resistivity value, and spontaneous potential value (units API, Ω•m, and mV, respectively). The 45-meter sequence input into the dynamic extrapolation model is a time step of length 45, with each time step corresponding to a three-dimensional vector (i.e., dimension 45×3), where 45 represents the number of depth points (interval 0.1 meters), and 3 represents the three geophysical parameters. This step correlates formation parameter variations with depth, giving the data spatial continuity. Based on these spatial sequences, a pre-defined time series prediction algorithm is used to train the drilled formation data to generate a dynamic extrapolation model. This model can capture the patterns of formation parameter variations with depth.

[0109] During drilling, real-time acquired and preprocessed logging data is input into a dynamic extrapolation model. Based on the input data, the model predicts formation parameters of un-drilled formations within a preset distance ahead of the drill bit. This step enables the prediction of unknown formation characteristics, providing a basis for subsequent decision-making.

[0110] Finally, the predicted formation parameters are coupled with a pre-set geomechanical model for analysis. Through this coupling analysis, a dynamic safe drilling window containing safe ranges for drilling parameters is calculated and generated. This safe window provides real-time guidance for parameter adjustments during drilling operations, helping to avoid potential drilling risks.

[0111] The entire process forms a closed loop: from data acquisition, preprocessing, model training to predictive analysis, and then to the generation of safety parameters. This dynamic prediction and real-time adjustment mechanism can effectively improve the safety and efficiency of drilling operations.

[0112] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0113] In subsea drilling operations, logging-while-drilling (LOD) equipment is installed in the drill string assembly, close to the drill bit. This equipment includes a gamma-ray detector, a resistivity meter, and a spontaneous potential meter. As drilling progresses, these devices continuously acquire formation data at sampling intervals of 0.1 meters.

[0114] The collected raw data underwent preprocessing steps. First, outliers, such as sudden zero values ​​or data points outside the normal range, were identified and removed using statistical methods. Then, wavelet transform algorithms were applied to denoise the data, selecting appropriate wavelet basis functions and decomposition levels to retain effective signals while removing high-frequency noise. Finally, the min-max normalization method was used to standardize the data from different logging curves to the 0-1 interval.

[0115] The preprocessed data are arranged in depth order to construct a spatial sequence. A Long Short-Term Memory (LSTM) network is selected as the time series prediction algorithm. The LSTM network structure consists of an input layer, hidden layers, and an output layer, with two hidden layers, each containing 64 neurons. During training, data from the past 50 meters is used as input to predict the geological parameters for the next 5 meters.

[0116] During actual drilling, the system continuously inputs the latest 5-meter logging data into the trained LSTM model. The model outputs predicted results of formation lithology, porosity, and stress state distribution within a range of 1-5 meters in front of the drill bit. These predictions are then input into a pre-set geomechanical model.

[0117] The geomechanical model is constructed based on elasticity theory and empirical formulas, taking into account factors such as formation pressure, rock strength, and stress field distribution. Through coupled analysis, the model calculates safe drilling fluid density and drilling rate ranges. These ranges are combined into a dynamic safe drilling window, which is displayed in real time on the monitoring interface of the drilling control system for drilling engineers to reference and adjust drilling parameters.

[0118] This application further proposes a method for preprocessing well logging curve data, including: cleaning the well logging curve data to remove outliers caused by equipment failure; using wavelet transform denoising method to denoise the cleaned well logging curve data to remove random noise in the data; and standardizing the denoised well logging curve data to convert the well logging curve data to a uniform numerical range.

[0119] Data cleaning involves identifying and removing outlier data points that deviate from the normal range by setting thresholds or using statistical analysis methods. Examples include records with gamma values ​​exceeding a preset upper limit or resistivity below a lower limit. Wavelet transform denoising employs multi-scale decomposition and reconstruction techniques, separating high-frequency noise components from the effective signal by selecting appropriate wavelet basis functions and decomposition levels. Standardization uses linear transformation methods to map gamma, resistivity, and natural potential data to the zero-to-one range. For example, gamma values ​​are normalized by subtracting the minimum value and dividing by the difference between the maximum and minimum values.

[0120] Specifically, during the data cleaning phase, real-time monitoring of logging curve data is conducted. Values ​​exceeding 300 API or resistivity below 0.1 ohm-meters are identified as outliers due to equipment malfunction and removed. Wavelet transform denoising employs a Daubechies wavelet basis for three-level decomposition, using soft thresholding of high-frequency coefficients to suppress noise while retaining low-frequency components reflecting formation characteristics. In the standardization phase, the original data for different logging parameters are normalized by range. For example, the original range of gamma values ​​is compressed to zero to one, and resistivity values ​​are converted to a proportional value between zero and one by dividing by the maximum range. The logging curve data processed in this way eliminates the influence of equipment malfunctions, random noise, and dimensional differences, resulting in a higher signal-to-noise ratio and consistency in the subsequently constructed spatial sequences, thereby improving the prediction accuracy of the dynamic extrapolation model for undrilled formation parameters.

[0121] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0122] Data cleaning is performed on well logging data to remove outliers caused by equipment malfunctions. The data cleaning process includes setting a threshold range, identifying data points outside this range, and marking these data points as outliers. Outliers can be replaced using interpolation or by averaging the values ​​of nearby valid data points.

[0123] Furthermore, wavelet transform denoising is employed to remove random noise from the cleaned logging curve data. Specifically, an appropriate wavelet basis function and decomposition level are selected to perform wavelet decomposition on the logging curve data. Then, thresholding is applied to the wavelet coefficients to suppress noise components. Finally, wavelet reconstruction is performed to obtain the denoised logging curve data.

[0124] The denoised logging data is standardized to convert it to a uniform numerical range. For example, a min-max standardization method can be used to linearly map the data to the 0-1 interval. This allows different types of logging data to be compared and analyzed on the same scale.

[0125] This application further proposes selecting a Long Short-Term Memory (LSTM) network as the time series prediction algorithm; selecting spatial sequences at a first preset depth from drilled formations as training samples, including historical data sequences as model inputs and target data sequences as model outputs; training the LTM network using these training samples, adjusting the number of hidden layers, neurons, and learning rate until the model's prediction error is less than a preset error threshold, thus obtaining a dynamic extrapolation model. The dynamic extrapolation model is continuously updated during drilling, rather than being trained and used only once. Updates are automatically triggered every 5 meters of drilling depth or when a new 5-meter logging data segment is received. A sliding window mechanism is used to determine the training data range during updates, using only the latest 50-meter formation data (covering the 50 meters before the current drill bit position), discarding historical data outside this range through a forgetting mechanism to ensure data timeliness. The update operation employs incremental learning, fine-tuning the existing model weights using new data (reducing the learning rate to 1 / 10 of the initial value), rather than completely retraining, to reduce computational time and maintain model stability.

[0126] The Long Short-Term Memory (LSTM) network controls information flow through a gating mechanism, enabling it to capture long-term dependencies in well logging data along the depth axis. A sliding window mechanism is used to select training samples, extracting continuous data segments at fixed depth intervals from drilled formations. The depth ratio of historical data sequences to target data sequences is set to nine to one to ensure spatial continuity between model input and output. The number of hidden layers is set to two to four, and the number of neurons is dynamically adjusted based on the complexity of the training samples. The initial learning rate is set to 0.001, and an adaptive learning rate algorithm is used for optimization.

[0127] Specifically, during training, well logging curve data at a depth of 45 meters was input into the model as historical data sequences, and the model output a target data sequence at a depth of 5 meters. The mean square error between the predicted and actual values ​​was calculated using the backpropagation algorithm. When the error exceeded a preset threshold, the number of hidden layers or neurons was gradually increased to improve model capacity, or the learning rate was reduced to avoid gradient oscillations. After iterative training, the model's prediction error on the validation set stabilized below 3%, indicating that it could accurately fit the nonlinear variation of well logging curves with depth. The resulting dynamic extrapolation model can effectively predict formation parameters of un-drilled formations ahead of the drill bit, providing a reliable data foundation for subsequent calculations of safe drilling windows.

[0128] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0129] Long Short-Term Memory (LSTM) networks were chosen as the time series forecasting algorithm. LSTM networks are a special type of recurrent neural network that can effectively handle long-term dependencies in long-sequence data.

[0130] A spatial sequence at a depth of 50 meters from the drilled strata was selected as the training sample. The training sample includes historical data sequences as input to the model and target data sequences as output to the model. Specifically, the spatial sequence at the first 45 meters was set as the historical data sequence, and the spatial sequence at the last 5 meters and the corresponding stratum parameters were set as the target data sequence.

[0131] The Long Short-Term Memory (LSTM) network is trained using training samples. By adjusting the number of hidden layers, the number of neurons, and the learning rate of the LTM network, a dynamic extrapolation model is obtained until the prediction error of the model is less than a preset error threshold.

[0132] Furthermore, the hidden states and cell states of the Long Short-Term Memory (LSTM) network are initialized before training. The target hidden state corresponding to the input data is calculated using the initialized LSM network and the trained model parameters, and the output data is determined based on the target hidden state.

[0133] Therefore, the dynamic extrapolation model can predict the geophysical parameters and formation characteristics of un-drilled formations ahead of the drill bit based on the spatial sequence data of drilled formations.

[0134] This application further proposes to select the spatial sequence of the past 50 meters from the drilled strata as training samples, set the spatial sequence of the first 45 meters as the historical data sequence, and set the spatial sequence of the last 5 meters and the corresponding stratum parameters as the target data sequence.

[0135] The training sample depth is limited to 50 meters, a range that balances data continuity and computational efficiency. The first 45 meters of historical data sequence provide the model with sufficient contextual information to learn how formation parameters change with depth. The last 5 meters of target data sequence maps to the historical data sequence, ensuring the correspondence between model output and input. The depth ratio of historical data sequence to target data sequence is set to 9:1, reducing the difficulty of model training by shortening the target sequence length.

[0136] Specifically, during model training, the logging-while-drilling (LOD) equipment continuously collects gamma, resistivity, and spontaneous potential logging curve data at a depth of 50 meters. The data from the first 45 meters, after preprocessing, is arranged in depth order to form a historical data sequence, which serves as input to the Long Short-Term Memory (LSTM) network. The data from the last 5 meters, along with the corresponding formation parameters, serves as the target data sequence for calculating the model's prediction error. By adjusting the number of hidden layers, neurons, and learning rate, the error between the model's output prediction at the 5-meter depth and the target data sequence is kept below a preset threshold. This training method enables the model to accurately capture the correlation between the 45-meter historical data and the 5-meter target data. Therefore, in real-time prediction, only the data from the most recent 45 meters is needed to predict the formation parameters of the un-drilled formations in the preceding 5 meters. For example, when the drill bit reaches a certain depth, the model predicts the formation lithology and porosity of the subsequent 5 meters based on the data from the first 45 meters, providing high-precision input for the generation of dynamic safe drilling windows.

[0137] As a preferred embodiment, the solution of this application is implemented as follows: A spatial sequence of the past 50 meters from the drilled strata is selected as a training sample. The spatial sequence of the first 45 meters is set as the historical data sequence, and the spatial sequence of the last 5 meters and the corresponding stratum parameters are set as the target data sequence. This selection method can fully utilize the information from the drilled strata, providing the model with sufficient historical data for learning and prediction.

[0138] Specifically, the historical data series includes gamma, resistivity, and spontaneous potential logging data within the first 45 meters of depth. This data, after preprocessing, forms a multidimensional time series. The target data series includes logging data within the last 5 meters of depth, along with corresponding formation parameters such as formation lithology, porosity, and stress distribution.

[0139] Therefore, the model can learn the mapping relationship from historical data to target data, and then predict the formation parameters within a five-meter range in the future. This method of selecting training samples enables the model to capture the variation patterns of formation parameters over short distances, thereby improving the accuracy and reliability of predictions.

[0140] This application further proposes to use the logging curve data of the most recent second preset depth, which is collected and preprocessed in real time, as input to the dynamic extrapolation model; the dynamic extrapolation model outputs the formation parameters of the undrilled formation within one to five meters in front of the drill bit, including formation lithology, porosity values ​​and stress state distribution.

[0141] The second preset depth is set to 45 meters. The well logging curve data at the nearest 45-meter depth is used as the input sequence to ensure that the model input data covers sufficient formation variation characteristics. The dynamic extrapolation model adopts a sliding window mechanism. Each time standardized well logging data at a depth of 45 meters is input, the predicted formation parameters within the range of one to five meters are output. The formation lithology prediction is based on the combined characteristics of gamma curves and resistivity curves. The porosity value is calculated through the correlation model of resistivity and spontaneous potential curves. The stress state distribution is derived by combining porosity and formation pressure gradient.

[0142] Specifically, during drilling, the logging-while-drilling (LOD) equipment continuously collects gamma, resistivity, and spontaneous potential data at a depth of 45 meters. After data cleaning, denoising, and standardization, a unified dimensional input sequence is formed. This sequence is input into a dynamic extrapolation model. The model analyzes the spatial variation patterns of the logging curves within the 45-meter depth to identify trends in formation lithology transitions, abrupt changes in porosity, and stress accumulation. The model output layer employs a multi-task learning structure, simultaneously generating lithology classification results, porosity value ranges, and stress state distribution heatmaps within a range of one to five meters. For example, when the model detects a steep drop in the resistivity curve at the end of the 45-meter sequence, combined with the low-value characteristics of the gamma curve, it predicts the existence of a high-porosity sandstone layer three meters ahead and calculates the critical collapse pressure of this layer based on the correlation equation between porosity and formation pressure. The resulting predictions are directly correlated with the safety thresholds of drilling fluid density and drilling rate, providing a quantitative basis for real-time adjustment of drilling parameters.

[0143] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0144] The most recent 45 meters of logging data, acquired and preprocessed in real time, is used as input to the dynamic extrapolation model. The dynamic extrapolation model outputs formation parameters for the un-drilled formation within a range of one to five meters in front of the drill bit. Formation parameters include formation lithology, porosity values, and stress state distribution.

[0145] Specifically, the dynamic extrapolation model employs a long short-term memory (LSTM) network structure, comprising an input layer, a hidden layer, and an output layer. The input layer receives a 45-meter-long sequence of logging curve data, while the hidden layer contains multiple LTM units to capture long-term and short-term dependencies in the data. The output layer generates formation parameter predictions for a future range of five meters.

[0146] During the prediction process, the well logging data is first normalized, then divided into forty-five time steps at one-meter intervals and input into the network. The network processes the data at each time step progressively, updates its internal state, and finally generates the prediction results for the next five time steps at the output layer.

[0147] The predicted results are inversely normalized and then converted into actual formation parameter values. The 'formation parameters 1 to 5 meters ahead' output by the model are discrete depth points, namely five discrete points at 1m, 2m, 3m, 4m, and 5m. Each output point is a vector containing multiple parameters. Stratigraphic lithology (e.g., classification labels for sandstone / shale / mudstone); Porosity value (unit: %) The stress state distribution includes the magnitudes of three principal stresses: vertical principal stress σv, maximum horizontal principal stress σH, and minimum horizontal principal stress σh, in MPa. The formation lithology is obtained through softmax classification, porosity values ​​are obtained through regression methods, and the stress state distribution is predicted using a multi-task learning approach to simultaneously predict the magnitude and direction of the principal stresses.

[0148] This application further proposes to input the predicted formation lithology, porosity values, and stress state distribution into a geomechanical model. The geomechanical model is pre-constructed and takes into account the geostress and rock mechanical properties of the submarine strata. Through calculation and analysis by the geomechanical model, the safe range of drilling fluid density and drilling speed corresponding to the predicted formation parameters is determined. The safe ranges of drilling fluid density and drilling speed are combined to form a dynamic safe drilling window.

[0149] The geomechanical model pre-calibrates the compressive strength and fracturing pressure thresholds of different lithological formations through rock mechanics experiments, establishes a mapping relationship between porosity and formation permeability, and integrates geostress field distribution data from the seafloor region. Formation lithology is mapped to corresponding rock strength parameters, porosity values ​​are converted into permeability indices, and stress state distribution is analyzed as the direction and magnitude of principal stresses within the formation. The safe range for drilling fluid density is determined by comparing the difference between predicted formation pore pressure and fracturing pressure, and the safe range for drilling speed is calculated using the dynamic balance between rock compressive strength and drill bit cutting force.

[0150] Specifically, when the formation lithology is shale, the rock strength parameter is set to the shear strength value under low permeability conditions. The porosity value is converted to the permeability coefficient using Darcy's formula, and the stress state distribution is decomposed into vertical and horizontal stress components. The geomechanical model superimposes the vertical stress and pore pressure to calculate the lower limit of formation collapse pressure, while comparing the horizontal stress with the rock tensile strength to determine the upper limit of formation fracturing pressure. The drilling fluid density is limited between the collapse pressure and the fracturing pressure to prevent wellbore instability. The drilling rate is dynamically adjusted based on real-time feedback from the shale's compressive strength and drill bit torque to ensure that the drilling pressure does not exceed the rock's bearing capacity limit. By synchronously inputting formation lithology, porosity data, and stress state distribution parameters into the geomechanical model, the drilling fluid density and drilling rate are linked for control, forming a real-time updated safe operating range.

[0151] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0152] Predicted formation parameters are input into the geomechanical model, which is pre-constructed and takes into account the in-situ stress and rock mechanical properties of the seafloor strata. Through calculation and analysis using the geomechanical model, the safe ranges for drilling fluid density and drilling speed corresponding to the predicted formation parameters are determined. These safe ranges for drilling fluid density and drilling speed are then combined to form a dynamic safe drilling window.

[0153] The pre-built geomechanical model is a customized model for the well site area, based on the following data: regional geological data, such as stratigraphic stratification and tectonic stress field; actual drilling data from adjacent wells, well logging curves, and core test results; laboratory rock mechanics experiments, including measurements of parameters such as compressive strength and cohesion.

[0154] Specifically, the predicted formation parameters include formation lithology, porosity, and stress distribution. These parameters are predicted by a dynamic extrapolation model based on real-time acquired and preprocessed well logging data. The geomechanical model uses the finite element method to consider the heterogeneity and anisotropy of the seafloor formations and simulates the wellbore stress distribution under different drilling parameters.

[0155] The coupling process is as follows: After inputting the predicted formation parameters into the model, the lithology is matched with the rock mechanical parameters through a preset mapping table. For example, shale corresponds to a compressive strength of 30-50 MPa and an internal friction angle of 25°, while sandstone corresponds to a compressive strength of 60-80 MPa and an internal friction angle of 30°. Porosity is correlated with permeability through an empirical formula (k=10^(-0.01×porosity+2) μm). 2 The pore pressure (the pore pressure increases by 0.05 MPa for every 1% increase in porosity) and the stress concentration factor around the well wall are calculated using finite element analysis. The collapse pressure and rupture pressure are determined by combining the predicted σv, σH, and σh.

[0156] Furthermore, the safe range for drilling fluid density is determined by calculating formation fracturing pressure and pore pressure. The safe range for drilling speed is determined based on wellbore stability analysis and rock fracturing mechanism. For example, for sandstone formations, the lower limit for drilling fluid density is 1.05 times the pore pressure, and the upper limit is 0.95 times the formation fracturing pressure; the drilling speed range is 20-60 meters per hour.

[0157] Formula for calculating the safe range of drilling fluid density: Lower limit = Collapse pressure / 1000 (g / cm³) 3 Upper limit = Rupture pressure / 1000 (g / cm) 3 ); The safe range of drilling speed is based on the rock drillability model (drilling speed = 0.5 × compressive strength^(-0.3) × drilling pressure^0.8, unit m / h).

[0158] Therefore, the dynamic safety drilling window is presented graphically, with the horizontal axis representing drilling fluid density and the vertical axis representing drilling speed. Safe operating areas are marked in green. Preferably, the window updates every 5 seconds to provide drilling engineers with real-time guidance.

[0159] The dynamic nature of the safe drilling window is reflected in the fact that it is updated in real time as the predicted formation parameters ahead of the drill bit change. Each time the prediction module outputs new formation parameters of 1-5 meters, the update is triggered immediately. The update frequency is synchronized with the logging data acquisition frequency (updated once every 0.1 meters of depth).

[0160] The safe range displayed in the window is an interval that varies with the predicted depth, such as a density of 1.2-1.3 g / cm³ at 1 meter. 3 The density at 5 meters is 1.3-1.4 g / cm³. 3 The parameters are presented as curves. This window automatically drives the drilling control system: it uses a PID control algorithm to convert the density range into mud pump frequency commands (e.g., the lower limit of density corresponds to a pump frequency of 45Hz, and the upper limit corresponds to 50Hz), and converts the drilling speed range into drilling pressure setpoints (e.g., the lower limit of drilling speed corresponds to a drilling pressure of 18kN, and the upper limit corresponds to 22kN). The system includes a real-time feedback loop that continuously monitors the deviation between the actual drilling parameters and the safety window, and automatically corrects the control commands when the deviation exceeds 5%.

[0161] This application further proposes to feed back the dynamic safety drilling window to the drilling control system in real time, providing a basis for decision-making on the adjustment of drilling parameters.

[0162] The dynamic safety drilling window establishes a communication link with the drilling control system via a data interface, employing the industrial Ethernet protocol to achieve millisecond-level latency transmission. The drilling control system has a built-in parameter parsing module that converts the safe ranges of drilling fluid density and drilling speed into executable hydraulic pump frequency adjustment commands and drilling pressure control commands. The drilling parameter adjustment process uses a feedforward control strategy, correcting control parameters in advance based on predicted formation parameters.

[0163] Specifically, after the dynamic safe drilling window is generated, the safe range values ​​of drilling fluid density and drilling speed are encapsulated into structured data packets using a preset data transmission protocol. These data packets are transmitted via industrial Ethernet to the central processing unit (CPU) of the drilling control system. The CPU then uses a parameter parsing algorithm to convert the safe range into hydraulic pump speed thresholds and drill pipe advance rate ranges. For example, when the predicted formation porosity exceeds a critical value, the safe range for drilling fluid density is interpreted as requiring the hydraulic pump frequency to be increased to 45-50Hz, and the safe range for drilling speed is interpreted as requiring the drilling pressure to be reduced to 18-22kN. Based on the parsing results, the control system automatically generates control commands to drive the hydraulic pump and drilling rig actuators to adjust synchronously, achieving dynamic optimization of drilling fluid density and drilling speed. This process eliminates human intervention delays through real-time closed-loop control, ensuring that drilling parameters remain within the safe window range and effectively preventing wellbore instability and formation fracturing.

[0164] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0165] The dynamic safety drilling window provides real-time feedback to the drilling control system, offering a basis for decision-making regarding drilling parameter adjustments. Specifically, the dynamic safety drilling window is transmitted via a data transmission network to the central control room of the drilling platform. The central control room is equipped with a real-time monitoring system that displays the received dynamic safety drilling window information graphically on a screen. Drilling engineers can visually observe whether the current drilling fluid density and drilling speed are within safe ranges.

[0166] Furthermore, the real-time monitoring system is equipped with an early warning mechanism. When drilling parameters are detected to be about to exceed the safe range, the system automatically issues an alarm signal. For example, when the drilling fluid density approaches the lower limit of the safe range, the system prompts to increase the drilling fluid density; when the drilling speed approaches the upper limit of the safe range, the system suggests reducing the drilling speed.

[0167] Therefore, drilling engineers can adjust key parameters such as drilling fluid density and drilling speed in a timely manner based on the information provided by the dynamic safety drilling window. Specific adjustments include changing the drilling fluid density by adjusting the drilling fluid formula and adjusting the drilling speed by controlling the drill bit speed and pressure on the drill bit. These adjustments are executed by the drilling control system to ensure that drilling operations are always conducted within safe limits.

[0168] This application further proposes to uniformly transform well logging curve data of gamma, resistivity and spontaneous potential with different units and dimensions into a preset numerical range of zero to one.

[0169] The standardization process employs a min-max normalization method, mapping the original data to the zero-to-one range through a linear transformation. The original unit for gamma-ray logging data is API, for resistivity logging data it's ohm-meter, and for spontaneous potential logging data it's millivolt. The numerical ranges of these parameters differ significantly. For example, gamma-ray logging data might fall within the range of 0 to 150 API, while resistivity logging data might fall within the range of 0 to 200 ohm-meter. By pre-setting a unified numerical range, the original values ​​of each parameter are converted into dimensionless relative values, ensuring comparability between different parameters during model training.

[0170] Specifically, standardization is achieved through the following formula: For a given logging parameter, its normalized value equals the original value minus the minimum value of that parameter in the drilled formation, divided by the difference between the maximum and minimum values ​​of that parameter. This process compresses all logging curve data to the range of zero to one. When the standardized data is input into the dynamic extrapolation model, the model does not need to adjust the weights to balance the magnitude differences between different parameters, thereby improving the model's convergence speed. Simultaneously, the standardized data can more accurately reflect the changing trends of formation parameters, avoiding oversensitivity of the model to high-magnitude parameters due to differences in numerical scale. For example, in gamma logging data, when the value changes from 50 API to 100 API, the normalized change range is 0.33 to 0.67, consistent with the change range in resistivity logging data from 50 ohm-meters to 100 ohm-meters, thus ensuring a balanced model response to changes in various parameters.

[0171] As a preferred embodiment, the specific implementation of this application is as follows: During the standardization process, the raw values ​​of gamma logging data recorded in API units are transformed to the zero-to-one interval using a linear normalization method, where the maximum value of gamma is set to 200 API and the minimum value is set to 20 API; the raw values ​​of resistivity logging data recorded in ohm-meters are mapped to the zero-to-one interval, where the maximum value of resistivity is set to 100 ohm-meters and the minimum value is set to 0.1 ohm-meters; the raw values ​​of spontaneous potential logging data recorded in millivolts are linearly transformed using the range method, where the maximum value of spontaneous potential is set to 200 millivolts and the minimum value is set to -100 millivolts. After the above transformations, the values ​​of different logging curves form a standardized dataset with uniform dimensions, where the numerical scale differences of each parameter are eliminated.

[0172] This application further proposes to standardize the denoised logging curve data, transforming the gamma, resistivity and spontaneous potential logging curve data of different units and dimensions into a preset numerical range of zero to one, so as to eliminate the numerical scale differences between different geophysical parameters.

[0173] The standardization process employs a min-max scaling method, mapping the original data to a zero-to-one range through a linear transformation. For example, the original values ​​for gamma-ray logging curves might range from 30 to 150 API units, resistivity logging curves from 0.1 to 20 ohm-meters, and spontaneous potential logging curves from -50 to +50 millivolts. By calculating the minimum and maximum values ​​of each parameter, a formula is used to convert each data point to a value between zero and one.

[0174] Specifically, the standardization process first calculates the minimum and maximum values ​​for the gamma, resistivity, and spontaneous potential logging curves in their respective datasets. Then, the minimum value is subtracted from each data point, and the result is divided by the difference between the maximum and minimum values ​​to obtain the normalized value. This unifies the numerical scale of different parameters, avoiding parameter weight imbalances during model training due to differences in dimensions. For example, the standardized gamma logging data has a numerical distribution of 0.2 to 0.8, resistivity logging data 0.05 to 0.95, and spontaneous potential logging data 0.3 to 0.7. This process ensures the comparability of different geophysical parameters in the model input, improves the efficiency of the Long Short-Term Memory network in extracting joint features of multiple parameters, and thus enhances the accuracy of the dynamic extrapolation model in predicting parameters of undrilled formations.

[0175] As a preferred embodiment, the specific implementation of this application is as follows: Before the training phase of the Long Short-Term Memory (LSTM) network begins, the hidden states and cell states are initialized with zero values. The hidden state tensor is initialized as a zero matrix, with its dimension matching the number of neurons in the hidden layer. The cell state tensor is also initialized as a zero matrix, with its dimension matching the number of neurons in the hidden layer. After initialization, historical data sequences from the training samples are input into the network, and the target hidden state is calculated through forward propagation. The update process of the hidden state and cell state at each time step of the input sequence is implemented using a gating mechanism. During backpropagation, the network weight parameters are adjusted using a gradient descent algorithm based on the error between the predicted output and the target data sequence until the model converges. After training, in the prediction phase, real-time collected well logging curve data is input into the initialized network, the target hidden state is calculated based on the trained parameters, and the target hidden state is mapped to the formation parameter prediction values ​​through a fully connected layer.

[0176] In summary, this application, by real-time acquisition of geophysical parameters such as gamma and resistivity, constructs a spatial sequence after preprocessing, and uses a long short-term memory network to train a dynamic extrapolation model, accurately predicts the lithology, porosity, and stress state of un-drilled formations 1-5 meters ahead of the drill bit. Furthermore, it combines this with a geomechanical model to generate a dynamic safe drilling window that includes safe ranges for drilling fluid density and velocity, feeding this information back to the drilling control system in real time. This provides a basis for parameter adjustments, thereby proactively mitigating risks such as wellbore collapse and leakage. This effectively solves the problem that existing logging-while-drilling technology can only reflect formation information where the drill bit is "already" or "just passed through," and lacks sufficient ability to predict geological anomalies 1-5 meters ahead. This significantly improves the safety and operational efficiency of subsea oil and gas drilling.

[0177] like Figure 8 As shown, this application further proposes a system for predicting drilling risks using geophysical data, including a data acquisition module, a data preprocessing module, a model training module, a prediction module, and a coupling analysis module.

[0178] The system comprises several modules: a data acquisition module that acquires real-time gamma, resistivity, and spontaneous potential logging curves using logging-while-drilling equipment; a data preprocessing module connected to the acquisition module that performs data cleaning, wavelet transform denoising, and standardization; a model training module that constructs a spatial sequence of depth-based logging curves from the preprocessed data and uses a long short-term memory network (LSTM) as the time series prediction algorithm to train a dynamic extrapolation model based on the drilled formation spatial sequence; and a prediction module that receives the real-time preprocessed logging curves and inputs them into the dynamic extrapolation model to predict un-drilled formation parameters within a range of one to five meters ahead of the drill bit. A coupled analysis module that inputs the predicted formation lithology, porosity values, and stress distribution into a pre-set geomechanical model to calculate the safe ranges for drilling fluid density and drilling speed, forming a dynamic safe drilling window.

[0179] Specifically, the data acquisition module obtains real-time logging curve data from the logging-while-drilling (LWD) equipment. The data preprocessing module cleans the data to remove outliers caused by equipment malfunctions, uses wavelet transform to remove random noise, and standardizes logging curve data of different units and dimensions to the zero-to-one range. The model training module selects the spatial sequence of drilled formations over the past 50 meters as training samples, with the first 45 meters as the historical data sequence and the last 5 meters as the target data sequence. By adjusting the number of hidden layers, neurons, and learning rate of the Long Short-Term Memory (LSTM) network, the prediction error is kept below a preset threshold. The prediction module inputs the preprocessed data from the second most recent preset depth into the dynamic extrapolation model and outputs parameters such as the lithology, porosity, and stress distribution of the undrilled formation. The coupling analysis module calculates the safe drilling window based on the geomechanical model and feeds the results back to the drilling control system in real time. For example, the safe range for drilling fluid density is set to 1.2-1.5 g / cm³ via a preset interface. 3The drilling speed is controlled between 0.5 and 1.2 m / h.

[0180] As a preferred embodiment, the solution of this application is implemented as follows: The data acquisition module is configured near the drill bit position on the drill string, and acquires logging-while-drilling data in real time through a gamma sensor, an array inductive resistivity meter, and a spontaneous potential electrode. The sampling frequency is set to record the measurement value once every 10 centimeters of depth. The data preprocessing module is deployed in the downhole processor. It detects abrupt changes in the gamma curve using a sliding window algorithm, performs a five-level decomposition of the resistivity data using the sym4 wavelet basis function to remove high-frequency noise, and applies a minimum-maximum normalization method to linearly map the spontaneous potential data to the zero-to-one interval. The model training module runs on a surface server and adopts a long short-term memory network architecture with two hidden layers. Each hidden layer has 64 neurons. During training, the Adam optimizer is used to iteratively update the weight parameters with a learning rate of 0.001. The prediction module receives the preprocessed 45-meter logging data sequence, continuously inputs it into the dynamic extrapolation model through a sliding window mechanism, and outputs the predicted porosity value at a depth of 5 meters ahead in real time. The coupled analysis module integrates a geomechanical solver, calculates formation shear strength based on predicted porosity using the Mohr-Coulomb criterion, generates a safe range of drilling fluid density of 1.2 to 1.4 grams per cubic centimeter by combining it with the wellbore stability analysis model, and uploads this parameter range to the ground control terminal via mud pulse signals.

[0181] like Figure 9 As shown in the illustration, this application also provides a drilling risk prediction system 900 based on geophysical data, which can implement the above-described method. The system includes:

[0182] The data acquisition module 910 is used to acquire geophysical parameters in real time during the seabed drilling process using logging-while-drilling equipment. The geophysical parameters correspond to depth information and include logging curve data.

[0183] The data preprocessing module 920 is used to preprocess geophysical parameters and generate target well logging data;

[0184] The model training module 930 is used to construct the target logging data into a spatial sequence with depth as the axis, and to train and generate a dynamic extrapolation model based on the spatial sequence of drilled formations using a preset time series prediction algorithm.

[0185] Prediction module 940 is used to input real-time target logging data into a dynamic extrapolation model to predict formation parameters of un-drilled formations within a preset distance in front of the drill bit.

[0186] The coupling analysis module 950 is used to couple formation parameters with a preset geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters.

[0187] In some embodiments, the system may further include a model update module for performing the following operations:

[0188] Use the end depth of the first preset depth as the target depth;

[0189] To obtain the drilling depth at the target depth in seabed drilling;

[0190] When the drilling depth reaches the target depth, the spatial sequence corresponding to the target distance is used as a new training sample; historical data of spatial sequences that exceed the target distance are discarded through a forgetting mechanism.

[0191] The dynamic extrapolation model is fine-tuned using new training samples, and then updated based on the results of the fine-tuning.

[0192] Using the final depth of the target distance as the target depth, return to the step of obtaining the drilling depth of the seabed drilling distance to the target depth, and continuously update and adjust the dynamic extrapolation model.

[0193] In some embodiments, the system may also include a model building module for performing the following operations:

[0194] By pre-calibrating the compressive strength and fracture pressure threshold of different lithological strata through rock mechanics experiments, establishing the mapping relationship between porosity and stratum permeability, and integrating the geostress field distribution data of the seabed area, a geomechanical model was constructed.

[0195] In the geomechanical model, formation lithology is mapped to corresponding rock strength parameters, porosity values ​​are converted into permeability indices, and stress state distribution is analyzed as the direction and magnitude of principal stresses within the formation. The safe range of drilling fluid density is determined by comparing the difference between predicted formation pore pressure and fracture pressure, and the safe range of drilling speed is determined by the dynamic balance between rock compressive strength and drill bit cutting force.

[0196] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0197] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0198] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0199] like Figure 10 As shown, Figure 10 The hardware structure of an electronic device 1000 according to another embodiment is illustrated. The electronic device 1000 includes:

[0200] The processor 1001 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (aSIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0201] The memory 1002 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RaM). The memory 1002 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1002 and is called and executed by the processor 1001.

[0202] Input / output interface 1003 is used to implement information input and output;

[0203] The communication interface 1004 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0204] Bus 1005 transmits information between various components of the device (e.g., processor 1001, memory 1002, input / output interface 1003, and communication interface 1004);

[0205] The processor 1001, memory 1002, input / output interface 1003 and communication interface 1004 are connected to each other within the device via bus 1005.

[0206] The electronic device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0207] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0208] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0209] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0210] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0211] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0212] The drilling risk prediction method, system, electronic device, storage medium, and program product based on geophysical data provided in this application embodiment utilizes logging-while-drilling equipment to collect geophysical parameters in real time during the subsea drilling process. The geophysical parameters correspond to depth information and include logging curve data. The geophysical parameters are preprocessed to generate target logging data. The target logging data is constructed into a spatial sequence with depth as the axis. Based on the spatial sequence of drilled formations, a dynamic extrapolation model is trained using a preset time series prediction algorithm. The real-time target logging data is input into the dynamic extrapolation model to predict the formation parameters of undrilled formations within a preset distance in front of the drill bit. The formation parameters are coupled with a preset geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters. This application embodiment acquires geophysical parameters in real time, constructs spatial sequences through preprocessing, and then trains a dynamic extrapolation model. Based on the data association mapping relationship learned by the model, it can accurately predict the formation parameters of the un-drilled formation ahead of the drill bit. In addition, it combines the geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters, which can provide a basis for parameter adjustment and thus avoid risks such as wellbore collapse and leakage in advance. It effectively solves the problem that the existing logging-while-drilling technology can only reflect the formation information that the drill bit "has been" or "just passed", and has insufficient ability to predict geological anomalies ahead, which significantly improves the safety and operational efficiency of subsea oil and gas drilling.

[0213] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A drilling risk prediction method based on geophysical data, characterized in that, The method includes the following steps: Geophysical parameters during the subsea drilling process are collected in real time using logging-while-drilling equipment. These geophysical parameters correspond to depth information and include logging curve data. The geophysical parameters are preprocessed to generate target well logging data; The target logging data is constructed as a spatial sequence with depth as the axis, and a dynamic extrapolation model is generated by training the spatial sequence of the drilled formation using a preset time series prediction algorithm. The real-time target logging data is input into the dynamic extrapolation model to predict the formation parameters of the un-drilled formation within a preset distance in front of the drill bit. The formation parameters are coupled with a preset geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters. The process of generating a dynamic extrapolation model based on the spatial sequence of drilled strata using a preset time series prediction algorithm includes the following steps: The time series prediction algorithm is configured as a long short-term memory network; The spatial sequence at the nearest first preset depth is obtained from the drilled stratum as a training sample. The training sample includes a first spatial sequence at the beginning of the first preset depth and a second spatial sequence at the end of the first preset depth. The length of the first spatial sequence is greater than the length of the second spatial sequence. The first spatial sequence is used as input data, and the second spatial sequence and its corresponding stratum parameters are used as output labels. The Long Short-Term Memory (LSTM) network is trained using the training samples. The number of hidden layers, the number of neurons, and the learning rate of the LSM network are adjusted until the prediction error of the LSM network is less than a preset error threshold, thus obtaining the dynamic extrapolation model. The step of coupling the formation parameters with a preset geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters includes the following steps: The formation parameters are input into the geomechanical model, and the safe range of drilling fluid density and drilling speed corresponding to the formation parameters is determined by the analysis of the geomechanical model. The formation parameters include formation lithology, porosity values, and stress state distribution, and the geomechanical model is pre-constructed based on the geostress and rock mechanical properties of the submarine strata. The dynamic safety drilling window is formed by combining the safe range of the drilling fluid density and the drilling speed.

2. The method according to claim 1, characterized in that, The geophysical parameters also include gamma and resistivity. The preprocessing of the geophysical parameters includes the following steps: The well logging curve data is cleaned to remove outliers caused by equipment failure, resulting in the first well logging data. The first logging data is denoised using wavelet transform to remove random noise and obtain the second logging data. The gamma, resistivity, and second logging data of different units and dimensions are uniformly transformed to a preset numerical range to obtain the target logging data.

3. The method according to claim 1, characterized in that, The method further includes the following steps: Take the end depth of the first preset depth as the target depth; Obtain the drilling depth of the seabed drilling operation at a distance from the target depth; When the drilling depth reaches the target distance from the target depth, the spatial sequence corresponding to the target distance is used as a new training sample; wherein, historical data of the spatial sequence that exceeds the target distance is discarded through a forgetting mechanism; The dynamic extrapolation model is fine-tuned using the new training samples, and the dynamic extrapolation model is updated based on the results of the fine-tuning. Using the end depth of the target distance as the target depth, return to the step of obtaining the drilling depth of the seabed drilling distance to the target depth, and continuously update and adjust the dynamic extrapolation model.

4. The method according to claim 1, characterized in that, The step of inputting the real-time target logging data into the dynamic extrapolation model to predict the formation parameters of the un-drilled formation within a preset distance in front of the drill bit includes the following steps: The target logging data at the nearest second preset depth is obtained from the drilled formation as input data; The input data is input into the dynamic extrapolation model to predict the formation parameters of the un-drilled formation within the preset distance in front of the drill bit. The training samples for generating the dynamic extrapolation model include a first spatial sequence of the first segment and a second spatial sequence of the second segment of the first preset depth of the nearest drilled stratum. The first spatial sequence is used as the training input data, and the second spatial sequence and its corresponding stratum parameters are used as the output labels. The preset distance is the same as the depth range corresponding to the second spatial sequence.

5. The method according to claim 1, characterized in that, The method further includes the following steps: The compressive strength and fracture pressure threshold of different lithological strata were pre-calibrated through rock mechanics experiments, the mapping relationship between porosity and stratum permeability was established, and the geomechanical model was constructed by integrating the geostress field distribution data of the seabed area. In the geomechanical model, the formation lithology is mapped to the corresponding rock strength parameters, the porosity value is converted into a permeability index, and the stress state distribution is analyzed as the direction and magnitude of the principal stress within the formation. The safe range of drilling fluid density is determined by comparing the difference between the predicted formation pore pressure and the fracture pressure, and the safe range of drilling speed is determined by the dynamic balance between the rock compressive strength and the drill bit cutting force.

6. A drilling risk prediction system based on geophysical data, characterized in that, The system, applied to the method of claim 1, comprises: The data acquisition module is used to acquire geophysical parameters in real time during the seabed drilling process using logging-while-drilling equipment. The geophysical parameters correspond to depth information and include logging curve data. The data preprocessing module is used to preprocess the geophysical parameters to generate target well logging data; The model training module is used to construct the target logging data into a spatial sequence with depth as the axis, and to train and generate a dynamic extrapolation model based on the spatial sequence of the drilled formation using a preset time series prediction algorithm. The prediction module is used to input the real-time target logging data into the dynamic extrapolation model to predict the formation parameters of the un-drilled formation within a preset distance in front of the drill bit. The coupling analysis module is used to couple the formation parameters with a preset geomechanical model to generate a dynamic safe drilling window that includes the safe range of drilling parameters.

7. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 5.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 5.

Images