Soft rock vertical shaft deformation value prediction method based on machine learning

Abstract

The invention provides a soft rock vertical shaft deformation value prediction method based on machine learning, and relates to the technical field of soft rock vertical shaft deformation value prediction. Comprising the steps of obtaining to-be-measured soft rock shaft data; inputting the shaft data of the soft rock shaft to be measured into the constructed soft rock deep well deformation prediction model to obtain a prediction result; the construction method of the soft rock deep well deformation prediction model comprises the following steps: determining an initial data set according to engineering geological data; introducing noise into the initial data set to obtain a new noise data set; determining construction conditions of a soft rock deep well deformation prediction model, and constructing an initial soft rock deep well deformation prediction model set based on a random forest, XGBoost, LightGBM and k neighbor regression according to the new noise data set; and performing precision evaluation on the models in the initial soft rock deep well deformation prediction model set, and determining a final soft rock deep well deformation prediction model. The problem of low prediction accuracy of the shaft deformation value of the soft rock shaft is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of soft rock shaft deformation value prediction, and in particular to a soft rock shaft deformation value prediction method based on machine learning. Background Art

[0002] In underground engineering construction, the deformation prediction of soft rock shaft is a key issue to ensure the safety and smooth progress of the project. Traditional analysis methods have many limitations when dealing with the deformation problem of soft rock shaft under complex geological conditions, and cannot effectively update the model in time to ensure the accuracy and reliability of the prediction. Summary of the invention

[0003] In order to overcome the deficiencies of the prior art, the object of the present invention is to provide a method for predicting the deformation value of a soft rock vertical shaft based on machine learning.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for predicting deformation value of a soft rock vertical shaft based on machine learning, comprising:

[0006] Obtain the shaft data of the soft rock vertical shaft to be tested;

[0007] Inputting the soft rock vertical wellbore data to be tested into the constructed soft rock deep well deformation prediction model to obtain the prediction result;

[0008] The method for constructing the soft rock deep well deformation prediction model is as follows:

[0009] Obtain engineering geological data from on-site investigation;

[0010] Extracting intrinsic characteristic parameters and extrinsic characteristic parameters of the soft rock vertical shaft to be tested according to the engineering geological data to determine an initial data set;

[0011] Introducing noise into the initial data set to obtain a new noise data set;

[0012] Determine the construction conditions of the soft rock deep well deformation prediction model, and construct an initial soft rock deep well deformation prediction model set based on random forest, XGBoost, LightGBM and k-nearest neighbor regression according to the new noise data set;

[0013] The models in the initial soft rock deep well deformation prediction model set are evaluated for accuracy to determine a final soft rock deep well deformation prediction model.

[0014] Preferably, it also includes:

[0015] Calculating the spatial autocorrelation of the prediction results to obtain a correlation result;

[0016] The correlation result is judged to be continuous and sudden, and a judgment result is obtained. If the judgment result is yes, the measured deformation value and the depth position information of the actual engineering data are selected for comparison with the prediction result, and the measured deformation value and the depth position information of the actual engineering data are used to optimize the soft rock deep well deformation prediction model.

[0017] Preferably, the step of performing accuracy evaluation on the models in the initial soft rock deep well deformation prediction model set to determine the final soft rock deep well deformation prediction model comprises:

[0018] Calculate the RMSE, MAE and R of each model in the initial soft rock deep well deformation prediction model set 2 ;

[0019] According to the RMSE, MAE and R 2 Determine the final soft rock deep well deformation prediction model.

[0020] Preferably, RMSE, MAE and R 2 The calculation expressions are:

[0021]

[0022] Among them, RMSE is the root mean square error, MAE is the mean absolute error, y i is the observed value, is the predicted value, y is the average of the predicted values, and n is the number of sample sets.

[0023] Preferably, the intrinsic characteristic parameters include:

[0024] Elastic modulus, Poisson's ratio, cohesion, internal friction angle, and rock compressive strength.

[0025] Preferably, the external characteristic parameters include:

[0026] Maximum tangential stress of surrounding rock and compressive strength of rock mass.

[0027] The present invention discloses the following technical effects:

[0028] The present invention provides a method for predicting deformation value of a soft rock vertical shaft based on machine learning, comprising:

[0029] Obtain the borehole data of the soft rock vertical shaft to be tested; input the borehole data of the soft rock vertical shaft to be tested into the constructed soft rock deep well deformation prediction model to obtain the prediction result; the construction method of the soft rock deep well deformation prediction model is: obtain the engineering geological data of the field survey; according to the engineering geological data, extract the intrinsic characteristic parameters and extrinsic characteristic parameters of the soft rock vertical shaft to be tested to determine the initial data set; introduce noise to the initial data set to obtain a new noise data set; determine the construction conditions of the soft rock deep well deformation prediction model, and according to the new noise data set, construct an initial soft rock deep well deformation prediction model set based on random forest, XGBoost, LightGBM and k nearest neighbor regression; perform accuracy evaluation on the models in the initial soft rock deep well deformation prediction model set to determine the final soft rock deep well deformation prediction model. The present invention comprehensively evaluates the performance of each model and selects the best model for prediction. In addition, the input result of each time is included in the training set, thereby realizing the self-learning function of the model, and then continuously optimizing the prediction accuracy and reliability of the model. In this way, the deformation value of soft rock vertical shaft can be predicted, which can better adapt to the complex situations under different engineering geological conditions and provide technical support for the design, construction and safety assessment of vertical shaft. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0031] Figure 1 A flow chart of a method for predicting deformation value of a soft rock vertical shaft based on machine learning provided in an embodiment of the present invention;

[0032] Figure 2 A detailed schematic diagram of a method for predicting deformation values ​​of a soft rock vertical shaft based on machine learning provided in an embodiment of the present invention;

[0033] Figure 3 A geological model map reflecting real strata provided in an embodiment of the present invention;

[0034] Figure 4 A comparison diagram of the initial data set and the noise data set provided by an embodiment of the present invention;

[0035] Figure 5 A comparison chart of the actual value and the predicted value provided by the embodiment of the present invention;

[0036] Figure 6 A weight analysis diagram of characteristic parameters provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] like Figure 1 As shown, the present invention provides a method for predicting deformation value of a soft rock vertical shaft based on machine learning, comprising:

[0040] Step 100: Acquire the shaft data of the soft rock vertical shaft to be tested;

[0041] Step 200: inputting the soft rock vertical wellbore data to be tested into the constructed soft rock deep well deformation prediction model to obtain a prediction result;

[0042] The method for constructing the soft rock deep well deformation prediction model is as follows:

[0043] Step 201: Obtain engineering geological data from on-site investigation;

[0044] Step 202: extracting intrinsic characteristic parameters and extrinsic characteristic parameters of the soft rock vertical shaft to be tested according to the engineering geological data to determine an initial data set;

[0045] Step 203: introducing noise into the initial data set to obtain a new noise data set;

[0046] Step 204: determining the construction conditions of the soft rock deep well deformation prediction model, and constructing an initial soft rock deep well deformation prediction model set based on random forest, XGBoost, LightGBM and k-nearest neighbor regression according to the new noise data set;

[0047] Step 205: performing accuracy evaluation on the models in the initial soft rock deep well deformation prediction model set to determine a final soft rock deep well deformation prediction model.

[0048] Specifically, Figure 2 As shown, another way of expressing the corresponding method flow of this implementation is:

[0049] Step 1: Collect engineering geological data from on-site investigation, organize and analyze the geological data in detail, and provide reliable basic data for subsequent numerical simulation.

[0050] Specifically, engineering geological data include but are not limited to various physical and mechanical properties of rocks, stratigraphic structure, geological structure and other multifaceted information. The geological information table is shown in Table 1.

[0051] Table 1 Geological information table

[0052]

[0053] Step 2: Establish a geological model that reflects the real strata based on the actual geological conditions, set the corresponding boundary conditions and initial conditions, and extract the intrinsic and extrinsic characteristic parameters to form the initial data set.

[0054] Specifically, based on the geological data collected in step 1, the spatial scope of the geological model is determined, the top and bottom elevations of each layer and its structural information are clarified, and the external boundary and unsteady boundary of the model are set.

[0055] Specifically, a geological model reflecting the real strata is established based on the actual geological conditions based on the FLAC3D finite element numerical simulation software, such as Figure 3 shown.

[0056] 1000 sets of characteristic parameters are extracted from the model. The intrinsic characteristic parameters are parameters that reflect the properties of the formation, including elastic modulus, Poisson's ratio, cohesion, internal friction angle, and rock compressive strength. External characteristic parameters refer to parameters related to the external environment and changing conditions, including maximum tangential stress of surrounding rock and rock compressive strength. Table 2 is a table of characteristic parameters of the geological model, as follows:

[0057] Table 2 Geological model characteristic parameters

[0058]

[0059] The internal and external characteristic parameters are unified and integrated to form an initial data set. Each data item can accurately reflect the properties of a specific geological unit and convert the initial data set into CSV format.

[0060] Step 3: In order to consider the uncertainty in the actual data, such as the influence of the external environment and human measurement errors, noise is introduced to enhance the robustness of the simulation results and reduce the sensitivity of the model to uncertainty, thereby generating a noisy data set.

[0061] The process of generating a noisy dataset described in step 3 includes:

[0062] Since the eigenvalues ​​are usually obtained through sensors, multiplicative noise is used to amplify or reduce the numerical simulation results to generate a noisy data set.

[0063] The process of generating a noise data set in step 3 includes using multiplicative noise to amplify or reduce the numerical simulation results, thereby generating a noise data set. The multiplicative noise calculation formula is as follows: x′=x(1+η); The result of the initial data set after multiplicative noise processing is as follows Figure 4 shown.

[0064] Where x is the initial sample set, x′ is the sample set after noise processing, and η is the noise coefficient, which conforms to the normal distribution.

[0065] Step 4: Using the new noise dataset, select four models including random forest, XGBoost, LightGBM and k-nearest neighbor regression for training and accuracy evaluation, and select the optimal model for soft rock deep well deformation prediction.

[0066] Divide multiple datasets into multiple subsets and perform multiple training and testing to evaluate the average performance and stability of the model, and use independent training and test sets to evaluate the generalization ability of the model.

[0067] The model with the highest R2 and the lowest MSE, RMSE, and MAE was selected as the optimal model;

[0068] RMSE, MAE, and R 2 The calculation expressions are:

[0069]

[0070] RMSE is the root mean square error, MAE is the mean absolute error, yi is the observed value, is the predicted value, is the average value of the predicted value, and n is the number of sample sets.

[0071] The process of using the new noise data set and selecting the optimal algorithm for soft rock deep well deformation prediction described in step 4 includes:

[0072] The generated noise data set is input into the machine learning algorithm, and four models, including random forest, XGBoost, LightGBM and k-nearest neighbor regression, are used for training and accuracy evaluation. During this period, the data is subjected to Bayesian optimization and normalization, and the optimal hyperparameters and feature item weights are calculated. Based on the evaluation accuracy of the four models, the optimal algorithm is selected for soft rock deep well deformation prediction.

[0073] The principle of Bayesian optimization is shown in the following formula.

[0074]

[0075] Among them, f * is a random variable, X,f,x * are known conditions, where X and f may be some existing data or variable sets. Represented as a normal distribution, where μ * is the mean of the normal distribution, is the variance of the normal distribution.

[0076] The principle of normalization processing is shown in the following formula.

[0077]

[0078] Among them, x′ is the original data value, μ is the mean of the original data, σ is the standard deviation of the original data, and x″ is the standardized data value.

[0079] The characteristic parameter values ​​are shown in Table 3. Table 3 is as follows:

[0080] Table 3 Search range values ​​of selected hyperparameters:

[0081]

[0082]

[0083] Furthermore, the accuracy evaluation of the models in the initial soft rock deep well deformation prediction model set to determine the final soft rock deep well deformation prediction model includes:

[0084] Furthermore, the RMSE, MAE and R of each model in the initial soft rock deep well deformation prediction model set are calculated. 2 ;

[0085] According to the RMSE, MAE and R 2 Determine the final soft rock deep well deformation prediction model.

[0086] Specifically, the evaluation results are shown in Table 4. Table 4 is as follows:

[0087] Table 4 Evaluation results

[0088]

[0089]

[0090] In the training set, R 2 =0.9953, RMSE = 0.0006, MAE = 0.0003. In the 10-fold cross validation, R 2=0.9775, RMSE = 0.0015, MAE = 0.0007. The data shows that the performance evaluation of the training set, test set and cross-validation set is consistent, indicating that the model is not overfitting. The hyperparameter values ​​of the random forest model are: n_estimators = 16.57, max_depth = 19.61, min_samples_split = 2.38, max_features = 0.26, min_samples_leaf = 1.07. The prediction results are as follows Figure 5 shown.

[0091] Step 5: Apply the trained model to the prediction of actual engineering data to obtain preliminary prediction results of soft rock shaft deformation values. Analyze the prediction results to determine their rationality and reliability. If any abnormality or unexpected situation is found, promptly check whether there are problems in data input, model parameter setting, etc.

[0092] First, the spatial autocorrelation analysis method is used to calculate the spatial autocorrelation of the deformation values ​​at different depths of the shaft, and the prediction results are pre-analyzed. If the deformation is continuous in the depth direction without mutations, the measured deformation values ​​and the depth position information of the actual engineering data are selected, and compared with the prediction results to determine their reliability; the new data in the actual project are continuously included in the training set to realize the dynamic update and self-learning of the model. With the accumulation of data, the model can better adapt to the changes in engineering geological conditions and various complex situations, further optimize the prediction performance, and improve the prediction accuracy and reliability. The prediction effect of the model is regularly evaluated and verified, and it can be compared and analyzed with the actual monitored wellbore deformation values. According to the evaluation results, the model is adjusted and improved. The dynamic update and self-learning of the model is to use historical data and newly collected data to construct a training set, and re-evaluate and adjust the hyperparameters of the model.

[0093] Specifically, the rationality and reliability described in step 5 include: dividing multiple data sets into multiple subsets, performing multiple training and testing to evaluate the average performance and stability of the model, and using independent training and test sets to evaluate the generalization ability of the model.

[0094] Step 6: Evaluate and verify the prediction effect of the model. The process includes: when some features originally included in the model do not contribute much to the prediction results in multiple evaluations, or even interfere with them, they can be considered to be eliminated. At the same time, if there are some new factors that have an important impact on wellbore deformation but have not yet been included in the model, such as the special properties of specific geological layers, the impact of new engineering activities in the surrounding area, etc., the features related to these factors can be added to the model, so as to continuously improve the prediction ability of the model, make it more in line with the actual situation, and more accurately predict wellbore deformation.

[0095] Specifically, engineering geological data includes but is not limited to various information such as various physical and mechanical properties of rocks, stratigraphic structure, and geological structure.

[0096] Specifically, a geological model reflecting the real strata is established based on the actual geological conditions according to claim 1 based on finite element numerical simulation software, and n groups of characteristic parameters are extracted from the model, including: elastic modulus, Poisson's ratio, cohesion, internal friction angle, rock compressive strength, maximum tangential stress of surrounding rock, and rock compressive strength to form an initial data set, n>1000 groups.

[0097] The process of improving prediction accuracy and reliability described in step 6 includes: collecting user feedback on the model prediction results, understanding the performance and shortcomings of the model in actual applications, further improving the model and self-learning process, and continuously optimizing and adjusting the model.

[0098] Furthermore, the prediction effect of the model can be evaluated and verified regularly by comparing and analyzing the actual monitored wellbore deformation values. According to the evaluation results, the model can be adjusted and improved, such as adjusting model parameters and optimizing feature selection.

[0099] Specifically, when some features originally included in the model do not contribute much to the prediction results in multiple evaluations, or even interfere with them, they can be considered for removal. At the same time, if there are some new factors that have an important impact on wellbore deformation but have not yet been included in the model, such as the special properties of a specific geological layer, the impact of new engineering activities in the surrounding area, etc., the features related to these factors can be added to the model, so as to continuously improve the prediction ability of the model, make it more in line with the actual situation, and predict wellbore deformation more accurately. The weights of each model parameter are as follows: Figure 5 shown.

[0100] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0101] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting deformation value of soft rock vertical shaft based on machine learning, characterized in that: include: Obtain the shaft data of the soft rock vertical shaft to be tested; Inputting the soft rock vertical wellbore data to be tested into the constructed soft rock deep well deformation prediction model to obtain the prediction result; The method for constructing the soft rock deep well deformation prediction model is as follows: Obtain engineering geological data from on-site investigation; Extracting intrinsic characteristic parameters and extrinsic characteristic parameters of the soft rock vertical shaft to be tested according to the engineering geological data to determine an initial data set; Introducing noise into the initial data set to obtain a new noise data set; Determine the construction conditions of the soft rock deep well deformation prediction model, and construct an initial soft rock deep well deformation prediction model set based on random forest, XGBoost, LightGBM and k-nearest neighbor regression according to the new noise data set; The models in the initial soft rock deep well deformation prediction model set are evaluated for accuracy to determine a final soft rock deep well deformation prediction model.

2. The method for predicting deformation value of a soft rock vertical shaft based on machine learning according to claim 1, characterized in that: Also includes: Calculating the spatial autocorrelation of the prediction results to obtain a correlation result; The correlation result is judged to be continuous and sudden, and a judgment result is obtained. If the judgment result is yes, the measured deformation value and the depth position information of the actual engineering data are selected for comparison with the prediction result, and the measured deformation value and the depth position information of the actual engineering data are used to optimize the soft rock deep well deformation prediction model.

3. The method for predicting deformation value of a soft rock vertical shaft based on machine learning according to claim 1, characterized in that: The step of evaluating the accuracy of the models in the initial soft rock deep well deformation prediction model set to determine the final soft rock deep well deformation prediction model includes: Calculate the RMSE, MAE and R of each model in the initial soft rock deep well deformation prediction model set 2 ; According to the RMSE, MAE and R 2 Determine the final soft rock deep well deformation prediction model.

4. The method for predicting deformation value of a soft rock vertical shaft based on machine learning according to claim 1, characterized in that: RMSE, MAE, and R 2 The calculation expressions are: Among them, RMSE is the root mean square error, MAE is the mean absolute error, y i is the observed value, is the predicted value, is the average value of the predicted value, and n is the number of sample sets.

5. The method for predicting deformation value of a soft rock vertical shaft based on machine learning according to claim 1, characterized in that: The intrinsic characteristic parameters include: Elastic modulus, Poisson's ratio, cohesion, internal friction angle, and rock compressive strength.

6. The method for predicting deformation value of a soft rock vertical shaft based on machine learning according to claim 1, characterized in that: The external characteristic parameters include: Maximum tangential stress of surrounding rock and compressive strength of rock mass.

Images