CN-121810974-B - Three-dimensional reconstruction method for layered geological structure

CN121810974BCN 121810974 BCN121810974 BCN 121810974BCN-121810974-B

Abstract

The invention discloses a three-dimensional reconstruction method of a layered geological structure, which comprises the following steps of 1, constructing a three-dimensional voxel model of a research site, collecting drilling data, constructing a first-stage parameter sample set, wherein the input of samples in the first-stage parameter sample set is voxel space position information, the output is stratum type, 2, constructing a first-stage structure identification model, training and optimizing by using the first-stage parameter sample set, 3, using the first-stage structure identification model which is completed by training and optimizing, outputting a prediction result of the stratum type of the whole domain, and outputting a final prediction result through layering sequence constraint, 4, constructing a second-stage parameter sample set, wherein the input of samples in the second-stage parameter sample set is the prediction result and while-drilling monitoring characteristics of the stratum type of the whole domain of the research site, and the output is soil parameters, 5, constructing a second-stage parameter prediction model, and training and optimizing by using the second-stage parameter sample set.

Inventors

CHEN XIN
PAN TONG
Cheng Tianshuo
FU ZHENHAO
DENG JIA
ZHANG QINGSONG
Chang Luyi
WU MEITING
LIU ZHENDONG
HU YUEPENG
SHEN JIAHAO
MA HONGYI

Assignees

山东大学

Dates

Publication Date: 20260508
Application Date: 20260311

Claims (9)

1. A method for three-dimensional reconstruction of a layered geological structure, comprising the steps of: Step 1, constructing a three-dimensional voxel model of a research site, collecting drilling data, and constructing a first-stage parameter sample set, wherein the input of samples in the first-stage parameter sample set is voxel space position information, and the input is stratum category; Step 2, constructing a first-stage structure identification model, and training and optimizing by using a first-stage parameter sample set; step 3, using a first stage structure recognition model with training and optimizing completed, outputting a prediction result of the global stratum category, and outputting a final prediction result through layering sequence constraint; step 4, constructing a second-stage parameter sample set, wherein the input of samples in the second-stage parameter sample set is the prediction result and the while-drilling monitoring characteristic of the stratum category of the research site universe, and the output is the soil property parameter; step 5, constructing a second-stage parameter prediction model, and training and optimizing by using a second-stage parameter sample set; Step 6, predicting stratum types of the whole domain of the target site by using the first-stage structure recognition model with the training optimization completed, and outputting a final prediction result by layering sequence constraint; The step3 comprises the following steps: 3.1, predicting and outputting probabilities of stratum categories corresponding to all voxel space positions except the voxel space positions of the drilling data in a research site by using a first stage structure recognition model with training and optimizing completed; step 3.2, arranging and screening the combination of stratum categories of which all positions on each vertical voxel column meet layering sequence constraint, wherein the layering sequence constraint is that the stratum categories of all positions of the vertical voxel column from top to bottom meet the sequence change of 0-1-2; and 3.3, calculating the overall cost of each combination of each vertical voxel column according to the arrangement screening result in the step 3.2, and selecting the combination with the minimum overall cost as a final result, wherein the overall cost calculation formula is as follows: ; In the formula (i), For voxel space position The probability of the formation class c.
2. The method of three-dimensional reconstruction of a layered geological structure according to claim 1, wherein step 1 comprises the steps of: Step 1.1, acquiring the space range and boundary information of a research field, dividing the research field into regular voxel units, constructing a three-dimensional voxel model, and enabling each voxel unit to correspond to a unique space position; step 1.2, obtaining drilling data, including drilling plane position, drilling depth, drilling inclination angle and stratum type, and carrying out format unification, abnormal rejection and integrity check on the drilling data; step 1.3, discretizing the drilling data along the drilling depth direction according to preset intervals, converting the drilling data into depth-by-depth samples, wherein each sample comprises voxel space position information And the formation category corresponding to z, and for vertical holes, And, for the inclined hole, 、、 Wherein x 0 、y 0 , In order to drill the coordinates of the hole at the surface, For the depth of the borehole to be drilled, For the inclination of the borehole, Is azimuth; and 1.4, constructing a first-stage parameter sample set with voxel space position information as input and stratum category as output based on the drilling depth-by-depth record.
3. The method of claim 2, further comprising the step of constructing a true value formation having upper and lower layers and an interface relief in step 1.1: step 1.1.1 constructing two interface Functions by random Functions 、 Respectively representing the fluctuation of the first interface and the second interface at different spatial positions, wherein ; Step 1.1.2 generating voxel class according to three layers of rules from top to bottom when The formation type is 0 when The formation type is 1 when In this case, the formation type is 2.
4. A method of three-dimensional reconstruction of a layered geological structure according to claim 1, wherein step 2 comprises the steps of: step 2.1 for voxel space position information And (3) performing scale unification and normalization processing to eliminate the influence of different scales on the model training process, wherein the formula is as follows: ; In the formula, x, y and z are original space coordinates of the voxel unit, The minimum and maximum values of the spatial extent in the x-direction respectively, The minimum and maximum values of the spatial extent in the y-direction respectively, The minimum and maximum values of the spatial extent in the z-direction respectively, Respectively normalized dimensionless characteristic values with a value range of ; Step 2.2, setting the type of the first-stage structure identification model and setting initial parameters of the first-stage structure identification model; training the first-stage structure identification model by using the first-stage parameter sample set to enable the first-stage structure identification model to learn the corresponding relation between voxel space position information and stratum category; in the training process of the first-stage structure identification model, a grid search parameter optimization method is introduced, traversal test is carried out on key parameter combinations of the first-stage structure identification model, and model performances under different parameter combinations are recorded; and 2.4, selecting and adjusting the model according to the evaluation result of the model performance, and completing training when the classification accuracy of the model reaches the preset requirement to form a final first-stage structure identification model.
5. The three-dimensional reconstruction method of a stratified geological structure as claimed in claim 4, wherein in step 2.2, the first stage structure recognition model is set as a random forest classification model, and initial parameters of the random forest model are set as n_ estimators =200, max_depth=15, min_samples_leaf=3, class_weight=bandwidth, range_state=0, and the model is output as probability vectors of corresponding voxels belonging to each stratum class, p= (P0, P1, P2); In step 2.4, the parameter search range is n_ estimators epsilon {150, 300}, max_depth epsilon {12, none }, min_samples_leaf epsilon {1, 3}, the cross-validation is KFold, and the fold number is 3.
6. A method of three-dimensional reconstruction of a layered geological structure according to claim 1, wherein step 4 comprises the steps of: Step 4.1, constructing a second-stage parameter sample set, wherein each sample in the second-stage parameter sample set corresponds to a voxel space position in the form as follows: Sample of ; In the formula (i), For the spatial location of the voxel, For the class of formation to which the location corresponds, In order to monitor the characteristics while drilling, Is an earthen parameter; Step 4.2, carrying out unit unification, scale normalization processing and missing value processing on the data in the constructed second-stage parameter sample set; and 4.3, checking the space consistency, the class consistency and the parameter rationality of the samples, and eliminating obvious abnormal or conflict samples.
7. A method of three-dimensional reconstruction of a layered geological structure according to claim 1, wherein step 5 comprises the steps of: step 5.1, setting the type of the second-stage parameter prediction model and setting initial parameters of the second-stage parameter prediction model; Setting a plurality of candidate model parameter combinations in the training process, evaluating the prediction performance of the model under different parameter combinations through parameter traversal and cross verification, selecting the parameter combination with the minimum prediction error or the optimal stability as the final model configuration, and completing the training of the second-stage parameter prediction model under the parameter configuration; And 5.3, carrying out prediction performance evaluation on the trained second-stage parameter prediction model, and completing training when the model prediction error meets the preset precision requirement to form a final second-stage parameter prediction model.
8. The method of claim 7, wherein in step 5.1, the second stage structure identification model is set to XGBoost, the initial parameters are default parameters, and in step 5.2, the XGBoost regression model sets candidate parameter combinations ：n_estimators：{300,500,800},learning_rate：{0.03,0.05,0.1},max_depth：{4,6,8},min_child_weight：{1,3,5},subsample：{0.7,0.8,1.0},colsample_bytree：{0.7,0.8,1.0},reg_lambda：{0.5,1.0,2.0},reg_alpha：{0.0,0.1,0.5},gamma：{0.0,0.1,0.3}, to perform cross-validation on each set of parameter combinations.
9. The three-dimensional reconstruction method of a layered geological structure according to claim 7, wherein in step 5.2, the second-stage parameter sample set is randomly divided into n subsets which are not overlapped with each other, k=1 to n for the kth training, the kth subset is selected as a verification set, the remaining n-1 subsets are used as training sets, one model training and evaluation are completed, and finally, the 5 evaluation results are averaged to be used as the overall performance index of the parameter combination.

Description

Three-dimensional reconstruction method for layered geological structure Technical Field The invention relates to the technical field of geotechnical engineering, in particular to a three-dimensional reconstruction method of a layered geological structure. Background In the geotechnical engineering fields such as tunnel engineering, mine engineering, underground space development and the like, accurately acquiring the underground layered geological structure and the soil parameters thereof is a basic work for carrying out engineering design, construction and numerical simulation analysis. At present, the modeling of the underground geological structure mainly depends on drilling data, field disclosure information and manual interpretation of engineering personnel, and is completed through two-dimensional section extrapolation or three-dimensional modeling software construction. The process is highly dependent on manual experience, so that the workload is large, the time consumption is long, modeling results of different persons have large differences, and the requirements of rapid modeling and objective modeling in engineering practice are difficult to meet. In addition, in the engineering tunneling and construction process, numerical simulation calculation has become an important means for analyzing stability of surrounding rock, designing support and evaluating risk, and soil property parameters are indispensable key inputs in numerical simulation. At present, soil parameters are usually obtained by on-site sampling and indoor test or in-situ test, if sampling and test are carried out again in each area needing to carry out numerical analysis, the period is long, the cost is high, the soil parameters are difficult to obtain in time under the condition of rapid construction promotion, and the application efficiency of numerical simulation in engineering decision is restricted. In summary, in the prior art, on one hand, quick and automatic modeling of a layered geological structure is difficult to realize, on the other hand, on the premise of ensuring rationality, soil property parameter distribution covering a whole area is difficult to quickly obtain, and the dual requirements of quick modeling and quick parameter determination in engineering practice cannot be met. The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present invention and is not intended to represent an admission that the foregoing is prior art. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides a three-dimensional reconstruction method of a layered geological structure, which can quickly construct a three-dimensional layered geological structure model based on the existing drilling data and forecast and obtain the earth parameters of a coverage universe on the basis, so as to provide an intelligent method for directly inputting available parameters for numerical simulation, reduce the workload of manual modeling, reduce the cost of repeated experiments and improve the engineering analysis and decision-making efficiency. In order to achieve the above purpose, the present invention further adopts the following technical scheme: a method for three-dimensional reconstruction of a layered geological structure, comprising the steps of: Step 1, constructing a three-dimensional voxel model of a research site, collecting drilling data, and constructing a first-stage parameter sample set, wherein the input of samples in the first-stage parameter sample set is voxel space position information, and the input is stratum category; Step 2, constructing a first-stage structure identification model, and training and optimizing by using a first-stage parameter sample set; step 3, using a first stage structure recognition model with training and optimizing completed, outputting a prediction result of the global stratum category, and outputting a final prediction result through layering sequence constraint; step 4, constructing a second-stage parameter sample set, wherein the input of samples in the second-stage parameter sample set is the prediction result and the while-drilling monitoring characteristic of the stratum category of the research site universe, and the output is the soil property parameter; step 5, constructing a second-stage parameter prediction model, and training and optimizing by using a second-stage parameter sample set; and 6, predicting the stratum category of the whole domain of the target site by using the first-stage structure identification model with the training and optimizing, outputting a final prediction result by layering sequence constraint, and predicting the soil property parameter of the target site by using the second-stage parameter prediction model with the training. Further, step 1 includes the steps of: Step 1.1, acquiring the space range and boundary information of a research field, dividing the research field in