Rock mass weathering identification method based on machine learning
Through machine learning-based methods, a rock mass weathering prediction model using drilling data is constructed, which solves the problem of low recognition efficiency of rock mass weathering in the existing technology, and achieves accurate prediction and recognition efficiency of rock mass weathering.
Patent Information
- Application Number
- CN202411852541.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-16
- Publication Date
- 2025-04-08
AI Technical Summary
The prior art is difficult to quickly and accurately use data while drilling to predict the degree of weathering of rock mass, resulting in inefficient identification of weathering.
Using machine learning-based methods, a rock mass weathering prediction model using feature extraction layer and recognition layer is constructed through data outliers and standardized processing, and a random search and cross-validation method is combined for model training.
Accurate prediction of the weathering degree of unexcavated palmate rock mass is achieved, and the identification efficiency and generalization ability and robustness of the model are improved.
Smart Images
Figure CN119782947A_ABST
Abstract
Description
Technical Field
[0001] The present specification relates to the technical field of rock weathering identification, and in particular to a rock weathering identification method based on machine learning. Background Art
[0002] Rock weathering is the process of rock mass being broken and decomposed by physical, chemical and biological effects, which affects its mechanical properties. The degree of weathering can be divided by qualitative (such as appearance characteristics) and quantitative methods (such as wave velocity ratio, etc.). Traditional judgment relies on manual discrimination, laboratory analysis and field measurement. Although the rebound hammer is highly accurate, it is often time-consuming and difficult to monitor in real time. Measurement while drilling (MWD) technology obtains key data such as drilling pressure and rotation speed in real time through sensors, providing key information for the analysis of rock mass mechanical properties. It has the characteristics of high precision, multi-information and integration, and can significantly improve the recognition efficiency. In recent years, the measurement while drilling technology has been widely used in the characterization of rock mass excavation process, including rock type, construction area stability, support design and rock mass strength parameters such as uniaxial compressive strength and shear strength. Therefore, it has become a reality to use the data of drilling while drilling to predict the degree of rock mass weathering in real time. The widespread application of machine learning has promoted the development of data-driven rock mass characteristic assessment. However, there are still few studies on the direct correlation between MWD data and rock weathering process, and data-driven intelligent models for rapid and accurate assessment of weathering degree have not been widely established. It is urgent to explore how to use machine learning technology to achieve intelligent prediction of rock weathering degree. Summary of the invention
[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a rock weathering identification method based on machine learning, which solves the problem that it is difficult to accurately identify the degree of rock weathering.
[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a rock weathering identification method based on machine learning, comprising:
[0005] S1: removing outliers from rock mass data through data outlier and standardization processing to obtain rock mass weathering sample data; the rock mass weathering sample data includes a training set and a test set;
[0006] S2: constructing a rock weathering prediction model using a feature extraction layer and an identification layer; wherein the identification layer includes a plurality of decision trees;
[0007] S3: inputting the training set into the rock weathering prediction model, and performing training using a random search and cross-validation method to obtain a trained rock weathering prediction model;
[0008] S4: Use the trained rock weathering prediction model to analyze the drilling data, obtain the rock weathering identification results, and complete the identification of rock weathering.
[0009] Further, the S1 includes:
[0010] Based on the rock mass acquisition data, an isolated decision tree is constructed by randomly selecting features and segmentation points to obtain an average path length of each data point;
[0011] Using the average path length of each data point, an anomaly score is obtained through calculation;
[0012] Based on the anomaly score, the outliers of the rock mass acquisition data are removed to obtain normal rock mass acquisition data;
[0013] The normal rock mass collected data is standardized to obtain rock mass weathering sample data.
[0014] Furthermore, the expression of the average path length of each data point is:
[0015]
[0016] Among them, E(h(x i )) represents x i The average path length, h(x i ) represents x i Relative to the path length of other data points, x i represents the data collected from the ith rock mass, H represents the harmonic number, and N represents the total number of samples;
[0017] The expression of the abnormality score is:
[0018]
[0019] Among them, s(x i ,N) represents the anomaly score, c(N) represents the normalization factor;
[0020] The expression of the rock weathering sample data is:
[0021]
[0022] in, Represents rock weathering sample data, x drill_i represents the normal rock mass acquisition data, μ drill_i Represents the mean value of the data characteristics of normal rock mass, σ drill_i Represents the standard deviation of the data characteristics collected for normal rock mass.
[0023] Furthermore, the rock weathering prediction model includes:
[0024] A feature extraction layer is used to extract features from the rock weathering sample data to obtain a rock weathering feature vector;
[0025] The identification layer is used to classify the rock weathering feature vectors using decision trees, and obtain rock weathering identification results by voting on all decision trees.
[0026] Further, the S3 includes:
[0027] S310: randomly combining hyperparameters of the rock weathering prediction model to obtain multiple candidate rock weathering prediction models;
[0028] S320: inputting the training set into the candidate rock weathering prediction model for training to obtain a corresponding trained rock weathering prediction model;
[0029] S330: Use the test set to perform performance evaluation on the trained rock weathering prediction model to obtain a performance evaluation result; when the performance evaluation results are all lower than a threshold, the trained rock weathering prediction model with the highest performance evaluation result is input into S310 for retraining; when the performance evaluation result is not lower than a threshold, the corresponding trained rock weathering prediction model is used as a trained rock weathering prediction model.
[0030] Furthermore, the expression of the performance evaluation result is:
[0031]
[0032] Among them, MacroF1 represents the performance evaluation result, n represents the total number of rock weathering degree categories, Represents the performance evaluation result of the kth category.
[0033] The beneficial effects of the present invention are as follows: a rock weathering identification method based on machine learning, which uses the MWD data collected during the drilling and blasting process to predict the weathering degree of the rock mass in the unexcavated face. (1) By removing outliers, the statistical characteristics and distribution stability of the rock weathering sample data can be improved, so that the rock weathering prediction model can more effectively capture and learn the key weathering degree characteristics; (2) The rock weathering prediction model is constructed by using multiple independent decision trees, which reduces the instability of a single decision tree and improves the generalization ability and robustness of the overall rock weathering prediction model; (3) The rock weathering prediction model is trained by using random search and cross-validation methods, which effectively reduces the model's dependence on specific data sets, enhances the generalization ability and reliability of the rock weathering prediction model, and improves performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:
[0035] Figure 1 It is an exemplary flow chart of a rock weathering identification method based on machine learning according to some embodiments of this specification. DETAILED DESCRIPTION
[0036] The specific implementation modes of the present invention are described below to facilitate those skilled in the art to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0037] Example
[0038] Figure 1 is an exemplary flow chart of a rock weathering identification method based on machine learning according to some embodiments of this specification. Figure 1 As shown, the process includes the following steps. In some embodiments, the process can be executed by a processor.
[0039] S1: Through data outlier and standardization processing, outliers of rock collection data are removed to obtain rock weathering sample data.
[0040] Rock mass acquisition data refers to parameter data that affects rock mass weathering during the drilling process. For example, rock mass acquisition data may include thrust velocity, impact pressure, thrust pressure, rotation pressure, water pressure, and water flow.
[0041] In some embodiments, the processor may utilize a data while drilling experiment to obtain measurement while drilling (MWD) data of drill and blast construction as rock mass acquisition data.
[0042] Rock weathering sample data is a set of sample data used to train and test rock weathering prediction models. For example, rock weathering sample data may include a training set and a test set.
[0043] In some embodiments, the processor can implement S1 based on the following steps: based on the rock mass acquisition data, construct an isolated decision tree by randomly selecting features and segmentation points to obtain the average path length of each data point; use the average path length of each data point to obtain an anomaly score through calculation; based on the anomaly score, remove the outliers of the rock mass acquisition data to obtain normal rock mass acquisition data; standardize the normal rock mass acquisition data to obtain rock weathering sample data.
[0044] The average path length of each data point is the average of the distances from each data point to other data points.
[0045] In some embodiments, the processor may construct an isolated decision tree, collect data for each rock mass, recursively segment the data by randomly selecting features and segmentation points until each data point is individually isolated, and calculate the average path length of each data point.
[0046] In some embodiments, the expression for the average path length of each data point may be:
[0047]
[0048] Among them, E(h(x i )) represents x i The average path length, h(x i ) represents x i Relative to the path length of other data points, x i represents the data collected from the ith rock mass, H represents the harmonic number, and N represents the total number of samples.
[0049] Anomaly score is a score used to evaluate the abnormality of a data point.
[0050] In some embodiments, the expression of the anomaly score may be:
[0051]
[0052] Among them, s(x i ,N) represents the anomaly score, and c(N) represents the normalization factor.
[0053] The normalization factor is used to normalize the average path length to the interval between 0 and 1.
[0054] Normal rock mass acquisition data is the rock mass acquisition data after outliers are removed.
[0055] In some embodiments, the processor may regard data points with anomaly scores not less than 0.5 as anomalies, and data points with anomaly scores less than 0.5 as normal points, remove the anomalies of the rock mass acquisition data, and obtain normal rock mass acquisition data.
[0056] The rock weathering sample data is standardized normal rock collection data. For example, the rock weathering sample data may include a training set and a test set.
[0057] In some embodiments, the processor may randomly divide the rock weathering sample data into a training set and a test set of equal size in a ratio of 9:1.
[0058] In some embodiments, the expression of rock weathering sample data may be:
[0059]
[0060] in, Represents rock weathering sample data, x drill_i represents the normal rock mass acquisition data, μ drill_i Represents the mean value of the data characteristics of normal rock mass, σ drill_i Represents the standard deviation of the data characteristics collected for normal rock mass.
[0061] S2: Use the feature extraction layer and identification layer to build a rock weathering prediction model.
[0062] The rock weathering prediction model is used to predict the weathering of the rock mass. There are many types of rock weathering prediction models. For example, the types of rock weathering prediction models may include random forest classifiers, deep neural network models, etc.
[0063] In some embodiments, the input of the rock weathering prediction model may be rock weathering sample data, and the output of the rock weathering prediction model may be a rock weathering identification result.
[0064] In some embodiments, the structure of the rock weathering prediction model is as follows: the rock weathering prediction model includes a feature extraction layer and a recognition layer. The output of the feature extraction layer is used as the input of the recognition layer, and the output of the recognition layer is used as the final output of the rock weathering prediction model.
[0065] The feature extraction layer is used to extract features from the rock weathering sample data to obtain a rock weathering feature vector. The input of the feature extraction layer may include the rock weathering sample data, and the output may include the rock weathering feature vector.
[0066] The rock mass weathering characteristic vector is a vector of characteristic components that reflects the weathering conditions of the rock mass.
[0067] The identification layer is used to classify the rock weathering feature vector using a decision tree, and obtain the rock weathering identification result by voting on all decision trees. The input of the identification layer may include the rock weathering feature vector, and the output may include the rock weathering identification result.
[0068] In some embodiments, the recognition layer may include multiple decision trees.
[0069] The rock mass weathering identification result is an identification result that reflects the weathering condition of the sample data. For example, the rock mass weathering identification result may include strong weathering, medium weathering, and slight weathering.
[0070] S3: Inputting the training set into the rock weathering prediction model, and training using random search and cross-validation methods to obtain a trained rock weathering prediction model.
[0071] In some embodiments, the processor may implement S3 based on the following steps.
[0072] S310: Randomly combine hyperparameters of the rock weathering prediction model to obtain multiple candidate rock weathering prediction models.
[0073] S320: Inputting the training set into the candidate rock weathering prediction model for training to obtain a corresponding trained rock weathering prediction model.
[0074] In some embodiments, the processor can input the labeled training set into the initial rock weathering prediction model, construct a loss function through the label and the result of the initial rock weathering prediction model, and iteratively update the parameters of the initial rock weathering prediction model based on the loss function through gradient descent or other methods. When the preset conditions are met, the model training is completed and a trained rock weathering prediction model is obtained. Among them, the preset conditions can be the convergence of the loss function, the number of iterations reaches a threshold, etc., the label can be the corresponding weathering degree, and the label can be manually marked.
[0075] S330: Use the test set to perform performance evaluation on the trained rock weathering prediction model to obtain a performance evaluation result; when the performance evaluation results are all lower than a threshold, the trained rock weathering prediction model with the highest performance evaluation result is input into S310 for retraining; when the performance evaluation result is not lower than a threshold, the corresponding trained rock weathering prediction model is used as a trained rock weathering prediction model.
[0076] The performance evaluation results are accuracy indicators used to evaluate the performance of the trained rock weathering prediction model.
[0077] In some embodiments, the expression of the performance evaluation result may be:
[0078]
[0079] Among them, MacroF1 represents the performance evaluation result, n represents the total number of rock weathering degree categories, Represents the performance evaluation result of the kth category.
[0080] S4: Use the trained rock weathering prediction model to analyze the drilling data, obtain the rock weathering identification results, and complete the identification of rock weathering.
[0081] In some embodiments, the processor can input the drilling data into a trained rock weathering prediction model, use multiple isolated decision trees to classify the drilling data separately, and obtain the prediction results of each decision tree; based on the prediction results of each decision tree, the rock weathering identification results are determined by the majority voting principle to complete the identification of rock weathering.
[0082] In some embodiments, the expression of rock weathering identification result can be:
[0083]
[0084] Among them, MacroF1 represents the performance evaluation result, n represents the total number of rock weathering degree categories, Represents the performance evaluation result of the kth category.
[0085] In some embodiments of the present specification, a rock weathering identification method based on machine learning uses the MWD data collected during drilling and blasting to predict the weathering degree of the unexcavated face. (1) By removing outliers, the statistical characteristics and distribution stability of the rock weathering sample data can be improved, so that the rock weathering prediction model can more effectively capture and learn the key weathering degree characteristics; (2) The rock weathering prediction model is constructed by using multiple independent decision trees, which reduces the instability of a single decision tree and improves the generalization ability and robustness of the overall rock weathering prediction model; (3) The rock weathering prediction model is trained and optimized by using random search and cross-validation methods, which effectively reduces the model's dependence on specific data sets, enhances the generalization ability and reliability of the rock weathering prediction model, and improves performance.
Claims
1. A rock weathering identification method based on machine learning, characterized in that: include: S1: removing outliers from rock mass data through data outlier and standardization processing to obtain rock mass weathering sample data; the rock mass weathering sample data includes a training set and a test set; S2: constructing a rock weathering prediction model using a feature extraction layer and an identification layer; wherein the identification layer includes a plurality of decision trees; S3: inputting the training set into the rock weathering prediction model, and performing training using a random search and cross-validation method to obtain a trained rock weathering prediction model; S4: Use the trained rock weathering prediction model to analyze the drilling data, obtain the rock weathering identification results, and complete the identification of rock weathering.
2. The rock weathering identification method based on machine learning according to claim 1 is characterized in that: The S1 includes: Based on the rock mass acquisition data, an isolated decision tree is constructed by randomly selecting features and segmentation points to obtain an average path length of each data point; Using the average path length of each data point, an anomaly score is obtained through calculation; Based on the anomaly score, the outliers of the rock mass acquisition data are removed to obtain normal rock mass acquisition data; The normal rock mass collected data is standardized to obtain rock mass weathering sample data.
3. The rock weathering identification method based on machine learning according to claim 2 is characterized in that: The expression of the average path length of each data point is: Among them, E(h(x i )) represents x i The average path length, h(x i ) represents x i Relative to the path length of other data points, x i represents the data collected from the ith rock mass, H represents the harmonic number, and N represents the total number of samples; The expression of the abnormality score is: Among them, s(x i ,N) represents the anomaly score, c(N) represents the normalization factor; The expression of the rock weathering sample data is: in, Represents rock weathering sample data, x drill_i represents the normal rock mass acquisition data, μ drill_i Represents the mean value of the data characteristics of normal rock mass, σ drill_i Represents the standard deviation of the data characteristics collected for normal rock mass.
4. The rock weathering identification method based on machine learning according to claim 1 is characterized in that: The rock weathering prediction model includes: A feature extraction layer is used to extract features from the rock weathering sample data to obtain a rock weathering feature vector; The identification layer is used to classify the rock weathering feature vectors using decision trees, and obtain rock weathering identification results by voting on all decision trees.
5. The rock weathering identification method based on machine learning according to claim 1 is characterized in that: The S3 includes: S310: randomly combining hyperparameters of the rock weathering prediction model to obtain multiple candidate rock weathering prediction models; S320: inputting the training set into the candidate rock weathering prediction model for training to obtain a corresponding trained rock weathering prediction model; S330: Use the test set to perform performance evaluation on the trained rock weathering prediction model to obtain a performance evaluation result; when the performance evaluation results are all lower than a threshold, the trained rock weathering prediction model with the highest performance evaluation result is input into S310 for retraining; when the performance evaluation result is not lower than a threshold, the corresponding trained rock weathering prediction model is used as a trained rock weathering prediction model.
6. The rock weathering identification method based on machine learning according to claim 5 is characterized in that: The expression of the performance evaluation result is: Among them, MacroF1 represents the performance evaluation result, n represents the total number of rock weathering degree categories, Represents the performance evaluation result of the kth category.