Surrounding rock parameter while-drilling dynamic inversion method based on migration enhancement multi-mode learning
Through the method of enhancing multimodal learning through migration, sensor calibration and data fusion are optimized, and surrounding rock parameters are monitored and dynamically updated, which solves the shortcomings in surrounding rock parameter measurement in the existing technology, and achieves high-precision and real-time surrounding rock parameter inversion, improving the safety and efficiency in fields such as geological exploration.
Patent Information
- Application Number
- CN202510417746.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-03
- Publication Date
- 2025-08-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, in geological exploration, mining and tunnel construction, the surrounding rock parameter measurement method is time-consuming and labor-intensive and difficult to reflect changes in the drilling process in real time. There are shortcomings in data acquisition, fusion processing, feature extraction and mapping, parameter inversion calculation, etc., which affects the safety and efficiency of the project.
The method of transfer-enhanced multimodal learning is adopted to monitor and dynamically update surrounding rock parameters through sensor calibration, data fusion, feature alignment and physical law constraints, including data acquisition optimization, fusion processing, feature extraction and mapping, parameter inversion calculation and dynamic update.
It improves the accuracy and efficiency of surrounding rock parameter measurement, ensures that the results comply with physical laws, has high output reliability, and improves the safety and efficiency in fields such as geological exploration.
Smart Images
Figure CN120541744A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of surrounding rock parameter measurement technology in the fields of geological exploration, mining, tunnel construction, etc., and in particular to a method for dynamic inversion of surrounding rock parameters while drilling based on transfer-enhanced multimodal learning. Background Art
[0002] In the fields of geological exploration, mining, tunnel construction, etc., accurate acquisition of surrounding rock parameters is of great significance for ensuring project safety and optimizing construction plans. Traditional surrounding rock parameter measurement methods mostly rely on laboratory testing or on-site sampling and analysis. These methods are not only time-consuming and labor-intensive, but also difficult to reflect the changes in surrounding rock parameters during the drilling process in real time. With the development of measurement while drilling technology, although some real-time data can be obtained during the drilling process, the existing technology still has shortcomings in terms of data acquisition accuracy, data processing depth, and parameter inversion precision.
[0003] Existing technologies have deficiencies in data acquisition, fusion processing, feature extraction and mapping, parameter inversion calculation, and dynamic updating, making it difficult to meet the needs of high-precision, real-time surrounding rock parameter inversion, affecting the efficiency and safety of geological exploration and other related work. Summary of the Invention
[0004] The main purpose of the present invention is to provide a dynamic inversion method for surrounding rock parameters while drilling based on transfer enhanced multimodal learning, which can effectively solve the problems existing in data acquisition, fusion processing, feature extraction and mapping, parameter inversion calculation and dynamic update.
[0005] To achieve the above object, the technical solution adopted by the present invention is: a method for dynamic inversion of surrounding rock parameters while drilling based on transfer-enhanced multimodal learning, comprising the following steps:
[0006] S1. Data acquisition optimization: Acquire sensor signals during drilling, compare them with the preset calibration model, adjust sensor parameters based on deviation values, dynamically adjust acquisition frequency based on drilling depth and formation changes, and generate calibrated acquisition data;
[0007] S2. Data fusion processing: Based on the calibration acquisition data, the formation density and hardness parameters in the geological exploration data are quantified, and the drilling characteristics and geological parameters are associated through feature correlation matrix mapping to obtain fused feature data;
[0008] S3. Feature extraction and mapping: Extract the main feature vectors from the fused feature data, use the feature alignment algorithm to match the feature vectors of the source domain and the target domain, adjust the model mapping relationship, and generate the aligned feature vectors.
[0009] S4, parameter inversion calculation: Call the aligned eigenvectors, integrate physical laws into the inversion model, monitor and automatically adjust model parameters in real time to ensure that the results conform to physical laws and are highly accurate, and output uncertainty assessment indicators;
[0010] S5. Dynamic update and adjustment: As drilling progresses, new calibration data is received and integrated to form an incremental data set. The inversion model is updated online using an incremental learning algorithm to generate an incremental updated model.
[0011] Preferably, in the data acquisition optimization step, the calculation formula of the deviation value is:
[0012]
[0013] Among them, S i Indicates the current output signal of the sensor, M i represents the corresponding signal of the preset calibration model, and n represents the number of sampling points.
[0014] Preferably, in the data fusion processing step, the characteristic correlation matrix is constructed using the principal component analysis method, and its calculation formula is:
[0015] A=X T X;
[0016] Among them, X represents the fusion feature data matrix, A represents the feature correlation matrix, X T Represents the transpose of the matrix X, that is, swapping the rows and columns of the original matrix.
[0017] Preferably, in the feature extraction and mapping steps, the feature alignment algorithm adopts Procrustes analysis, and its calculation formula is:
[0018]
[0019] Among them, F s represents the source domain feature vector, F t represents the target domain feature vector, represents the alignment transformation matrix, argmin Q Indicates the value of the corresponding independent variable (here is the matrix Q) when the following expression is minimized. The square of the Frobenius norm of a matrix is used to measure the degree of difference between two matrices.
[0020] Preferably, in the parameter inversion calculation step, the rock mechanics equilibrium equation of the physical law constraint condition is:
[0021]
[0022] in, It represents the divergence of the stress tensor σ, σ represents the stress tensor, b represents the body force, and 0 means that in the static equilibrium state, the stress and body force inside the rock are balanced.
[0023] Preferably, in the dynamic updating and adjusting step, the incremental learning algorithm adopts a small batch gradient descent strategy, and its update formula is:
[0024]
[0025] Among them, w k represents the model parameters of the kth iteration, η represents the learning rate, represents the gradient of the loss function.
[0026] Preferably, the data acquisition optimization step also includes automatically adjusting the layout and installation position of the sensors according to the operating status of the drilling equipment and the drilling plan, and the collected data includes but is not limited to extracting the mean, variance and peak value of drill bit pressure, torque and drilling speed.
[0027] Preferably, the parameter inversion calculation step further includes outputting an uncertainty evaluation index of the inversion result, the calculation formula of which is:
[0028]
[0029] Where P represents the physical constraint inversion result, and Var(P) represents the variance of the result.
[0030] Compared with the prior art, the present invention has the following beneficial effects:
[0031] 1. In the present invention, the method optimizes the sensor calibration process, uses deviation values to adjust parameters, and dynamically adjusts the acquisition frequency in combination with drilling depth and formation changes to generate high-precision calibration data, thereby improving the accuracy and timeliness of data acquisition. By extracting the statistical characteristics of drill bit pressure, torque, and drilling speed and quantifying geological exploration data, and using the feature association matrix to achieve mapping association, the fused data is made more comprehensive and representative, thereby enhancing the depth and breadth of data fusion.
[0032] 2. In the present invention, the method adopts a feature alignment algorithm to match and align the feature vectors of the source domain and the target domain in feature extraction and mapping, generates an aligned feature vector, and further improves the accuracy and adaptability of the features. In the digital inversion calculation, the physical laws are integrated into the model as constraints, and the parameters are monitored and automatically adjusted in real time to ensure that the results conform to the physical laws and are highly accurate. At the same time, the uncertainty assessment index is output to quantify the reliability of the results.
[0033] 3. This method utilizes an incremental learning algorithm to update the inversion model as drilling progresses, ensuring real-time and adaptability. Overall, this approach significantly improves the accuracy, efficiency, and reliability of dynamic inversion of surrounding rock parameters while drilling, providing stronger technical support for geological exploration and other related fields. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION
[0035] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.
[0036] In the first embodiment, in the data acquisition optimization step, during the drilling process, sensors (including pressure sensors, torque sensors, and drilling speed sensors) collect output signals in real time. The sampling frequency of these sensors can be set according to actual needs, generally between 1 and 5 times per second. The system compares the collected sensor signals with a preset calibration model. The preset calibration model is established based on laboratory calibration and historical drilling data and can provide accurate reference signals. The deviation value is calculated as follows:
[0037]
[0038] Among them, S i Indicates the current output signal of the sensor, M i represents the corresponding signal of the preset calibration model, and n represents the number of sampling points.
[0039] For example, in a drilling experiment, the pressure sensor collected 10 signal points, namely 101, 102, 100, 103, 104, 99, 100, 101, 102, 103, with the unit being kPa. The preset calibration model corresponds to the signal of 100, 100, 100, 100, 100, 100, 100, 100, 100, 100. The deviation value is calculated as:
[0040]
[0041] This result indicates that the sensor's current output signal deviates from the preset calibration model, necessitating adjustments to the sensor's acquisition parameters, such as gain and bias. Based on a deviation of 1.6, the system automatically adjusts these parameters. If the deviation exceeds a preset threshold (e.g., 0.5), the sensor is deemed to require calibration. The system also dynamically adjusts the acquisition frequency based on drilling depth and formation changes. In critical formations (e.g., those with high hardness or complex geological structures), the acquisition frequency is increased to 2 times per second; in less complex formations, the acquisition frequency is maintained at 1 time per second. This method generates calibration data.
[0042] In the second embodiment, in the data fusion processing step, based on the calibration data, the system extracts the mean, variance, and peak of the drill bit pressure, torque, and drilling speed. For the drill bit pressure data, the formula for calculating its mean is:
[0043]
[0044] Where Pi represents the i-th data point of the drill bit pressure, and n represents the total number of data points. Assuming that the 10 data points of the drill bit pressure in the calibration data are 101, 102, 100, 103, 104, 99, 100, 101, 102, and 103 kPa, the mean is calculated as:
[0045]
[0046] The results show that the average drill bit pressure is 101.5 kPa, which can be used as the basic data for subsequent analysis to quantify the formation density and hardness parameters in geological exploration data. For example, the formation density data is obtained through laboratory measurement in units of g / cm 3 The hardness data is measured by a Mohs hardness tester to obtain the hardness grade value. The drilling process characteristics are mapped and associated with the geological parameters through the characteristic correlation matrix to obtain the fusion feature data. The characteristic correlation matrix is constructed using the principal component analysis method, and its calculation formula is:
[0047] A=X T X
[0048] Among them, X represents the fusion feature data matrix, A represents the feature correlation matrix, X T Represents the transpose of the matrix X, that is, swapping the rows and columns of the original matrix.
[0049]
[0050] Then the feature association matrix A is calculated as:
[0051]
[0052] The results show that the feature correlation matrix constructed by the principal component analysis method can effectively map the drilling process characteristics with geological parameters, providing a basis for subsequent feature extraction and mapping steps.
[0053] In the third embodiment, the feature extraction and mapping step extracts the main feature vectors from the fused feature data, uses a feature alignment algorithm to align the source domain feature vectors with the target domain feature vectors, adjusts the mapping relationship of the model, and generates an aligned feature vector. The feature alignment algorithm uses Procrustes analysis, and its calculation formula is:
[0054]
[0055] Among them, F s represents the source domain feature vector, F t represents the target domain feature vector, represents the alignment transformation matrix, argmin Q Indicates the value of the corresponding independent variable (here is the matrix Q) when the following expression is minimized. It represents the square of the Frobenius norm of the matrix and is used to measure the degree of difference between two matrices. In this way, the feature vectors of different domains can be effectively aligned to improve the accuracy and adaptability of the features.
[0056] In the fourth embodiment, the parameter inversion calculation step calls the aligned eigenvectors, incorporates physical laws such as geomechanics and rock mechanics as constraints into the parameter inversion model, monitors the degree of conformity of the model output results with the physical laws in real time, and automatically adjusts the model parameters or structure when deviations occur to obtain the physical constraint inversion results. The rock mechanics equilibrium equation under the physical law constraints is:
[0057]
[0058] in, It represents the divergence of the stress tensor σ, σ represents the stress tensor, b represents the body force, and 0 means that in the static equilibrium state, the stress and body force inside the rock are balanced.
[0059] At the same time, the uncertainty evaluation index of the inversion result is output, and its calculation formula is:
[0060]
[0061] Among them, P represents the physical constraint inversion result, and Var(P) represents the variance of the result. In this way, it is ensured that the inversion result conforms to the physical laws and has high accuracy, while the reliability of the result is quantified.
[0062] In the fifth embodiment, the dynamic update and adjustment step is to continuously receive new calibration data as drilling progresses, integrate the new data with the previous data to form an incremental data set, and use the incremental learning algorithm to update the inversion model online to generate an incremental update model. The incremental learning algorithm adopts a small batch gradient descent strategy, and its update formula is:
[0063]
[0064] Among them, w k represents the model parameters of the kth iteration, η represents the learning rate, It represents the gradient of the loss function. In this way, the real-time and adaptability of the model are guaranteed, and it can timely reflect the changes in surrounding rock parameters during the drilling process.
[0065] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.
Claims
1. A dynamic inversion method for surrounding rock parameters while drilling based on transfer-enhanced multimodal learning, characterized by: The following steps are involved: S1. Data acquisition optimization: Acquire sensor signals during drilling, compare them with the preset calibration model, adjust sensor parameters based on deviation values, dynamically adjust acquisition frequency based on drilling depth and formation changes, and generate calibrated acquisition data; S2. Data fusion processing: Based on the calibration acquisition data, the formation density and hardness parameters in the geological exploration data are quantified, and the drilling characteristics and geological parameters are associated through feature correlation matrix mapping to obtain fused feature data; S3. Feature extraction and mapping: Extract the main feature vectors from the fused feature data, use the feature alignment algorithm to match the feature vectors of the source domain and the target domain, adjust the model mapping relationship, and generate the aligned feature vectors. S4, parameter inversion calculation: Call the aligned eigenvectors, integrate physical laws into the inversion model, monitor and automatically adjust model parameters in real time to ensure that the results conform to physical laws and are highly accurate, and output uncertainty assessment indicators; S5. Dynamic update and adjustment: As drilling progresses, new calibration data is received and integrated to form an incremental data set. The inversion model is updated online using an incremental learning algorithm to generate an incremental updated model.
2. The method for dynamic inversion of surrounding rock parameters while drilling based on transfer-enhanced multimodal learning according to claim 1 is characterized in that: In the data acquisition optimization step, the calculation formula of the deviation value is: Among them, S i Indicates the current output signal of the sensor, M i represents the corresponding signal of the preset calibration model, and n represents the number of sampling points.
3. The method for dynamic inversion of surrounding rock parameters while drilling based on transfer-enhanced multimodal learning according to claim 1 is characterized in that: In the data fusion processing step, the characteristic correlation matrix is constructed using the principal component analysis method, and its calculation formula is: A=X T X; Among them, X represents the fusion feature data matrix, A represents the feature correlation matrix, X T Represents the transpose of the matrix X, that is, swapping the rows and columns of the original matrix.
4. The method for dynamic inversion of surrounding rock parameters while drilling based on transfer-enhanced multimodal learning according to claim 1 is characterized in that: In the feature extraction and mapping steps, the feature alignment algorithm uses Procrustes analysis, and its calculation formula is: Among them, F s represents the source domain feature vector, F t represents the target domain feature vector, represents the alignment transformation matrix, argmin Q Indicates the value of the corresponding independent variable (here is the matrix Q) when the following expression is minimized. The square of the Frobenius norm of a matrix is used to measure the degree of difference between two matrices.
5. The method for dynamic inversion of surrounding rock parameters while drilling based on transfer-enhanced multimodal learning according to claim 1 is characterized in that: In the parameter inversion calculation step, the rock mechanics equilibrium equation under the physical law constraint condition is: in, It represents the divergence of the stress tensor σ, σ represents the stress tensor, b represents the body force, and 0 means that in the static equilibrium state, the stress and body force inside the rock are balanced.
6. The method for dynamic inversion of surrounding rock parameters while drilling based on transfer-enhanced multimodal learning according to claim 1, characterized in that: In the dynamic update and adjustment steps, the incremental learning algorithm adopts a small batch gradient descent strategy, and its update formula is: Among them, w k represents the model parameters of the kth iteration, η represents the learning rate, represents the gradient of the loss function.
7. The method for dynamic inversion of surrounding rock parameters while drilling based on transfer-enhanced multimodal learning according to claim 1, characterized in that: The data acquisition optimization step also includes automatically adjusting the layout and installation position of sensors according to the operating status of the drilling equipment and the drilling plan. The collected data includes but is not limited to extracting the mean, variance and peak value of drill bit pressure, torque and drilling speed.
8. The method for dynamic inversion of surrounding rock parameters while drilling based on transfer-enhanced multimodal learning according to claim 1 is characterized in that: The parameter inversion calculation step also includes outputting an uncertainty evaluation index of the inversion result, the calculation formula of which is: Where P represents the physical constraint inversion result, and Var(P) represents the variance of the result.