A three-dimensional geological space modeling method based on random forest

Abstract

A three-dimensional geological space modeling method based on random forest, which combines random forest algorithm in machine learning with sparse drilling exploration data to establish a complete three-dimensional geological model, realizes accurate judgment of the geological conditions of the tunnel section, and the implementation steps include: first, obtaining the sparse drilling geological information of the target engineering tunnel along the line and the local encryption measuring point; second, identifying and extracting the target engineering geological feature matrix from the local encryption measuring point geological profile information or similar engineering geological data; third, model training and prediction according to the feature matrix to construct a local two-dimensional geological profile; fourth, considering the anisotropy of the stratum, constructing a two-dimensional geological profile array, and integrating the array to obtain a local three-dimensional geological model of the engineering; fifth, considering the tunnel excavation planning axis, integrating the three-dimensional geological models of multiple local ranges to obtain a complete three-dimensional geological model within the range along the tunnel.

Description

Technical Field

[0001] This invention relates to the fields of machine learning, advanced geological prediction, and 3D spatial modeling. Background Technology

[0002] With the increasing density of cities in my country, the demand for urban engineering systems and underground space is also growing. Constructing shield tunnels within cities and between cities is an effective way to improve travel efficiency and promote regional development. However, conducting advanced geological forecasting before tunnel construction is an essential step, crucial for ensuring tunnel construction safety and improving efficiency.

[0003] The main purpose of geological exploration and forecasting is to ascertain the engineering geological conditions, characteristics, distribution range, and development trend of adverse geological processes within the construction area; to provide geotechnical parameters within the exploration depth along the proposed project route; to analyze and evaluate the degree of hazard and its impact on the project; and to propose prevention and control suggestions. During shield tunneling, appropriate preparations and reinforcement measures should be taken based on the geological weaknesses observed in different sections according to the exploration feedback. The main exploration methods include on-site drilling, seismic reflection waves, and resistivity testing. However, due to the complexity and unique nature of the shield tunneling environment, it is difficult to directly apply geophysical and geochemical exploration methods such as reflection wave testing; and borehole exploration data is very sparse and cannot meet the needs of advanced geological forecasting for shield tunneling.

[0004] Therefore, in the field of tunnel boring machine (TBM) construction, finding a suitable method for advanced geological prediction is a highly valuable issue. Currently, commonly used methods include:

[0005] 1. Increasing the number and depth of borehole exploration points. This method can clarify the main engineering geological conditions within the construction area to a certain extent, but the final geological exploration effect depends on the actual number and depth of boreholes. Furthermore, this method consumes a lot of manpower and resources, and requires consideration of various complex surface conditions along the tunnel boring machine route, which greatly increases construction costs and construction period. In addition, although increasing the number of boreholes can obtain more detailed geological information of the area, an excessive number of boreholes may also pose certain hidden dangers to subsequent tunnel boring machine construction.

[0006] 2. Refer to the construction experience of projects with similar geological conditions, and simultaneously implement shield machine reinforcement and emergency response measures. Since there is currently no suitable method for advanced geological forecasting, in actual construction, the experience of similar projects is often referenced, and the tunneling speed is appropriately reduced during the initial excavation. Adjustments are made based on the stress and deformation feedback from the shield machine. The advantage of this approach is that it does not require very detailed advanced geological forecasting information, but the problem is that it heavily relies on past construction experience and the judgment of on-site personnel, and its safety and reliability are not high. Sudden geological changes at the tunnel face could potentially lead to safety issues.

[0007] The above-mentioned methods all have some shortcomings and cannot meet the requirements of shield tunnel construction for detailed geological conditions of the tunnel sections. With the increasing demand for urban underground space, shield tunnels are also developing towards increasingly longer tunneling routes and more complex geological conditions in the tunnel sections; therefore, how to achieve detailed and advanced geological forecasting for shield tunnels has become an urgent problem to be solved. Summary of the Invention

[0008] To overcome the need for detailed and advanced geological forecasting in tunnel boring machine (TBM) construction, this invention provides a method that combines the random forest algorithm in machine learning with sparse borehole exploration data to establish a complete three-dimensional geological model, thereby enabling accurate judgment of the geological conditions of the tunnel section.

[0009] This method consists of five steps: First, obtain sparse borehole geological information of the target tunnel along the entire route and in local soft areas; second, identify and extract the target engineering geological feature matrix from the geological profile information of the local densely packed measuring points or similar engineering geological data; third, perform model training and prediction based on the feature matrix to construct a local two-dimensional geological profile; fourth, considering the anisotropy of the strata, construct a two-dimensional geological profile array, and integrate the array to obtain a local three-dimensional geological model of the project; fifth, considering the tunnel excavation planning axis, integrate multiple local three-dimensional geological models to obtain a complete three-dimensional geological model along the tunnel route.

[0010] The specific implementation steps of this invention are as follows.

[0011] A. Obtain geological information from sparse boreholes:

[0012] A1. Select the target tunnel and determine parameters such as the tunnel's starting shaft, receiving shaft, tunnel depth, and tunneling axis.

[0013] A2. Based on the spatial location of the tunnel and the surface conditions, determine the location and depth of borehole exploration points along the tunnel route and collect geological information;

[0014] A3. Based on the geological conditions along the route collected from actual borehole exploration, conduct a preliminary analysis, select areas with weaker geology to increase the density of measuring points, and collect detailed geological information within a certain range. If there is existing geological information from similar projects, it should also be collected and organized as much as possible.

[0015] A4. Summarize and integrate all collected geological information, including geological types, soil and rock parameters, mechanical properties, and geological images, and save it to a spreadsheet.

[0016] B. Extract the geological feature matrix:

[0017] B1. First, all geological data along the tunnel are divided into several intervals according to their actual spatial location. The size of each interval is comparable to that of the area with denser monitoring points in soft soil.

[0018] B2. Select geological information from the densely packed measuring point area or existing geological information from similar projects. Based on the principle of actual spatial distance and similarity between sparse boreholes in other areas along the route, extract some geological data from the densely packed measuring point area as the feature matrix for the first round of prediction.

[0019] B3. After the first round of prediction, a portion of the prediction results is extracted and added to the original feature matrix as the feature matrix for the second round of prediction, and similar operations are repeated in subsequent operations.

[0020] C. Construct a local two-dimensional geological profile:

[0021] C1. Use the geological data of the densely packed measuring points as the training set, and the geological data of other sparsely packed measuring points along the route as the test set;

[0022] C2. Based on the random forest algorithm in the ensemble algorithm and the feature matrix extraction method in steps B2 and B3, the test set data is used for the first round of prediction to obtain partial prediction results between adjacent sparse boreholes.

[0023] C3. Based on the above prediction results, a second round of predictions will be conducted to obtain more regional geological data;

[0024] C4. Repeat the above prediction steps to obtain a complete two-dimensional geological profile image within a certain interval;

[0025] C5. Use appropriate iterative methods (such as genetic algorithms, particle swarm optimization, etc.) to optimize the prediction model parameters, improve the prediction accuracy of the algorithm, and obtain the optimized algorithm model and two-dimensional geological profile image.

[0026] D. Construct a local three-dimensional geological model:

[0027] D1. Select a test interval and set the two-dimensional geological profile obtained by the methods in steps C4 and C5 as the x-direction; then, along the y-direction, take similar steps to divide the dataset, train the data, and predict the two-dimensional geological profile in the y-direction.

[0028] D2. Considering anisotropy, different training models are used along the x and y directions to predict multiple two-dimensional geological profiles, resulting in a two-dimensional geological profile array within the interval.

[0029] D3. Based on the prediction of the two-dimensional geological profile array, continue to perform two-dimensional profile prediction to finally obtain the three-dimensional geological prediction model in the local area.

[0030] E-integrated complete 3D geological model:

[0031] E1. Based on the spatial location of the tunnel's overall excavation planning axis, multiple intervals are divided. In each interval, the local three-dimensional geological model is constructed and optimized in steps B, C, and D to obtain multiple interval three-dimensional geological prediction models.

[0032] E2. Considering the three-dimensional spatial relationship, multiple local three-dimensional geological models are integrated to finally obtain an overall three-dimensional geological model of the tunnel section that takes into account uncertainties.

[0033] The main advantages of this method are as follows.

[0034] 1. Very detailed geological data is not required; the original geological drilling data of the project can basically meet the prediction requirements. The data required for model training can be based on geological data from some densely packed measuring points, or existing geological data from similar projects can be used.

[0035] 2. The prediction method based on the random forest algorithm integrates the prediction results of multiple single models in machine learning, resulting in higher prediction accuracy;

[0036] 3. The algorithm prediction does not require much manual operation. After the model is built, only the initial geological data needs to be input to obtain the prediction results, saving labor costs.

[0037] 4. The modeling supports multi-source heterogeneous data input. Existing geological image data can be converted into spreadsheet data input using the model's image recognition module for training and prediction.

[0038] 5. The output prediction model data can be directly used to draw 3D model images, or imported into other modeling software such as finite element method for subsequent tunnel stress and settlement analysis;

[0039] 6. This local geological modeling method can directly predict and output three-dimensional spatial geological data after borehole exploration and before tunnel excavation, and reduce the capital cost and engineering construction cycle required for advanced geological exploration;

[0040] 7. Compared with advanced geological exploration methods such as reflected waves and resistivity testing, this method is not affected by electromagnetic or spatial interference from the tunnel boring machine during tunneling; Attached Figure Description

[0041] Figure 1 This is a schematic diagram illustrating the training principle of the algorithm model of the present invention.

[0042] Figure 2 This is a flowchart of the three-dimensional geological model prediction process of the present invention.

[0043] Figure 3 This is a flowchart illustrating the overall implementation of the present invention.

[0044] Figure 4 This is a comparison chart of the prediction results of the present invention.

[0045] Legend:

[0046] 1—Encrypted borehole data at measuring points;

[0047] 2—Sparse borehole data to be predicted;

[0048] 3—Principles of the Random Forest Algorithm;

[0049] 4—Two-dimensional geological prediction profile (xz plane, yz plane);

[0050] 5—Two-dimensional geological profile array;

[0051] 6—Three-dimensional geological prediction model. Detailed Implementation

[0052] The specific embodiments of the present invention are further described below with reference to the accompanying drawings.

[0053] The specific steps of the three-dimensional uncertain geological space modeling method based on the random forest algorithm are as follows:

[0054] A. Obtain geological information from sparse boreholes:

[0055] A1. Select the target tunnel and determine parameters such as the tunnel's starting shaft, receiving shaft, tunnel depth, and tunneling axis.

[0056] A2. Based on the spatial location of the tunnel and the surface conditions, determine the location and depth of borehole exploration points along the tunnel route and collect geological information;

[0057] A3. Based on the preliminary analysis of the geological conditions along the drilling route collected from actual borehole exploration, select areas with relatively weak geological conditions to establish additional monitoring points and collect detailed geological information within a certain range. If there is existing geological information from similar projects, it should also be collected and organized as much as possible. The data from these additional monitoring points or geological data from similar projects will serve as... Figure 1 The initial training data in the dataset is used to train the prediction model using the random forest algorithm.

[0058] A4. Summarize and integrate all collected geological information, including geological type, soil and rock parameters, mechanical properties and geological images, and save it to a spreadsheet; except for training data, the remaining geological borehole data are used as data to be predicted.

[0059] B. Extract the geological feature matrix:

[0060] B1. First, all geological data along the tunnel are divided into several intervals according to their actual spatial location. The size of each interval is comparable to that of the area with denser monitoring points in soft soil.

[0061] B2. Select geological information from the densely packed measuring point area or existing geological information from similar projects. Based on the principle of actual spatial distance and similarity between sparse boreholes in other areas along the route, extract some geological data from the densely packed measuring point area as the feature matrix for the first round of prediction.

[0062] B3. After the first round of prediction, a portion of the prediction results is extracted and added to the original feature matrix as the feature matrix for the second round of prediction. Similar operations are repeated in subsequent operations to build a pre-trained model by extracting the geological feature matrix in this way.

[0063] C. Construct a local two-dimensional geological profile:

[0064] C1. Use the geological data of the densely packed measuring points as the training set, and the geological data of other sparsely packed measuring points along the route as the test set;

[0065] C2. Based on the random forest algorithm in the ensemble algorithm and the feature matrix extraction method in steps B2 and B3, the first round of prediction is performed on the test set data, and the first round of prediction results are obtained from the borehole data to be predicted.

[0066] C3. Based on the above prediction results, a second round of prediction is conducted, and the second round of prediction results are obtained from the first round of prediction results, thereby obtaining more regional geological data;

[0067] C4. Repeat the above prediction steps to obtain a complete two-dimensional geological profile image within a certain interval;

[0068] C5. Appropriate iterative methods (such as genetic algorithms, particle swarm optimization, etc.) can be used to optimize the prediction model parameters, improve the prediction accuracy of the algorithm, and obtain the optimized algorithm model and two-dimensional geological profile image.

[0069] D. Construct a local three-dimensional geological model:

[0070] D1. As Figure 2 As shown, a test interval is selected, and the two-dimensional geological profile obtained by the method in steps C4 and C5 is set as the two-dimensional geological prediction profile along the x direction (i.e., the two-dimensional geological prediction profile in the xz plane); then, along the y direction, similar steps are taken to divide the dataset, train the data, and predict the two-dimensional geological profile in the y direction (the two-dimensional geological prediction profile in the yz plane).

[0071] D2. Considering anisotropy, different training models are used along the x and y directions to predict multiple two-dimensional geological profiles, resulting in the two-dimensional geological profile array in the interval shown in Figure 5.

[0072] D3. Based on the two-dimensional geological profile array prediction, continue to perform two-dimensional profile prediction, and finally obtain the three-dimensional geological prediction model in the local interval shown in Figure 6.

[0073] E-integrated complete 3D geological model:

[0074] E1. Based on the spatial location of the tunnel's overall excavation planning axis, multiple intervals are divided. In each interval, the local three-dimensional geological model is constructed and optimized in steps B, C, and D to obtain multiple interval three-dimensional geological prediction models.

[0075] E2. Considering the three-dimensional spatial relationship, multiple local three-dimensional geological models are integrated to finally obtain an overall three-dimensional geological model of the tunnel section that takes into account uncertainties.

[0076] A specific example is as follows:

[0077] (1) A dense survey point exploration was carried out on a rectangular area in the project. The geological data obtained was used as training data. In this example, it was stored in Excel in the form of a 50*100*100 data matrix. The data of each point represents the geological conditions of the corresponding spatial location.

[0078] (2) Build a random forest prediction model in Python, input training data, extract geological feature matrix by learning the distribution law inside the strata, and complete the training of the model;

[0079] (3) Select sparse borehole data of size 50*5 as test data, input it into the model for prediction, and obtain complete data of size 50*100 as the two-dimensional geological prediction result along the corresponding direction.

[0080] (4) Considering the anisotropy of the strata, multiple rounds of prediction are carried out along different directions. The results of multiple two-dimensional geological predictions are integrated into three-dimensional prediction results (50*100*100) to obtain the three-dimensional geological model of the corresponding section. By integrating multiple three-dimensional geological models, a complete three-dimensional geological model within multiple sections can be obtained.

[0081] (5) Figure 4 The comparison between the 3D geological model predicted by random forest and the actual results is shown. The prediction accuracy is calculated by the ratio of the number of predicted results that match the actual results for each data point in the 3D data matrix to the total amount of data. In this example, the prediction accuracy can reach 81%, which is a high level of accuracy for 3D geological conditions with scarce initial data.

[0082] In summary, the random forest prediction model established according to this embodiment can achieve relatively accurate predictions of three-dimensional geological models.

[0083] The contents described in this specification are merely an enumeration of the implementation forms of the inventive concept. The scope of protection of this invention is not limited to the specific forms stated in the implementation examples, but also includes equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A three-dimensional geological spatial modeling method based on random forest, comprising the following steps: A. Obtain geological information from sparse boreholes; A1. Select the target tunnel and determine the tunnel's starting shaft, receiving shaft, tunnel depth, and excavation axis parameters; A2. Based on the spatial location of the tunnel and the surface conditions, determine the location and depth of borehole exploration points along the tunnel route and collect geological information; A3. Based on the geological conditions along the route collected from actual borehole exploration, conduct a preliminary analysis, select geologically weak locations to increase the density of measuring points, collect detailed geological information within a certain range, and also collect and organize existing geological information from similar projects. A4. Summarize and integrate all collected geological information, including geological types, soil and rock parameters, mechanical properties, and geological image data, and save it to a spreadsheet; B. Extract the geological feature matrix; B1. First, all geological data along the tunnel are divided into several intervals according to their actual spatial location. The size of each interval is comparable to that of the area with denser monitoring points in soft soil. B2. Select geological information from the densely packed measuring point area or existing geological information from similar projects. Based on the principle of actual spatial distance and similarity between sparse boreholes in other areas along the route, extract some geological data from the densely packed measuring point area as the feature matrix for the first round of prediction. B3. After the first round of prediction, a portion of the prediction results is extracted and added to the original feature matrix as the feature matrix for the second round of prediction, and similar operations are repeated in subsequent operations. C. Construct a local two-dimensional geological profile; C1. Use the geological data of the densely packed measuring points as the training set, and the geological data of other sparsely packed measuring points along the route as the test set; C2. Based on the random forest algorithm in the ensemble algorithm and the feature matrix extraction method in steps B2 and B3, perform the first round of prediction on the test set data to obtain partial prediction results between adjacent sparse boreholes; C3. Based on the above prediction results, a second round of predictions will be conducted to obtain more regional geological data; C4. Repeat the above prediction steps to obtain a complete two-dimensional geological profile image within a certain interval; C5. Use appropriate iterative methods to optimize the prediction model parameters, improve the prediction accuracy of the algorithm, and obtain the optimized algorithm model and two-dimensional geological profile image; appropriate iterative methods include genetic algorithm and particle swarm optimization algorithm. D. Construct a local three-dimensional geological model; D1. Select a test interval and set the two-dimensional geological profile obtained by the methods in steps C4 and C5 as the x-direction; then, along the y-direction, take similar steps to divide the dataset, train the data, and predict the two-dimensional geological profile in the y-direction. D2. Considering anisotropy, different training models are used along the x and y directions to predict multiple two-dimensional geological profiles, resulting in a two-dimensional geological profile array within the interval. D3. Based on the prediction of the two-dimensional geological profile array, continue to perform two-dimensional profile prediction to finally obtain the three-dimensional geological prediction model in the local area; E-integrated complete 3D geological model; E1. Based on the spatial location of the tunnel's overall excavation planning axis, multiple intervals are divided. In each interval, the local three-dimensional geological model is constructed and optimized in steps B, C, and D to obtain multiple interval three-dimensional geological prediction models. E2. Considering the three-dimensional spatial relationship, multiple local three-dimensional geological models are integrated to finally obtain an overall three-dimensional geological model of the tunnel section that takes into account uncertainties.

Images