An intelligent prediction method for surrounding rock grade based on TSP-geological information fusion

Abstract

The present application relates to the field of tunnel advanced geological prediction, and provides a surrounding rock grade intelligent prediction method based on TSP-geological information fusion, first extracts TSP trend characteristics, and carries out spatial registration with the face mileage in the geological information; then, abnormal data screening rules are formulated for the TSP trend characteristics, and abnormal data is eliminated; on this basis, the processed TSP trend characteristics and the surrounding rock grade of the construction surface are used as model inputs, different construction surface categories are divided to model, and the TSP detection data distribution under different construction surface conditions is different, a surrounding rock grade prediction model is established by combining knowledge driving and data driving. Finally, the surrounding rock information in the design stage is added to the TSP single prediction result for intelligent decision-making, so as to realize the surrounding rock grade prediction. The present application can effectively improve the prediction accuracy, and has important significance for protecting the life and property safety of tunnel construction.

Description

Technical Field

[0001] This invention relates to the field of tunnel advanced geological prediction, and in particular to an intelligent prediction method for surrounding rock level based on TSP-geological information fusion. Background Technology

[0002] Rock mass level prediction accompanies the entire tunnel excavation process. It is a crucial comprehensive indicator in advanced geological prediction, requiring consideration of multiple factors, including the utilization of geological survey and design data, the interpretation of various geophysical data, and the analysis of actual groundwater conditions. The involvement of diverse and heterogeneous detection data poses a significant challenge to intelligent rock mass level prediction. Traditional methods rely on professionals to assess rock mass level based on specific detection data combined with experience and expertise. This approach is highly subjective, costly, and inefficient. With technological advancements, more researchers are using neural network technology to build intelligent prediction models for rock mass level prediction. However, current models still have several shortcomings: 1. The heterogeneous nature of the detection data makes it difficult to extract effective features, resulting in generally low data quality; 2. Single prediction methods based on TSP (Tunneling Seed Probe) suffer from multiple solutions, leading to insufficient prediction accuracy. Therefore, to ensure the safety and efficiency of tunnel construction, it is urgent to research an intelligent rock mass level prediction method based on TSP-geological information fusion. Summary of the Invention

[0003] To address the challenges of existing tunnel advanced geological prediction and detection data, such as the coexistence of multi-source heterogeneous data, the lack of significant differences in data characteristics leading to difficulties in effectively predicting surrounding rock levels, and the ambiguity of single prediction methods, this invention provides an intelligent prediction method for surrounding rock levels based on TSP-geological information fusion. This method improves the quality of TSP detection data and enhances the accuracy of surrounding rock level prediction by extracting features from TSP detection data, filtering data based on expert knowledge, considering different sub-models based on the construction face layering approach, and incorporating surrounding rock information from the design phase for intelligent decision-making. This is of great significance for ensuring the safety and efficiency of tunnel construction.

[0004] The technical solution of the present invention is as follows:

[0005] A smart prediction method for surrounding rock level based on TSP-geological information fusion includes the following steps:

[0006] S1: The TSP trend features are obtained by using the feature extraction method of the application surface difference and then normalized.

[0007] S2: Obtain different tunnel face mileage points from geological information, spatially register them with the TSP trend features obtained in S1, and form a dataset of TSP trend features and surrounding rock level. For the obtained dataset, formulate TSP trend feature abnormal data filtering rules and remove abnormal data that do not meet the requirements.

[0008] S3: Based on the aforementioned feature extraction and abnormal data filtering methods, obtain the effective feature data corresponding to each working face; stratify the surrounding rock level of the working face, and adopt a combination of knowledge-driven and data-driven approaches to establish a TSP surrounding rock level prediction sub-model according to the stratified levels, based on the trend characteristics and information of the surrounding rock level of the working face; finally, incorporate the surrounding rock information from the design stage and merge it with the TSP surrounding rock level prediction sub-model to form a TSP-geological fusion model for intelligent decision-making;

[0009] S4: Use the trained TSP surrounding rock level prediction sub-model and TSP-geological fusion model to perform advanced geological prediction for actual tunnel projects, and verify the model.

[0010] Furthermore, step S1 specifically includes the following processes:

[0011] S1.1: Extract the trend features of the TSP probe data, as shown below:

[0012]

[0013] in, For TSP data trend characteristics, TSP probes raw data for each point. This refers to the TSP detection data of the operation surface during this TSP detection.

[0014] S1.2: Then, the extracted trend features are standardized to obtain the final TSP trend feature library. The standardization formula is as follows:

[0015]

[0016] Where Z represents the standardized result of the TSP trend feature. Trend features extracted from TSP detection data. The mean of the TSP trend characteristics. The standard deviation represents the trend characteristics of TSP.

[0017] Furthermore, step S2 specifically includes the following processes:

[0018] S2.1: Spatially align the trend characteristics of TSP probe data with the tunnel face mileage, as shown below:

[0019]

[0020] in The tunnel mileage is aligned with the actual working face mileage. This refers to the actual mileage at the working face. Let be the tunnel mileage corresponding to the i-th TSP trend feature data. The direction of tunnel excavation is indicated by 0 for a small mileage and 1 for a large mileage.

[0021] S2.2: When using TSP trend characteristic data for advanced geological prediction, if the surrounding rock grade of the working face has been determined, and the grade increases relative to the working face but the actual surrounding rock condition is worse, or decreases relative to the working face but the actual surrounding rock condition is better, such trend characteristic data that do not conform to the mechanism should be removed. The specific mechanism should meet the following conditions:

[0022]

[0023]

[0024] in, For the P-wave velocity at a certain mileage point, The longitudinal wave velocity of the application surface, This represents the change in P-wave velocity relative to the work surface at a given mileage point. Data representing the abnormal trend characteristics of TSPs that have been removed. The actual surrounding rock grade, The grade of the surrounding rock at the construction surface.

[0025] Furthermore, step S3 specifically includes the following processes:

[0026] S3.1: Based on the feature matrix extraction and abnormal data filtering method, obtain the TSP trend features for each working face, form a dataset and divide it into training set, validation set and test set;

[0027] S3.2: The surrounding rock level of the construction face is divided into four levels: I, II, III, IV, V and VI. The TSP trend feature matrix and the surrounding rock level are used to build a model. For construction faces that can be directly divided, expert experience is used for judgment. For those that cannot be directly divided, a data-driven approach is used for modeling to obtain the TSP surrounding rock level prediction sub-model.

[0028] S3.3: Based on the TSP surrounding rock level prediction sub-model established for different construction faces, a surrounding rock result is obtained. Then, the surrounding rock information in the design stage is incorporated into the TSP surrounding rock level model prediction result for intelligent decision-making, resulting in a surrounding rock level prediction model based on TSP-geological fusion. The specific decision-making mechanism is as follows:

[0029] If the TSP forecast result differs too much from the surrounding rock level in the design stage, the TSP forecast result shall prevail; if the TSP forecast result is not much different from the surrounding rock level in the design stage, different weights shall be assigned for comprehensive decision-making, with the weights of the surrounding rock level in the design stage and the TSP forecast model being 0.4 and 0.6, respectively.

[0030] Furthermore, step S4 specifically includes the following processes:

[0031] S4.1: Use the TSP surrounding rock level prediction sub-model and the TSP-geological fusion model to predict the test set data in S3 and verify the effectiveness of the model;

[0032] S4.2: Use the test set for validation, with the following validation metrics:

[0033]

[0034] in It is the accuracy on the test set. It is the number of correctly predicted samples. It represents the total number of samples in the test set.

[0035] The beneficial effects of the technical solution provided by this invention are as follows:

[0036] This invention proposes a method for extracting trend features from TSP (Transformation and Spatial Probing) data. To improve data quality, an outlier screening method for TSP trend features is proposed based on expert experience. Considering the ambiguity of single forecasting methods and their difficulty in accurately predicting the surrounding rock grade ahead of the tunnel face, an intelligent forecasting method for surrounding rock grade based on TSP-geological information fusion is proposed. First, the method is stratified according to different working faces. A knowledge-driven and data-driven approach is used to establish a forecasting model for surrounding rock grade under different geological conditions of TSP data and working faces. Then, surrounding rock information from the design stage is incorporated for intelligent decision-making to achieve accurate forecasting of surrounding rock grade. The model's prediction effect is verified by simulation using test set data, which is beneficial for the application of this invention in actual production. Attached Figure Description

[0037] Figure 1 This is an architecture diagram of a smart prediction method for surrounding rock level based on TSP-geological information fusion in an embodiment of the present invention.

[0038] Figure 2 These are the test results of the TSP sub-model and the TSP-geological fusion prediction model in the embodiments of this invention. Detailed Implementation

[0039] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, the technical solution of the present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0040] Figure 1 This is an architecture diagram of the intelligent prediction method for surrounding rock levels based on TSP-geological information fusion provided by this invention. The intelligent prediction method for surrounding rock levels is divided into three parts. The first part performs operations such as feature extraction, outlier screening, and spatial registration on the acquired TSP detection data to form a dataset of TSP trend feature data and surrounding rock levels. The second part first stratifies the surrounding rock according to different levels of the working face, and establishes a model for predicting surrounding rock levels based on TSP detection data using a combination of expert experience and convolutional neural networks. The third part incorporates surrounding rock information from the design stage and the surrounding rock levels predicted by the TSP model for auxiliary decision-making, realizing the fusion prediction of surrounding rock levels. Finally, the TSP sub-model and the TSP-geological fusion prediction model are simulated and verified using test set data.

[0041] Example:

[0042] A smart prediction method for surrounding rock level based on TSP-geological information fusion includes:

[0043] S1: For TSP detection data, the interpretation of TSP detection data is to make predictions by the rising or falling trend of TSP data between the prediction point and the application surface. In this embodiment, the feature extraction method of the application surface difference is used to obtain the TSP trend features and then normalize them.

[0044] S1.1: For structured TSP detection data, in actual engineering sites, geophysical personnel use the degree of rise or fall of TSP detection data relative to the working face (the face where geophysical exploration is conducted) to make advanced geological predictions. This paper considers extracting the trend characteristics of TSP detection data, as shown below:

[0045]

[0046] in, For TSP data trend characteristics, TSP probes raw data for each point. This refers to the TSP detection data of the operation surface during this TSP detection.

[0047] S1.2: Then, the extracted trend features are standardized to obtain the final TSP trend feature library. The formula for standardization is shown in 1-2.

[0048]

[0049] Where Z represents the standardized result of the TSP trend feature. Trend features extracted from TSP detection data. The mean of the TSP trend characteristics. The standard deviation represents the trend characteristics of TSP.

[0050] S2: For different tunnel face mileage points obtained from geological information, spatial registration is performed with the TSP trend features obtained in S1. Since the surrounding rock grade information of each tunnel face mileage point is known, the registration of surrounding rock grade and TSP trend features can be completed, forming a dataset of TSP trend features and surrounding rock grade. Then, for the obtained dataset, in order to improve data quality, TSP trend feature abnormal data screening rules are formulated to remove abnormal data that do not meet the requirements.

[0051] S2.1: For the TSP detection data trend characteristics, the corresponding tunnel mileage is the detection mileage. Therefore, it needs to be spatially aligned with the tunnel face mileage. This embodiment uses the nearest neighbor interpolation method to achieve this. Based on actual tunnel engineering experience, this method is related to the tunnel excavation direction. For example, for a tunnel excavated from a small mileage to a large mileage, the tunnel face mileage should correspond to the mileage smaller than it and the nearest mileage. Based on this, the mileage alignment can be completed, that is, the registration of the TSP trend characteristics with the surrounding rock grade is completed, as shown below:

[0052]

[0053] in The tunnel mileage is aligned with the actual working face mileage. This refers to the actual mileage at the working face. Let be the tunnel mileage corresponding to the i-th TSP trend feature data. The direction of tunnel excavation (0 for short mileage, 1 for long mileage).

[0054] S2.2: When using TSP trend characteristic data for advanced geological prediction, if the surrounding rock grade of the working face is already determined, if the grade increases relative to the working face but the actual surrounding rock condition is worse, or if the grade decreases relative to the working face but the actual surrounding rock condition is better, this type of trend characteristic data does not conform to the underlying mechanism. Therefore, this embodiment considers adding a mechanism to remove this type of data to improve data quality. The specific mechanism satisfies the following conditions:

[0055]

[0056]

[0057] in, For the P-wave velocity at a certain mileage point, The longitudinal wave velocity of the application surface, This represents the change in P-wave velocity relative to the work surface at a given mileage point. Data representing the abnormal trend characteristics of TSPs that have been removed. The actual surrounding rock grade, The grade of the surrounding rock at the construction surface.

[0058] S3: Based on the feature extraction and abnormal data filtering methods, obtain the effective feature data corresponding to each working face; consider the TSP trend characteristics and the surrounding rock level information of the working face to establish a surrounding rock level prediction model, and finally incorporate the surrounding rock information in the design stage and make intelligent decisions with the prediction results of the TSP sub-model.

[0059] S3.1: Based on the feature matrix extraction and abnormal data filtering method, obtain the TSP trend features for each working face, form a dataset and divide it into training set, validation set and test set;

[0060] S3.2: The surrounding rock level of the construction face is divided into four levels: I, II, III, IV, V and VI. The TSP trend feature matrix and the surrounding rock level are used to build a model. For construction faces that can be directly divided, expert experience is used for judgment. For those that cannot be directly divided, a data-driven approach is used for modeling to obtain the final surrounding rock level prediction model.

[0061] S3.3: Based on the TSP surrounding rock level prediction model established for different construction faces, a surrounding rock result is obtained; then, the surrounding rock information in the design stage is incorporated into the TSP surrounding rock level model prediction result for intelligent decision-making, resulting in a surrounding rock level prediction model based on TSP-geological fusion. The specific decision-making mechanism is as follows: if the difference between the TSP prediction result and the surrounding rock level in the design stage is too large, the TSP prediction result shall prevail; if the difference between the TSP prediction result and the surrounding rock level in the design stage is not large, different weights shall be assigned for comprehensive decision-making. The weights of the surrounding rock level in the design stage and the TSP prediction model are 0.4 and 0.6, respectively.

[0062] S4: Use the trained intelligent prediction model for tunnel surrounding rock level to perform advanced geological prediction for actual tunnel projects, and validate the model.

[0063] S4.1: Use the TSP sub-model and the TSP-geological fusion model to forecast the test set data in S3 and verify the effectiveness of the model;

[0064] S4.2: Use the test set for validation, with the following validation metrics:

[0065]

[0066] in It is the accuracy on the test set. It is the number of correctly predicted samples. It represents the total number of samples in the test set.

[0067] The TSP surrounding rock grade sub-model and the TSP-geological fusion prediction model were validated using test set data. Figure 2 The results show the accuracy of the surrounding rock level prediction model. The results demonstrate that the intelligent prediction method for surrounding rock level based on TSP-geological information fusion proposed in this invention can improve the accuracy of individual TSP predictions and reduce the ambiguity of individual TSP predictions, which is beneficial for the application of this invention in actual production.

[0068] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent prediction of surrounding rock level based on TSP-geological information fusion, characterized in that, Includes the following steps: S1: The TSP trend features are obtained by using the feature extraction method of the application surface difference and then standardized. S2: Obtain different tunnel face mileage points from geological information, spatially register them with the TSP trend features obtained in S1, and form a dataset of TSP trend features and surrounding rock level. For the obtained dataset, formulate TSP trend feature abnormal data filtering rules and remove abnormal data that do not meet the requirements. S3: Based on the feature extraction method and abnormal data filtering rules, obtain the effective feature data corresponding to each working face; stratify the surrounding rock level of the working face, and adopt a combination of knowledge-driven and data-driven approach to establish a TSP surrounding rock level prediction sub-model according to the stratified level, based on the trend characteristics and surrounding rock level information of the working face; finally, incorporate the surrounding rock information from the design stage and merge it with the TSP surrounding rock level prediction sub-model to form a TSP-geological fusion model for intelligent decision-making; S4: Use the trained TSP surrounding rock level prediction sub-model and TSP-geological fusion model to perform advanced geological prediction for actual tunnel projects, and verify the model. Step S2 specifically includes the following processes: S2.1: Spatially align the trend characteristics of TSP probe data with the tunnel face mileage, as shown below: in The tunnel mileage is aligned with the actual working face mileage. This refers to the actual mileage at the working face. Let be the tunnel mileage corresponding to the i-th TSP trend feature data. The direction of tunnel excavation is indicated by 0 for a small mileage and 1 for a large mileage. S2.2: When using TSP trend characteristic data for advanced geological prediction, if the surrounding rock grade of the working face has been determined, and the grade increases relative to the working face but the actual surrounding rock condition is worse, or decreases relative to the working face but the actual surrounding rock condition is better, such trend characteristic data that do not conform to the mechanism should be removed. The specific mechanism should meet the following conditions: in, For the P-wave velocity at a certain mileage point, The longitudinal wave velocity of the application surface, This represents the change in P-wave velocity relative to the work surface at a given mileage point. Data representing the abnormal trend characteristics of TSPs that have been removed. The actual surrounding rock grade, The grade of the surrounding rock at the construction surface; Step S3 details The process includes the following: S3.1: Based on the feature extraction method and abnormal data filtering rules, obtain the TSP trend features for each working face, form a dataset and divide it into training set, validation set and test set; S3.2: The surrounding rock level of the construction face is divided into four levels: I, II, III, IV, V and VI. The TSP trend feature matrix and the surrounding rock level are used to build a model. For construction faces that can be directly divided, expert experience is used for judgment. For those that cannot be directly divided, a data-driven approach is used for modeling to obtain the TSP surrounding rock level prediction sub-model. S3.3: Based on the TSP surrounding rock level prediction sub-model established for different construction faces, a surrounding rock result is obtained. Then, the surrounding rock information in the design stage is incorporated into the TSP surrounding rock level model prediction result for intelligent decision-making, resulting in a surrounding rock level prediction model based on TSP-geological fusion. The specific decision-making mechanism is as follows: If the TSP forecast result differs too much from the surrounding rock level in the design stage, the TSP forecast result shall prevail; if the TSP forecast result is not much different from the surrounding rock level in the design stage, different weights shall be assigned for comprehensive decision-making, with the weights of the surrounding rock level in the design stage and the TSP forecast model being 0.4 and 0.6, respectively.

2. The intelligent prediction method for surrounding rock level based on TSP-geological information fusion as described in claim 1, characterized in that: Step S4 in detail The process includes the following: S4.1: Use the TSP surrounding rock level prediction sub-model and the TSP-geological fusion model to predict the test set data in S3 and verify the effectiveness of the model; S4.2: Use the test set for validation, with the following validation metrics: in It is the accuracy on the test set. It is the number of correctly predicted samples. It represents the total number of samples in the test set.

Images