A two-dimensional geological profile prediction method based on a neural network

Abstract

A two-dimensional geological profile prediction method based on a neural network, which combines a neural network algorithm in machine learning with sparse drilling exploration data to predict a two-dimensional geological profile, realizes accurate judgment of the longitudinal profile geological conditions of a tunnel, and the implementation steps include the following: first, obtaining sparse drilling geological information of an encryption measurement point of a target engineering tunnel and a required prediction interval, and constructing a geological profile prediction database; second, preliminarily constructing and pre-training a neural network model; third, constructing a genetic algorithm basic framework and optimizing the neural network model; fourth, after training is completed, inputting test data into the optimized neural network model to predict a two-dimensional geological profile, and adopting accuracy and recall rate to evaluate the final result.

Description

Technical Field

[0001] This invention relates to the fields of machine learning, advanced geological prediction, and two-dimensional geological prediction, specifically to a two-dimensional geological profile prediction method based on neural networks. Background Technology

[0002] In recent years, with the rapid development of my country's national economy, urban population density has continued to increase, putting enormous pressure on urban surface transportation. Therefore, developing underground rail transit, represented by the shield tunneling method, has become an effective means to improve residents' travel efficiency and alleviate surface traffic pressure. Before constructing a tunnel, conducting advanced geological forecasting is an essential step. Its main purpose is to provide geological basis for tunnel excavation, and the quality of the geological forecast has a significant impact on the subsequent construction quality.

[0003] Currently, the main method for advanced geological forecasting is based on geological surveys of the surrounding area conducted before tunnel excavation, providing a rough prediction of the geological conditions along the tunnel route. This geological survey primarily involves borehole exploration along the tunnel route. This involves first arranging boreholes according to the overall route design, then using mud-wall drilling and rotary drilling techniques to obtain core samples for analysis. Other methods include in-situ tests such as longitudinal wave testing, resistivity testing, and standard penetration testing, as well as laboratory tests such as consolidation, direct shear, and particle size analysis. These geological surveys can preliminarily identify the topography, lithology, geological structure, and adverse geological phenomena of the tunnel site area; evaluate the stability and suitability of the engineering site; determine the types, distribution, and engineering characteristics of adverse geological and special soil and rock formations in the tunnel site area; analyze and evaluate their adverse impacts on the tunnel project; and propose engineering countermeasures and recommendations.

[0004] Geological predictions based on geological exploration generally employ traditional interpolation methods or combine them with the drafter's personal experience. Since the longitudinal length of a tunnel is much greater than its transverse length, a two-dimensional geological image along the tunnel's longitudinal profile can effectively reflect its geological conditions. When drawing two-dimensional geological profiles, borehole geological exploration data is often used, combined with methods such as linear interpolation, inverse distance weighted interpolation, and Kriging interpolation to fully reconstruct the geological conditions of the tunnel's location. Experienced professionals may also draw the profiles based on their personal experience. Specific considerations also depend on factors such as the importance of the project and the complexity of the geology.

[0005] While the above methods are relatively simple, they have the following problems: The most significant problem is the low accuracy of geological predictions. This is because the available borehole data is very sparse when drawing the profiles, making it difficult to reconstruct the true geological distribution patterns using ordinary interpolation methods, and it is also impossible to analyze the internal relationships from known geological data. Secondly, the methods lack versatility. Traditional prediction methods may yield ideal results or significant deviations under different geological conditions, which is largely related to the specific method chosen. Finally, the method of drawing profiles based on personal experience is highly subjective and heavily relies on the individual skill level of the practitioners.

[0006] Machine learning methods, represented by neural networks, are prediction methods that have emerged in recent years. They have a very wide range of applications and high accuracy, and can provide more accurate prediction results based on existing sparse geological data. Summary of the Invention

[0007] This invention aims to overcome the aforementioned shortcomings of existing technologies and provide a method to improve the accuracy of prediction results in engineering geological forecasting. This method optimizes neural network algorithms using GA (genetic algorithm) and combines one-dimensional sparse borehole data to achieve high-accuracy prediction of two-dimensional geological profiles.

[0008] The two-dimensional geological profile prediction method based on GA optimized neural network proposed in this invention uses borehole information obtained from exploration as input variables and unknown geological information between adjacent boreholes as output variables, thereby obtaining complete and highly accurate two-dimensional geological profile information. During model training, geological data from densely packed measurement points are used as the training set, and borehole information obtained from normal geological exploration is used as the test set. The hyperparameters of the neural network algorithm are optimized using the GA optimization algorithm to train the model and obtain accurate prediction results.

[0009] This method consists of four steps: First, borehole information of sparse exploration points and dense measuring points within the target area is obtained to construct a geological profile prediction database; second, a prediction model based on a neural network algorithm is constructed and preliminarily trained using geological information from the dense measuring points; third, the hyperparameters of the neural network algorithm are optimized and the model is trained using a genetic algorithm to obtain an optimized prediction model; finally, based on the above-trained prediction model, geological parameters are input to predict a complete and accurate two-dimensional geological profile.

[0010] The present invention provides a two-dimensional geological profile prediction method based on neural networks, the specific implementation steps of which are as follows.

[0011] A. Construct a geological profile prediction database;

[0012] A1. Select the target tunnel and determine the location of the required two-dimensional geological profile and the relative positions of the borehole exploration points;

[0013] A2. Based on the determined spatial relationship of the boreholes and the borehole exploration depth, conduct initial geological exploration to collect one-dimensional sparse borehole information;

[0014] A3. Conduct preliminary analysis of the collected borehole exploration data, select representative locations for intensified measurement point exploration, collect detailed geological information within a certain range, and combine existing engineering case experience and theoretical methods to reconstruct the two-dimensional geological profile information of this part as a data training set.

[0015] A4. Summarize and integrate all collected geological data, including geological types and various mechanical parameters, to establish a geological profile prediction database;

[0016] B. Construct a neural network prediction model;

[0017] B1. Construct a neural network algorithm, select the two-dimensional geological profile information restored by the encrypted measuring point exploration in step A3 as training data, take the sparse borehole data corresponding to the ordinary geological exploration points as input parameters, and take the remaining geological data in the training data as output parameters.

[0018] B2. Perform initial algorithm training and adopt a multi-round iterative prediction method. In the first round of prediction, select the input parameters in step B1 as the input parameters for the first round of prediction, and take the one-dimensional unknown geological data at the location of each two columns of one-dimensional borehole data center as the output parameters. The prediction result is regarded as the completion of the first round of training prediction.

[0019] B3. After the first round of prediction is completed, the original input and output parameters are used as the input parameters for the second round of prediction. The one-dimensional unknown geological data at the center of the new input parameters is then used as the output parameter, thus completing the second round of prediction. This process is repeated until the final round of training and prediction is completed. Preliminary evaluation of the model is then performed using accuracy and recall.

[0020] C.GA optimization algorithm optimizes neural network models;

[0021] C1. Based on the neural network prediction model, construct the basic framework of the genetic algorithm and complete the steps of initial population generation and fitness evaluation and detection;

[0022] C2. Use the above optimization algorithm to optimize parameters such as hidden layer size, maximum number of iterations, activation function, and solver in the neural network to find the optimal hyperparameters;

[0023] C3. Based on existing multi-round iterative prediction methods, the optimal hyperparameters are assigned to the neural network algorithm model to obtain the optimized model, and the model is trained a second time.

[0024] D. Predicting two-dimensional geological profiles;

[0025] D1. Select the two-dimensional geological profile information reconstructed from the encrypted survey points as training data, and the remaining sparse geological survey data as test data. Input the sparse borehole data at the corresponding locations as input parameters to the neural network model optimized by the genetic algorithm obtained in step C3.

[0026] D2. Use the remaining unknown geological data in the training set as output parameters, and train and predict according to the prediction model established above.

[0027] D3. After training is complete, the test data is input into the optimized neural network model to predict two-dimensional geological profiles, and the final results are evaluated using accuracy and recall.

[0028] The advantages of this invention are:

[0029] 1. The requirements for geological exploration data are not high; predictions can be made based on the borehole exploration data of the project itself, and the accuracy of the predictions can be evaluated.

[0030] 2. The neural network algorithm used has high fitness and accuracy, and can obtain relatively stable and accurate prediction results for different geological conditions. After optimization and training by combining it with the genetic algorithm, even more accurate results can be obtained.

[0031] 3. The algorithm prediction requires only a portion of manual operation. After the model is established, only the initial geological data needs to be input to obtain complete prediction results and result analysis.

[0032] 4. The output prediction results are highly versatile and can be directly imported into software such as Excel and ABAQUS for subsequent data processing, finite element analysis, and other operations;

[0033] 5. This method can make predictions directly after obtaining geological exploration data, and can quickly obtain prediction results with low time cost; Attached Figure Description

[0034] Figure 1 This is a schematic diagram illustrating the prediction principle of the algorithm model of the present invention.

[0035] Figure 2 This is a flowchart of the two-dimensional geological profile prediction process of the present invention.

[0036] Figure 3 This is a flowchart illustrating the overall implementation of the present invention.

[0037] Figure 4 This is the training data matrix of this invention.

[0038] Figure 5 This is a comparison chart of the prediction results of the present invention.

[0039] Legend:

[0040] 1—Sparse borehole data to be predicted;

[0041] 2 – Results of the first round of predictions;

[0042] 3 — Prediction results for the (n-1)th round;

[0043] 4—Two-dimensional geological prediction profile;

[0044] 5 – Prediction in rounds 1, 2, and n; Detailed Implementation

[0045] The specific embodiments of the present invention are further described below with reference to the accompanying drawings.

[0046] A method for predicting two-dimensional geological profiles based on neural networks, the specific steps of which are as follows:

[0047] A. Constructing a geological profile prediction database:

[0048] A1. Select the target tunnel and determine the location of the required two-dimensional geological profile and the relative positions of the borehole exploration points;

[0049] A2. Based on the determined spatial relationship of the boreholes and the borehole exploration depth, conduct initial geological exploration to collect one-dimensional sparse borehole information;

[0050] A3. Conduct preliminary analysis of the collected borehole exploration data, select representative locations for intensified measurement point exploration, collect detailed geological information within a certain range, and combine existing engineering case experience and theoretical methods to reconstruct the two-dimensional geological profile information of this part as a data training set.

[0051] A4. Summarize and integrate all collected geological data, including geological types and various mechanical parameters, to establish a geological profile prediction database, as... Figure 1 Input layer data;

[0052] B. Constructing a neural network prediction model

[0053] B1. Construction Figure 1 The neural network algorithm shown selects the two-dimensional geological profile information restored by the encrypted measuring point exploration in step A3 as training data, takes the sparse borehole data corresponding to the ordinary geological exploration points as input parameters, and takes the remaining geological data in the training data as output parameters.

[0054] B2. Perform initial algorithm training, using a multi-round iterative prediction method. In the first round of prediction, select the input parameters from step B1 as the input parameters for the first round of prediction. Figure 1 Input layer data x iThe one-dimensional unknown geological data at the location of each pair of one-dimensional borehole data centers is used as the output parameters. Figure 1 Middle output layer data y i The predicted result is considered as completing the first round of training prediction, i.e., completing the prediction process. Figure 2 Steps shown in section 5A;

[0055] B3. After the first round of prediction is completed, the original input and output parameters are used as the input parameters for the second round of prediction. Figure 1 The neural network input layer is used, and the one-dimensional unknown geological data at the center of the new input parameters is used as the output parameters to complete the second round of prediction, i.e., complete the prediction. Figure 2 Follow the steps shown in section 5B. Repeat this process until the final round of training and prediction is complete, thus finishing the process. Figure 2 Following step 5C, a preliminary two-dimensional geological prediction profile is obtained at this stage. Then, accuracy and recall are used for preliminary model evaluation.

[0056] C.GA optimization algorithm optimizes neural network models.

[0057] C1. Based on the neural network prediction model, construct the basic framework of the genetic algorithm and complete the steps of initial population generation and fitness evaluation and detection;

[0058] C2. Using the above optimization algorithm to... Figure 1 In the neural network, parameters such as hidden layer size, maximum number of iterations, activation function, and solver are optimized to find the optimal hyperparameters;

[0059] C3. Based on existing multi-round iterative prediction methods, the optimal hyperparameters are assigned to the neural network algorithm model, the hidden layer is optimized, the optimized neural network model is obtained, and the model is trained a second time.

[0060] D. Predicting two-dimensional geological profiles

[0061] D1. Select the two-dimensional geological profile information reconstructed from the encrypted survey points as training data, and the remaining sparse geological survey data as test data. Input the sparse borehole data at the corresponding locations as input parameters to the neural network model optimized by the genetic algorithm obtained in step C3.

[0062] D2. Use the remaining unknown geological data in the training set as output parameters, and train and predict according to the prediction model established above.

[0063] D3. After training, input the test data into the optimized neural network model to predict two-dimensional geological profiles. The prediction steps are similar to steps B2 and B3. Precision and recall are used to evaluate the final results.

[0064] A specific example is as follows:

[0065] (1) The geological data obtained from the encrypted measurement points is used as training data. In this example, it is stored in an Excel spreadsheet as a 50*100 data matrix. The data for each point represents the geological conditions of the corresponding spatial location, such as... Figure 4 A portion of sparse borehole data (50*5) was selected for testing as initial test data, and then subjected to intensive exploration to obtain complete test data. The test data was stored in the same way as the training data.

[0066] (2) Build a neural network model in Python, import the training data and initial test data, normalize them, and perform preliminary training;

[0067] (3) Establish a GA optimization algorithm to optimize the hyperparameters of the neural network model and improve prediction accuracy;

[0068] (4) The trained neural network model is obtained by importing the training data matrix and optimization algorithm. The trained model is used to import the initial test data into the model for prediction. The prediction results are obtained. The results of all data points are compared with the test data one by one to verify the accuracy.

[0069] (5) Figure 5 The comparison between the prediction results of the trained neural network and the actual results is shown. The accuracy of the prediction results can reach about 85%. It can be seen that high prediction accuracy can be obtained by using neural networks.

[0070] In summary, the neural network prediction model established according to this embodiment is relatively complete and can achieve accurate prediction of two-dimensional geological profiles.

[0071] 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 two-dimensional geological profile prediction method based on neural networks, comprising the following steps: A. Construct a geological profile prediction database; A1. Select the target tunnel and determine the location of the required two-dimensional geological profile and the relative positions of the borehole exploration points; A2. Based on the determined spatial relationship of the boreholes and the borehole exploration depth, conduct initial geological exploration to collect one-dimensional sparse borehole information; A3. Conduct preliminary analysis of the collected borehole exploration data, select representative locations for intensified measurement point exploration, collect detailed geological information within the project area, and combine existing engineering case experience and theoretical methods to reconstruct two-dimensional geological profile information as a data training set; A4. Summarize and integrate all collected geological data, including geological types and various mechanical parameters, and establish a geological profile prediction database; B. Construct a neural network prediction model; B1. Construct a neural network algorithm, select the two-dimensional geological profile information restored by the encrypted measuring point exploration in step A3 as training data, take the sparse borehole data corresponding to the ordinary geological exploration points as input parameters, and take the remaining geological data in the training data as output parameters. B2. Perform initial algorithm training and adopt a multi-round iterative prediction method. In the first round of prediction, select the input parameters in step B1 as the input parameters for the first round of prediction, and take the one-dimensional unknown geological data at the location of each two columns of one-dimensional borehole data center as the output parameters. The prediction result is regarded as the completion of the first round of training prediction. B3. After the first round of prediction is completed, the original input parameters and output parameters are used as the input parameters for the second round of prediction, and the one-dimensional unknown geological data at the center of the new input parameters is used as the output parameters to complete the second round of prediction. Repeat this step until the final round of training and prediction is completed, and use accuracy and recall to conduct an initial evaluation of the model. C.GA optimization algorithm optimizes neural network models; C1. Based on the neural network prediction model, construct the basic framework of the genetic algorithm and complete the initial population generation and fitness evaluation and detection; C2. Use the above optimization algorithm to optimize the hidden layer size, maximum number of iterations, activation function, and solver parameters in the neural network to find the optimal hyperparameters; C3. Based on existing multi-round iterative prediction methods, the optimal hyperparameters are assigned to the neural network algorithm model to obtain the optimized model, and the model is trained a second time. D. Predicting two-dimensional geological profiles; D1. Select the two-dimensional geological profile information restored by the exploration of the encrypted measuring points as training data, and the remaining sparse geological exploration data as test data; and input the sparse borehole data at the corresponding location as input parameters, and input the neural network model optimized by the genetic algorithm obtained in step C3. D2. Use the remaining unknown geological data in the training set as output parameters, and train and predict according to the prediction model established above. D3. After training is complete, the test data is input into the optimized neural network model to predict two-dimensional geological profiles, and the final results are evaluated using accuracy and recall.

Images