Data and physical double-driven method and system for predicting rock mass strength while drilling

Abstract

The present application belongs to the technical field of rock strength prediction, and provides a method and system for predicting rock mass strength while drilling based on data and physical double driving, which fuses construction parameters collected by a drilling machine in real time and physical intermediate variables calculated by using a physical empirical model to form a physically enhanced high-dimensional feature vector; a double driving objective function under physical constraints is constructed, the physically enhanced high-dimensional feature vector is taken as input, a residual learning model pre-trained under the double driving objective function is used for prediction to obtain a preliminary prediction result, and the preliminary prediction result is subjected to physical constraint verification; if the physical constraint verification is passed, the preliminary prediction result is taken as the final result; otherwise, the current abnormal point is corrected by using prediction values of upstream and downstream depths, and the corrected prediction result is taken as the final result. The present application improves the stability and reliability of rock mass strength prediction in engineering application.

Description

Technical Field

[0001] This invention belongs to the field of rock strength prediction technology, specifically relating to a method and system for predicting rock mass strength while drilling based on a dual-drive approach of data and physics. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In deep-earth resource extraction, tunnel engineering, and oil drilling, rock mass strength (such as uniaxial compressive strength, UCS) is a key mechanical indicator for evaluating rock drillability, optimizing drilling parameters, and ensuring the stability of the wellbore / tunnel surrounding rock. Real-time and accurate acquisition of rock mass strength information in front of and around the drill bit is crucial for the automatic control of the drilling system.

[0004] Currently, the most common methods for obtaining rock mass strength include indoor experimental methods, physical model analysis methods, and pure data-driven methods. However, all of these methods have certain problems.

[0005] Specifically, laboratory testing methods generally rely on drilling cores for uniaxial or triaxial compression tests. These methods are time-consuming and costly, high-value samples are scarce, and test results often fail to reflect the actual mechanical state of geological rock masses. Physical model analysis, on the other hand, assumes the rock is an isotropic homogeneous material and the drill bit is in an ideal cutting state. This differs significantly from the reality of heterogeneous, anisotropic, and fractured formations, leading to large prediction errors in complex formations. Furthermore, physical model analysis struggles to quantify the impact of complex factors such as drill bit wear on energy efficiency, resulting in distorted results.

[0006] Pure data-driven methods are generally based on statistical laws and ignore the physical nature of rock fracturing. Under certain conditions, the model may output prediction results that violate common sense physics. Furthermore, when encountering unseen strata or extreme conditions, the prediction performance of pure data models is prone to a sharp decline and has poor robustness. Summary of the Invention

[0007] To address the aforementioned problems, this invention proposes a method and system for predicting rock mass strength during drilling based on both data and physical factors. This invention integrates drilling parameters and mechanical constraints to achieve dynamic prediction of rock mass strength ahead, reducing the uncertainty of pure data models and improving the stability and reliability of rock mass strength prediction in engineering applications.

[0008] According to some embodiments, the present invention adopts the following technical solution: A method for predicting rock mass strength while drilling based on both data and physics-driven approaches includes the following steps: The construction parameters collected in real time by the drilling rig are fused with the physical intermediate variables calculated using a physical empirical model to form a physically enhanced high-dimensional feature vector. Using the physically enhanced high-dimensional feature vector as input, a pre-trained residual learning model is used to make predictions and obtain preliminary prediction results. The preliminary prediction results are then verified by physical constraints. If the physical constraint verification passes, the preliminary prediction result shall be taken as the final result; otherwise, the predicted values ​​of the upstream and downstream depths shall be used to correct the current anomaly point to ensure that the rock mass strength curve in the preliminary prediction result conforms to the continuity characteristics of geological sedimentation, and the corrected prediction result shall be taken as the final result. In the training process of the residual learning model, it is trained with a dual-drive objective function under physical constraints. The dual-drive objective function includes two aspects: prediction loss and physical regularization term. It is trained in multiple rounds using the forward distribution algorithm. In each round of iteration, the residual between the prediction result of the previous round and the true label is fitted by growing a decision tree, the gain score of feature splitting is calculated, and the physical benchmark feature is preferentially selected as the main splitting node.

[0009] As an alternative implementation method, the process of fusing the construction parameters collected in real time by the drilling rig with the physical intermediate variables calculated using the physical experience model to form a physically enhanced high-dimensional feature vector includes: acquiring several data points from the current drilling pressure, rotation speed, torque, mechanical drilling speed and flushing fluid pressure during the drilling process; acquiring the actual strength of the drilling rig sampling; acquiring the physical theoretical rock mass strength and derived fracture specific work calculated by the physical experience model; acquiring the stuck drill risk indicator, breakthrough indicator and groundwater outburst indicator identified by expert experience; and splicing and fusing them to obtain the physically enhanced high-dimensional feature vector.

[0010] As a further defined implementation method, the process of calculating the physical theoretical rock mass strength using a physical empirical model includes: simplifying the torque work formula based on the Xianglushan mechanical specific energy formula to obtain the unit rock-breaking formula. Combined with the elastic strain energy formula Obtain the theoretical value of the rock mass strength at the current moment, i.e., the physical theoretical value:

[0011] in, This is the preset mechanical efficiency coefficient. The drilling speed of the drill pipe. For torque, The diameter of the drill bit. For drilling speed, Uniaxial compressive strength of rocks with different lithologies The elastic modulus is the average value for different lithologies. This is the energy conversion coefficient.

[0012] As an alternative implementation method, the process of constructing a dual-drive objective function under physical constraints includes: the dual-drive objective function is: ; in, The residual loss function is a second-order Taylor expansion, used to drive the model to fit the true value. y true Compared with physical theoretical values y phy Deviation; This is a model complexity regularization term used to prevent overfitting. f k For the model's first A decision tree; This is a physical consistency penalty term; it increases sharply when the predicted value violates the monotonicity of rock mechanics or becomes negative. These are the physical constraint weighting coefficients.

[0013] As an alternative implementation, during multi-round iterative training using the forward distribution algorithm, in the... In each iteration, the model does not directly predict the absolute value of rock mass strength, but instead grows a new decision tree. The residual between the previous prediction result and the true label is fitted. The gain score of feature splitting is calculated, and the physical theoretical value is selected as the main splitting node to ensure that the physical law occupies the dominant position in the tree structure. Drilling pressure and rotation speed are used to fine-tune the nonlinear deviation at the leaf nodes.

[0014] As an optional implementation method, the process of verifying the physical constraints of the preliminary prediction results includes: summing the scores of the leaf nodes of all iterative trees to obtain the preliminary rock mass strength prediction value, checking whether the preliminary rock mass strength prediction value falls within the preset valid value range, and if not, it is considered that the verification has failed. Calculate the gradient of the change between the current depth and the predicted value of the previous depth. If the gradient exceeds the threshold allowed by the stratigraphic deposition law, the verification is considered to have failed.

[0015] As an alternative implementation method, the process of correcting the current anomaly point using the predicted values ​​of upstream and downstream depths includes: using Kalman filtering or neighborhood weighted averaging to correct the current anomaly point using the predicted values ​​of upstream and downstream depths to ensure that the output rock mass strength curve conforms to the continuity characteristics of geological sedimentation.

[0016] A drilling rock mass strength prediction system based on both data and physics-driven approaches includes: The data acquisition module is configured to fuse the construction parameters collected in real time by the drilling rig with the physical intermediate variables calculated using the physical experience model to form a physically enhanced high-dimensional feature vector. The prediction and verification module is configured to take the physically enhanced high-dimensional feature vector as input, use the pre-trained residual learning model to make predictions, obtain preliminary prediction results, and perform physical constraint verification on the preliminary prediction results. The result output module is configured to take the preliminary prediction result as the final result if the physical constraint verification passes; otherwise, it will use the predicted values ​​of the upstream and downstream depths to correct the current anomaly point to ensure that the rock mass strength curve in the preliminary prediction result conforms to the continuity characteristics of geological sedimentation, and take the corrected prediction result as the final result. During the training process of the residual learning model, it is trained using a dual-drive objective function under physical constraints. The dual-drive objective function includes two aspects: prediction loss and physical regularization term. The forward distribution algorithm is used for multiple rounds of iterative training. In each round of iteration, the residual between the prediction result of the previous round and the true label is fitted by growing a decision tree, the gain score of feature splitting is calculated, and the physical benchmark feature is preferentially selected as the main splitting node.

[0017] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the steps in the above method.

[0018] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the steps in the method described above.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention integrates drilling parameters with mechanical constraints to achieve dynamic prediction of rock mass strength ahead, reducing the uncertainty of pure data models and improving the stability and reliability of rock mass strength prediction in engineering applications. It also solves the model training problems caused by difficult and costly on-site sampling and the scarcity of high-value samples, thereby reducing data acquisition costs.

[0020] This invention achieves a deep complementarity between physical mechanisms and data simulation, preserving the correctness of the macroscopic trends of the physical model while using the data model to refine local errors, thus improving prediction accuracy and interpretability compared to a single method.

[0021] This invention sets up physical constraint verification, integrates rock mechanics extreme boundary and stratigraphic sedimentary gradient rules, can intercept and smoothly correct abnormal prediction values ​​in real time, ensure that the output results are logically self-consistent and engineering-usable, and enhance the robustness of the system.

[0022] This invention has a certain learning ability. When new measured feedback data is entered at the engineering site, it can automatically trigger fine-tuning of model parameters, enabling the algorithm to dynamically adapt to the specific geological features of the current site.

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0024] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0025] Figure 1 A flowchart of a drilling rock mass strength prediction method according to one embodiment; Figure 2 This is a block diagram of a drilling rock mass strength prediction system according to one embodiment. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0027] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, 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 invention pertains.

[0028] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0029] Where there is no conflict, the embodiments and features described in this application may be combined with each other.

[0030] Example 1 A data- and physics-driven method for predicting rock mass strength during drilling, such as... Figure 1 As shown, it includes the following steps: Step 1: Construct a physically enhanced high-dimensional feature space. This involves fusing the real-time drilling parameters collected by the drilling rig with the physical intermediate variables calculated in previous steps. Specifically, this involves establishing an input feature vector. The vector contains the following feature subset: Original drilling characteristics set: including current weight on bit (WOB), rotational speed (RPM), torque (Torque), rate of penetration (ROP), and flushing fluid pressure; Strength simulation feature set: includes the actual strength sampled by the drilling rig. ; Physical empirical feature set: includes theoretical rock mass strength calculated from physical empirical models ( ), and the derived specific work of breaking (MSE); Expert experience feature set: Discrete state labels output by expert experience judgment, identifying stuck drill risk. Advance marker Groundwater outcrop marker The final enhanced input vector is: ; Step 2: Define the dual-driving objective function under physical constraints. Unlike traditional XGBoost which only minimizes the prediction error, this embodiment constructs a composite objective function that includes both a "prediction loss" and a "physical regularization term":

[0031] in: This is the residual loss function of the second-order Taylor expansion, used to drive the deviation between the model fitting the true value and the physical theoretical value; This is a model complexity regularization term used to prevent overfitting; This is a physical consistency penalty term; it increases sharply when the predicted value violates the monotonicity of rock mechanics or becomes negative. These are the physical constraint weighting coefficients; Step 3: Perform additive training based on residual approximation. Use the forward distribution algorithm for multiple rounds of iterative training. In the... In each iteration, the model does not directly predict the absolute value of the rock mass strength, but instead grows a new decision tree. The system fits the residual between the previous prediction result and the true label. It calculates the gain score of feature splitting, prioritizing physical baseline features. As the main splitting node, it ensures that physical laws dominate the tree structure, while other parameters are used to fine-tune the weights of the leaf nodes. Step 4: Output preliminary prediction results and perform physical logic post-processing. Accumulate the scores of the leaf nodes of all iterative trees to obtain preliminary rock mass strength prediction values. Then, the physical constraint verification begins: Extreme value verification: Check Does it fall within the valid value range? Gradient verification: Calculate the gradient of the change between the current depth and the predicted value at the previous depth. ,like If the threshold exceeds the allowable threshold of the stratigraphic depositional pattern, it is judged as "abnormal"; Step 5: Adaptive correction and smoothing output. If the verification in Step 4 passes, the final result is output directly. If the verification is "abnormal", the "correction / smoothing" subroutine is triggered: Kalman filtering or neighborhood weighted average method is used to correct the current abnormal point using the predicted values ​​of upstream and downstream depths to ensure that the output rock mass strength curve conforms to the continuity characteristics of geological sedimentation, and finally outputs the rock mass strength prediction value with clear physical meaning. Step Six: Visual Interaction and Online Model Updates. The final predicted rock mass strength values ​​are displayed on the visualization interface, a rock mass strength profile curve varying with depth is plotted, and the boundaries between soft and hard strata are marked with different colors. Corresponding recommended values ​​and confidence intervals for construction parameters are also provided.

[0032] Example 2 A drilling rock strength prediction system driven by both data and physics, such as Figure 2 As shown, the system physically includes a drilling rig and its supporting sensor group and a main control room host. The core software architecture of the system is divided into four major functional modules: drilling parameter acquisition module, data communication and transmission module, data processing module, and control terminal processing module.

[0033] The physical entity includes the drilling rig itself and its associated monitoring sensors used to generate raw mechanical response signals during the drilling process; The drilling parameter acquisition module is connected to the physical entity part via signal and is controlled by the control terminal processing module. Through the signal acquisition link, it acquires the analog or digital signals generated by the physical entity part in real time, and receives instructions from the control terminal processing module. Based on the instructions, it starts, stops, or adjusts the sampling frequency to ensure data synchronization.

[0034] The data communication transmission module is used to connect bidirectionally with the control terminal processing module through a communication link. Its main function is to transmit the standardized data acquired by the drilling parameter acquisition module to the control terminal processing module in real time, and to ensure data integrity and low latency during the transmission process. The control terminal processing module is the core scheduling hub of the system. It sends signals to the drilling parameter acquisition module, driving the underlying hardware to operate. It acquires real-time construction data through the communication interface. It sends instructions to the data processing module, distributes calculation tasks, receives processed feedback results, records and updates the data, and finally sends the final results to the visualization module. The data processing module is the core of the system's computation, interacting with the control terminal processing module through the task execution link. It receives raw drilling data distributed by the control terminal. Internally, it integrates a physical empirical formula solution unit, an expert experience discrimination unit, and an XGBoost residual learning model pre-trained based on a database. It verifies the model output against physical constraints and sends the verified high-confidence rock mass strength prediction back to the control terminal processing module. It also includes a visualization module, which connects to the control terminal processing module and other modules for human-computer interaction. It receives the final prediction data from the control terminal processing module, plots the drilling rock strength curve, formation profile, and confidence interval, allows users to input interactive commands such as parameter corrections, and feeds these commands back to the control terminal to form a closed-loop control.

[0035] The specific implementation steps include: Before the system goes into operation, the data processing module first establishes an initial model based on physical simulation.

[0036] A numerical model of drill bit-rock cutting was established using discrete element method (DEM) simulation software. Different rock micro-parameters and drilling conditions were set, generating over 5000 sets of virtual drilling data. The virtual data, along with a small amount of historical field-measured data, were integrated into a database. Parameters such as drill weight (WOB), rotational speed (RPM), and torque (Torque) were subjected to maximum-minimum normalization, uniformly mapping them to the [0,1] interval. The XGBoost algorithm was used to construct a basic predictor, with the objective function set to minimize the mean square error between the predicted value and the virtual rock mass strength, allowing the model to initially learn the physical and mechanical laws. Step 1: The control terminal processing module sends a control excitation signal to the drilling parameter acquisition module to initiate sampling. The acquisition module acquires the current drilling parameter vector in real time at a frequency of 1Hz (or higher). ; Step Two: The data communication transmission module transmits the drilling parameter vector to the data processing module. The data processing module first calls its internal physical empirical formula unit, and based on the Xianglushan mechanical specific energy formula, further simplifies the torque work to obtain the unit rock-breaking formula. Combined with the elastic strain energy formula We obtain the theoretical value of the rock mass strength at the current moment, i.e., the physical theoretical value.

[0037]

[0038] in: This is the preset mechanical efficiency coefficient. The drilling speed of the drill pipe. For torque, The diameter of the drill bit. For drilling speed, Uniaxial compressive strength of rocks with different lithologies The elastic modulus is the average value for different lithologies. This is the energy conversion coefficient. This value is not used as the final result, but rather as a strong feature input for the next stage.

[0039] Step 3: The data processing module will process the original parameter vector. The actual intensity of the drilling rig sampling and the calculated physical theoretical value Vector concatenation is performed, and a stuck drill risk indicator output by an expert experience judgment module is appended to the end. Advance marker Groundwater outcrop marker This allows for the construction of high-dimensional enhanced feature vectors. ; Step 4: The data processing module will... The input is fed into a pre-trained logistic constraint residual XGBoost learning model. The physical constraint residual XGBoost model is defined as learning the deviation between the true value and the physical theoretical value, and a physical consistency penalty term is introduced to meet the formation continuity requirement.

[0040] It should be noted that this model does not directly use the existing general XGBoost model to fit the absolute value of rock mass strength, but is defined as learning the true value. Compared with physical theoretical values The deviation between them. Based on the original loss term and complexity regularization term of the XGBoost model, a physical consistency penalty term is introduced to constrain the prediction results to meet the requirements of non-negativity of rock mass strength, a reasonable value range, and stratigraphic continuity. Multiple decision trees within the model determine paths based on the feature splitting gain. Because It includes the main physical trends, which the model automatically uses as the main split node, so that the decision tree learns the main trends determined by physical laws first, and then fine-tunes the local nonlinear deviations by construction parameters such as drilling pressure, rotation speed, and torque. The online fine-tuning process includes one or a combination of the following two methods.

[0041] The first method maps new samples to the leaf nodes of each tree in the current XGBoost model, and makes a slight correction to the output value of the leaf node based on the mean error of the new samples falling into the corresponding leaf node, so that the corrected leaf node output is more in line with the data distribution of the current construction section.

[0042] The second approach, while maintaining the original tree structure and most parameters unchanged, adds sample error. As a new learning objective, one or more shallow compensation decision trees are trained and appended to the output of the original model to compensate for local biases under the current operating conditions. After online fine-tuning, the system re-verifies the physical consistency of the updated model, and the verification is performed by outputting the predicted residual values ​​from the output model. By superimposing this value with the theoretical physical value, we obtain the preliminary prediction result: ; Step 5: Output Perform physical boundary verification. If the verification passes, output the final result directly. If the verification is "abnormal", trigger the "correction / smoothing" subroutine: use Kalman filtering or neighborhood weighted average method to correct the current abnormal point using the predicted values ​​of upstream and downstream depths.

[0043] The specific process of correcting using upstream and downstream depth prediction values ​​includes: reading the current abnormal depth point. Preliminary forecast value It reads the predicted values ​​of its neighboring upstream and downstream points; calculates the difference between the current point and the upstream and downstream points, as well as the deviation of the current point relative to the neighborhood mean; when the difference or deviation exceeds a preset threshold, the current point is determined to be a local anomaly; and constructs a neighborhood reference value based on the predicted values ​​of the upstream and downstream points. According to the preset fusion weights, and We perform weighted fusion to obtain the corrected prediction value. The corrected predicted values ​​are then re-input into the physical boundary verification module. If the verification passes, the final result is output; otherwise, smoothing correction is performed or the value is marked as a low-confidence point. This ensures that the output rock mass strength curve conforms to the continuity characteristics of geological sedimentation, and finally outputs rock mass strength prediction values ​​with clear physical meaning. Step Six: Visualization and Feedback: The control terminal processing module receives the final results and drives the visualization module. On the human-computer interface, a "dual-drive correction prediction curve" is drawn with a solid red line, while a "physical theory baseline curve" is drawn with a solid green line, visually demonstrating the model's correction effect on physical formula errors. When field engineers input new measured core data into the visualization interface, the control terminal automatically triggers the online transfer learning interface of the data processing module, using the new data to update the leaf node weights of the XGBoost model.

[0044] Example 3 A drilling rock mass strength prediction system based on both data and physics-driven approaches includes: The data acquisition module is configured to fuse the construction parameters collected in real time by the drilling rig with the physical intermediate variables calculated using the physical experience model to form a physically enhanced high-dimensional feature vector. The prediction and verification module is configured to take the physically enhanced high-dimensional feature vector as input, use the pre-trained residual learning model to make predictions, obtain preliminary prediction results, and perform physical constraint verification on the preliminary prediction results. The result output module is configured to take the preliminary prediction result as the final result if the physical constraint verification passes; otherwise, it will use the predicted values ​​of the upstream and downstream depths to correct the current anomaly point to ensure that the rock mass strength curve in the preliminary prediction result conforms to the continuity characteristics of geological sedimentation, and take the corrected prediction result as the final result. During the training process of the residual learning model, it is trained using a dual-drive objective function under physical constraints. The dual-drive objective function includes two aspects: prediction loss and physical regularization term. The forward distribution algorithm is used for multiple rounds of iterative training. In each round of iteration, the residual between the prediction result of the previous round and the true label is fitted by growing a decision tree, the gain score of feature splitting is calculated, and the physical benchmark feature is preferentially selected as the main splitting node.

[0045] Example 4 A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the steps in the method provided in Embodiment 1.

[0046] Example 5 An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the steps in the method provided in Embodiment 1.

[0047] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of one or more computer-usable storage media (including, but not limited to, disk storage, etc.) containing computer-usable program code. CD - ROM It takes the form of a computer program product implemented on (such as optical memory, etc.).

[0048] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0049] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0050] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0051] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art without creative effort within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting rock mass strength while drilling based on a dual-driven approach of data and physics, characterized in that, Includes the following steps: The construction parameters collected in real time by the drilling rig are fused with the physical intermediate variables calculated using a physical empirical model to form a physically enhanced high-dimensional feature vector. Using the physically enhanced high-dimensional feature vector as input, a pre-trained residual learning model is used to make predictions and obtain preliminary prediction results. The preliminary prediction results are then verified by physical constraints. If the physical constraint verification passes, the preliminary prediction result shall be taken as the final result; otherwise, the predicted values ​​of the upstream and downstream depths shall be used to correct the current anomaly point to ensure that the rock mass strength curve in the preliminary prediction result conforms to the continuity characteristics of geological sedimentation, and the corrected prediction result shall be taken as the final result. During the training process of the residual learning model, it is trained using a dual-drive objective function under physical constraints. The dual-drive objective function includes two aspects: prediction loss and physical regularization term. The forward distribution algorithm is used for multiple rounds of iterative training. In each round of iteration, the residual between the prediction result of the previous round and the true label is fitted by growing a decision tree, the gain score of feature splitting is calculated, and the physical benchmark feature is preferentially selected as the main splitting node.

2. The drilling rock mass strength prediction method based on data and physical dual-drive as described in claim 1, characterized in that, The process of fusing the construction parameters collected in real time by the drilling rig with the physical intermediate variables calculated using the physical experience model to form a physically enhanced high-dimensional feature vector includes: acquiring several data points from the current drilling pressure, rotation speed, torque, mechanical drilling speed and flushing fluid pressure during the drilling process; acquiring the actual strength of the drilling rig samples; acquiring the physical theoretical rock mass strength and derived fracture specific work calculated by the physical experience model; acquiring the stuck drill risk indicator, breakthrough indicator and groundwater outburst indicator identified by expert experience; and splicing and fusing these data to obtain the physically enhanced high-dimensional feature vector.

3. The method for predicting rock mass strength during drilling based on a dual data and physical drive as described in claim 2, characterized in that, The process of calculating the physical theoretical rock mass strength using a physical empirical model includes: simplifying the torque work formula based on the Xianglushan mechanical specific energy formula to obtain the unit rock breaking formula. Combined with the elastic strain energy formula Obtain the theoretical value of the rock mass strength at the current moment, i.e., the physical theoretical value: in: This is the preset mechanical efficiency coefficient. The drilling speed of the drill pipe. For torque, The diameter of the drill bit. For drilling speed, Uniaxial compressive strength of rocks with different lithologies The elastic modulus is the average value for different lithologies. This is the energy conversion coefficient.

4. The drilling rock mass strength prediction method based on data and physical dual-drive as described in claim 1, characterized in that, The process of constructing the dual-drive objective function under physical constraints includes: The dual-drive objective function is: ； in, The residual loss function is a second-order Taylor expansion, used to drive the model to fit the true value. y true Compared with physical theoretical values y phy Deviation; This is a model complexity regularization term used to prevent overfitting. f k For the model's first A decision tree; This is a physical consistency penalty term; it increases sharply when the predicted value violates the monotonicity of rock mechanics or becomes negative. These are the physical constraint weighting coefficients.

5. The drilling rock mass strength prediction method based on data and physical dual-drive as described in claim 1, characterized in that, During multi-round iterative training using the forward distribution algorithm, in the... In each iteration, the model does not directly predict the absolute value of rock mass strength, but instead grows a new decision tree. The residual between the previous prediction result and the true label is fitted. The gain score of feature splitting is calculated, and the physical theoretical value is selected as the main splitting node to ensure that the physical law occupies the dominant position in the tree structure. Drilling pressure and rotation speed are used to fine-tune the nonlinear deviation at the leaf nodes.

6. The drilling rock mass strength prediction method based on data and physical dual-drive as described in claim 1, characterized in that, The process of verifying the physical constraints of the preliminary prediction results includes: summing the scores of the leaf nodes of all iterative trees to obtain the preliminary rock mass strength prediction value; checking whether the preliminary rock mass strength prediction value falls within the preset valid value range; if not, it is considered that the verification has failed. Calculate the gradient of the change between the current depth and the predicted value of the previous depth. If the gradient exceeds the threshold allowed by the stratigraphic deposition law, the verification is considered to have failed.

7. The drilling rock mass strength prediction method based on data and physical dual-drive as described in claim 1, characterized in that, The process of correcting the current outlier using the predicted values ​​of upstream and downstream depths includes: using Kalman filtering or neighborhood weighted averaging to correct the current outlier using the predicted values ​​of upstream and downstream depths, ensuring that the output rock mass strength curve conforms to the continuity characteristics of geological sedimentation.

8. A drilling rock mass strength prediction system based on data and physical dual-drive, characterized in that, include: The data acquisition module is configured to fuse the construction parameters collected in real time by the drilling rig with the physical intermediate variables calculated using the physical experience model to form a physically enhanced high-dimensional feature vector. The prediction and verification module is configured to take the physically enhanced high-dimensional feature vector as input, use the pre-trained residual learning model to make predictions, obtain preliminary prediction results, and perform physical constraint verification on the preliminary prediction results. The result output module is configured to take the preliminary prediction result as the final result if the physical constraint verification passes; otherwise, it will use the predicted values ​​of the upstream and downstream depths to correct the current anomaly point to ensure that the rock mass strength curve in the preliminary prediction result conforms to the continuity characteristics of geological sedimentation, and take the corrected prediction result as the final result. In the training process of the residual learning model, it is trained with a dual-drive objective function under physical constraints. The dual-drive objective function includes two aspects: prediction loss and physical regularization term. It is trained in multiple rounds using the forward distribution algorithm. In each round of iteration, the residual between the prediction result of the previous round and the true label is fitted by growing a decision tree, the gain score of feature splitting is calculated, and the physical benchmark feature is preferentially selected as the main splitting node.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the steps in the drilling rock strength prediction method based on data and physical dual-drive as described in claim 1.

10. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the processor executes the computer instructions, it completes the steps in the data- and physical dual-driven drilling rock mass strength prediction method as described in claim 1.

Images