Tunnel over-excavation rate regulation and control method and device

Abstract

The invention provides a tunnel over-excavation rate regulation and control method and device, and relates to the field of tunnel blasting construction control. The method comprises the steps that the current shot hole number, the current explosive load, the current peripheral hole degree, the current auxiliary hole degree, the current underholing hole degree, the current excavation footage, the current surrounding rock grade, the current shot hole depth and the current tunnel face design area are input to a preset over-excavation rate prediction model, and an over-excavation rate prediction result is obtained; and under the condition that the over-excavation rate prediction result is larger than or equal to the preset control over-excavation rate, the current shot hole number, the current explosive load, the current peripheral hole degree, the current auxiliary hole degree and the current slotting hole degree are adjusted according to the regulation and control parameter adjustment sequence, and tunnel blasting construction is conducted according to the target regulation and control parameters. According to the method, through the accurate prediction model and parameter regulation and control, the over-excavation rate of the tunnel is remarkably reduced, the construction precision and efficiency are improved, and through reasonable parameter adjustment, unnecessary material waste and labor cost can be reduced.

Description

Technical Field

[0001] The invention relates to the field of tunnel blasting construction control, and in particular to a tunnel over-excavation rate control method and device. Background Art

[0002] The drilling and blasting method is economical and efficient, and is one of the main construction methods for tunnel excavation. However, due to differences in geological characteristics, blasting arrangements and construction environment, over-excavation and under-excavation of the excavation section often occur during the construction process. Over-excavation of the tunnel section will lead to additional earth and stone transportation, while increasing the amount of concrete for initial support, increasing the project cost, and affecting the construction progress. In addition, the additional damage to the soil around the excavation contour line caused by over-excavation will lead to a decrease in the mechanical properties of the surrounding rock, affecting the bearing capacity and stability of the surrounding rock. Therefore, controlling the tunnel over-excavation rate is very critical in tunnel blasting construction.

[0003] The prediction of over-excavation rate of traditional tunnel blasting construction section mainly relies on the experience and judgment of construction personnel, and the prediction results are subjective and uncertain. In recent years, artificial intelligence technology has been continuously applied to the prediction and control of complex engineering problems. However, when facing different data sets as input, the existing prediction models cannot make reasonable predictions based on the characteristics of different data sets, resulting in low model prediction performance.

[0004] In summary, there is currently no technical solution that can solve the above technical problems, and there is no method and device for controlling the tunnel over-excavation rate. Summary of the invention

[0005] The present invention provides a method and device for controlling tunnel over-excavation rate, which regulate key parameters in tunnel blasting construction through a preset over-excavation rate prediction model and a random forest model, so as to achieve the purpose of controlling the over-excavation rate.

[0006] In a first aspect, the present invention provides a method for controlling a tunnel over-excavation rate, comprising:

[0007] Input the current number of blastholes, the current charge, the current peripheral hole degree, the current auxiliary hole degree, the current slot hole degree, the current excavation footage, the current surrounding rock grade, the current blasthole depth and the current face design area into a preset over-excavation rate prediction model to obtain an over-excavation rate prediction result output by the preset over-excavation rate prediction model;

[0008] When the over-break rate prediction result is greater than or equal to the preset control over-break rate, the current number of blastholes, the current charge, the current peripheral hole degree, the current auxiliary hole degree, and the current groove hole degree are adjusted according to the control parameter adjustment order until the over-break rate prediction result is less than the preset control over-break rate, and the target number of blastholes, the target charge, the target peripheral hole degree, the target auxiliary hole degree, and the target groove hole degree are determined;

[0009] Taking the number of target blast holes, the target charge amount, the target perimeter hole degree, the target auxiliary hole degree, the target cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area as target control parameters to perform tunnel blasting construction according to the target control parameters;

[0010] The adjustment order of the control parameters is determined by sorting the number of blast holes, the charge amount, the perimeter hole degree, the auxiliary hole degree, and the cut hole degree from large to small according to the result of the feature importance parameter. The result of the feature importance parameter is obtained by inputting the number of blast holes, the charge amount, the perimeter hole degree, the auxiliary hole degree, the cut hole degree, the excavation footage, the surrounding rock grade, the blast hole depth, and the face design area into a preset random forest model and output by the preset random forest model;

[0011] The preset overbreak rate prediction model is determined after training according to the sample number of blast holes, the sample charge amount, the sample perimeter hole degree, the sample auxiliary hole degree, the sample cut hole degree, the sample excavation footage, the sample surrounding rock grade, the sample blast hole depth, the sample face design area, and the sample overbreak rate prediction result.

[0012] According to the tunnel overbreak rate control method provided by the present invention, when the adjustment order of the control parameters is successively the number of blast holes, the charge amount, the perimeter hole degree, the auxiliary hole degree, and the cut hole degree, adjusting the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, and the current cut hole degree according to the adjustment order of the control parameters includes:

[0013] Increasing the number of blast holes, inputting the increased number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into the preset overbreak rate prediction model to obtain a first prediction result output by the preset overbreak rate prediction model;

[0014] When the first prediction result is greater than or equal to the preset control overbreak rate, decreasing the charge amount, inputting the increased number of blast holes, the decreased charge amount, the current perimeter hole degree, the current auxiliary hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into the preset overbreak rate prediction model to obtain a second prediction result output by the preset overbreak rate prediction model;

[0015] When the second predicted result is greater than or equal to the preset control overbreak rate, reduce the number of perimeter holes, and input the increased number of blast holes, the reduced charge amount, the reduced perimeter hole degree, the current relief hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into the preset overbreak rate prediction model to obtain the third predicted result output by the preset overbreak rate prediction model;

[0016] When the third predicted result is greater than or equal to the preset control overbreak rate, increase the relief hole degree, and input the increased number of blast holes, the reduced charge amount, the reduced perimeter hole degree, the increased relief hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into the preset overbreak rate prediction model to obtain the fourth predicted result output by the preset overbreak rate prediction model;

[0017] When the fourth predicted result is greater than or equal to the preset control overbreak rate, increase the cut hole degree, and input the increased number of blast holes, the reduced charge amount, the reduced perimeter hole degree, the increased relief hole degree, the increased cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into the preset overbreak rate prediction model to obtain the fifth predicted result output by the preset overbreak rate prediction model.

[0018] According to the tunnel overbreak rate regulation method provided by the present invention, until the overbreak rate prediction result is less than the preset control overbreak rate, determining the target number of blast holes, the target charge amount, the target perimeter hole degree, the target relief hole degree, and the target cut hole degree includes:

[0019] When the fifth predicted result is greater than or equal to the preset control overbreak rate, increase the cut hole degree again, and input the increased number of blast holes, the reduced charge amount, the reduced perimeter hole degree, the increased relief hole degree, the cut hole degree increased again, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into the preset overbreak rate prediction model to obtain the sixth predicted result output by the preset overbreak rate prediction model;

[0020] Until the sixth predicted result is less than the preset control overbreak rate, determine that the target number of blast holes is the increased number of blast holes, determine that the target charge amount is the reduced charge amount, determine that the target perimeter hole degree is the reduced perimeter hole degree, determine that the target relief hole degree is the increased relief hole degree, and determine that the target cut hole degree is the cut hole degree increased again.

[0021] According to the tunnel overbreak rate regulation method provided by the present invention, before inputting the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into a preset overbreak rate prediction model, the method further includes:

[0022] Determine the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area according to the excavation site data in the tunnel blasting construction record;

[0023] Normalize all excavation site data to obtain test set data and training set data, calculate the range of the number of hidden layer neurons, define the value range of the regularization coefficient, divide the training set data into K folds for cross-validation, and use different activation functions to perform grid search on the regularization coefficient and the number of hidden layer neurons respectively to obtain the regularization coefficient and the number of hidden layer neurons with the optimal determination coefficient;

[0024] Import the test set data, perform cross-validation on the test set data, determine the determination coefficient and the absolute mean square error corresponding to each activation function, and determine the target activation function according to the determination coefficient and the absolute mean square error corresponding to different activation functions;

[0025] Construct the preset overbreak rate prediction model by using the target activation function.

[0026] According to the tunnel overbreak rate regulation method provided by the present invention, the different activation functions at least include the identity function, the logistic function, the hyperbolic tangent function, and the rectified linear unit;

[0027] The determining the target activation function according to the determination coefficient and the absolute mean square error corresponding to different activation functions includes:

[0028] Determine the activation function with the largest determination coefficient and the smallest absolute mean square error as the target activation function.

[0029] According to the tunnel overbreak rate regulation method provided by the present invention, when the excavation site data is the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area, determine that the target activation function is the rectified linear unit Relu.

[0030] According to the tunnel overbreak rate regulation method provided by the present invention, using different activation functions to perform grid search on the regularization coefficient and the number of hidden layer neurons respectively to obtain the regularization coefficient with the optimal determination coefficient and the number of hidden layer neurons, including:

[0031] For any activation function, based on the test set data, construct a three-layer BP neural network model including an input layer, a hidden layer, and an output layer. The configuration of the neural network model includes the number of input variables, the number of output variables, the number of hidden layers, the number of samples in the training sample set, the number of folds in K-fold cross-validation, the optimizer, the initial learning rate, the regularization method, and the maximum number of training steps;

[0032] Using a preset grid search algorithm, the range of the number of hidden layer neurons, and the range of values of the regularization coefficient, combined with K-fold cross-validation, perform a traversal search on the number of hidden layer neurons and the regularization coefficient of the multi-layer perceptron regression model. During the search process, perform multiple trainings on the data set, record the running time and performance score of each training, and obtain the regularization coefficient with the optimal determination coefficient and the number of hidden layer neurons.

[0033] According to the tunnel overbreak rate regulation method provided by the present invention, before adjusting the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, and the current cut hole degree according to the adjustment order of the regulation parameters, the method further includes:

[0034] Construct a preset random forest model with the input layer feature data of each section before tunnel blasting construction and the overbreak rate of the corresponding section after construction as the construction training set;

[0035] During the construction process, randomly extract samples from the construction training set multiple times to construct decision trees, and randomly select features for splitting at each node to construct multiple decision trees;

[0036] Calculate the contribution degree of each input layer feature data during the splitting process of the decision tree, and calculate the average value of the contribution degrees of the input layer feature data in all decision trees to obtain the feature importance parameter result of the input layer feature data;

[0037] Sort the number of blast holes, the charge amount, the perimeter hole degree, the auxiliary hole degree, and the cut hole degree from large to small according to the feature importance parameter result to determine the adjustment order of the regulation parameters.

[0038] In the second aspect, a tunnel overbreak rate regulation device is provided, including:

[0039] An input unit for inputting the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into a preset overbreak rate prediction model to obtain an overbreak rate prediction result output by the preset overbreak rate prediction model;

[0040] A determination unit for, when the overbreak rate prediction result is greater than or equal to a preset controlled overbreak rate, adjusting the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, and the current cut hole degree according to the adjustment order of the control parameters until the overbreak rate prediction result is less than the preset controlled overbreak rate, and determining the target number of blast holes, the target charge amount, the target perimeter hole degree, the target auxiliary hole degree, and the target cut hole degree;

[0041] A processing unit for using the target number of blast holes, the target charge amount, the target perimeter hole degree, the target auxiliary hole degree, the target cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area as target control parameters to perform tunnel blasting construction according to the target control parameters;

[0042] The adjustment order of the control parameters is determined after sorting the number of blast holes, the charge amount, the perimeter hole degree, the auxiliary hole degree, and the cut hole degree from large to small according to the characteristic importance parameter result, and the characteristic importance parameter result is determined by inputting the number of blast holes, the charge amount, the perimeter hole degree, the auxiliary hole degree, the cut hole degree, the excavation footage, the surrounding rock grade, the blast hole depth, and the face design area into a preset random forest model and output by the preset random forest model;

[0043] The preset overbreak rate prediction model is determined after being trained according to the sample number of blast holes, the sample charge amount, the sample perimeter hole degree, the sample auxiliary hole degree, the sample cut hole degree, the sample excavation footage, the sample surrounding rock grade, the sample blast hole depth, the sample face design area, and the sample overbreak rate prediction result.

[0044] Through precise prediction models and parameter regulation, the present invention can significantly reduce the over-excavation rate of tunnels, improve construction accuracy and efficiency. Through reasonable parameter adjustment, unnecessary material waste and labor costs can be reduced, construction resource allocation can be optimized. Precise blasting construction parameters contribute to reducing safety risks during construction and improving construction safety. Through an automated prediction and regulation process, the optimal construction parameters can be quickly determined, improving decision-making efficiency. The prediction model obtained by training a large number of sample data has good generalization ability and can adapt to tunnel blasting construction under different geological and construction conditions. According to the adjustment sequence determined by the feature importance parameter results, construction parameters can be adjusted in an orderly and efficient manner, avoiding time and resource waste caused by blind adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0046] Figure 1 It is a schematic flow chart of the method for regulating the over-excavation rate of tunnels provided by the present invention;

[0047] Figure 2 It is a heat map of the hyperparameters and coefficient of determination of the Identity activation function provided by the present invention;

[0048] Figure 3 It is a heat map of the hyperparameters and coefficient of determination of the Logistic activation function provided by the present invention;

[0049] Figure 4 It is a heat map of the hyperparameters and coefficient of determination of the Tanh activation function provided by the present invention;

[0050] Figure 5 It is a heat map of the hyperparameters and coefficient of determination of the Relu activation function provided by the present invention;

[0051] Figure 6 It is a comparison chart of predicted values and true values under different activation functions provided by the present invention;

[0052] Figure 7 It is a feature importance parameter diagram provided by the present invention;

[0053] Figure 8 It is a BP neural network diagram for over-excavation rate prediction provided by the present invention;

[0054] Figure 9 It is a schematic structural diagram of the device for regulating the over-excavation rate of tunnels provided by the present invention. Specific implementation mode

[0055] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0056] Figure 1 It is a schematic flowchart of a method for regulating the overbreak rate of a tunnel provided by the present invention. The method for regulating the overbreak rate of a tunnel includes:

[0057] Step 101: Input the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into a preset overbreak rate prediction model, and obtain the overbreak rate prediction result output by the preset overbreak rate prediction model;

[0058] Step 102: In the case where the overbreak rate prediction result is greater than or equal to a preset controlled overbreak rate, adjust the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, and the current cut hole degree according to the adjustment order of the regulation parameters until the overbreak rate prediction result is less than the preset controlled overbreak rate, and determine the target number of blast holes, the target charge amount, the target perimeter hole degree, the target auxiliary hole degree, and the target cut hole degree;

[0059] Step 103: Take the target number of blast holes, the target charge amount, the target perimeter hole degree, the target auxiliary hole degree, the target cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area as target regulation parameters, so as to perform tunnel blasting construction according to the target regulation parameters;

[0060] The adjustment order of the regulation parameters is determined by sorting the number of blast holes, the charge amount, the perimeter hole degree, the auxiliary hole degree, and the cut hole degree from large to small according to the result of the characteristic importance parameter. The result of the characteristic importance parameter is obtained by inputting the number of blast holes, the charge amount, the perimeter hole degree, the auxiliary hole degree, the cut hole degree, the excavation footage, the surrounding rock grade, the blast hole depth, and the face design area into a preset random forest model and output by the preset random forest model;

[0061] The preset overbreak rate prediction model is determined after training based on the number of sample blast holes, sample charge amount, sample perimeter hole degree, sample auxiliary hole degree, sample cut hole degree, sample excavation footage, sample surrounding rock grade, sample blast hole depth, sample face design area, and sample overbreak rate prediction results.

[0062] In step 101, by inputting the parameters of the current blasting construction (such as the number of blast holes, charge amount, hole degree, excavation footage, surrounding rock grade, blast hole depth, face design area), using the trained overbreak rate prediction model, the current overbreak rate can be predicted. This prediction provides data support for subsequent regulation, enabling construction personnel to understand the possible overbreak situation in advance and thus take corresponding measures for adjustment. Suppose the current number of blast holes is 100, the charge amount is 50 kg, the perimeter hole degree is 45°, the auxiliary hole degree is 30°, the cut hole degree is 60°, the excavation footage is 3 m, the surrounding rock grade is grade III, the blast hole depth is 2.5 m, and the face design area is 100 m². Inputting these parameters into the overbreak rate prediction model, the predicted overbreak rate is 0.5%.

[0063] In step 102, when the predicted overbreak rate is greater than or equal to the preset controlled overbreak rate, according to the sorting of the characteristic importance parameter results, the number of blast holes, charge amount, perimeter hole degree, auxiliary hole degree, and cut hole degree are adjusted in sequence until the overbreak rate prediction result is less than the preset controlled overbreak rate. By gradually adjusting the parameters, precise control of the overbreak rate is achieved, avoiding resource waste and safety hazards caused by excessive overbreak.

[0064] In step 103, the adjusted parameters are used as target regulation parameters to guide tunnel blasting construction, which can ensure that the overbreak rate during construction is controlled within the preset range, improving construction quality and efficiency. After determining the parameters such as the target number of blast holes and target charge amount, tunnel blasting construction is carried out according to these parameters. During the construction process, the overbreak situation should be closely monitored and fine-tuned according to the actual situation to ensure construction quality and safety.

[0065] Optionally, by inputting a large amount of historical data into the preset random forest model, the model can output the importance ranking of each feature (such as the number of blast holes, charge amount, etc.) for overbreak rate prediction, providing a basis for parameter adjustment in step 102. By collecting a large amount of historical data (including the number of sample blast holes, sample charge amount, etc.) and the corresponding overbreak rate prediction results, and using machine learning algorithms (such as neural networks, support vector machines, etc.) for training, a model that can accurately predict the overbreak rate is obtained. Algorithms such as BP neural networks and random forests have powerful non-linear mapping capabilities and self-learning capabilities, and can handle complex data relationships. Applying them to the field of overbreak rate prediction of tunnel blasting construction excavation sections has important research value and application prospects.

[0066] Optionally, when the adjustment order of the control parameters is the number of blast holes, the charge amount, the degree of perimeter holes, the degree of auxiliary holes, and the degree of cut holes in sequence, the adjustment of the current number of blast holes, the current charge amount, the current degree of perimeter holes, the current degree of auxiliary holes, and the current degree of cut holes according to the adjustment order of the control parameters includes:

[0067] Increase the number of blast holes, input the increased number of blast holes, the current charge amount, the current degree of perimeter holes, the current degree of auxiliary holes, the current degree of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into the preset overbreak rate prediction model, and obtain the first prediction result output by the preset overbreak rate prediction model;

[0068] When the first prediction result is greater than or equal to the preset control overbreak rate, decrease the charge amount, input the increased number of blast holes, the decreased charge amount, the current degree of perimeter holes, the current degree of auxiliary holes, the current degree of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into the preset overbreak rate prediction model, and obtain the second prediction result output by the preset overbreak rate prediction model;

[0069] When the second prediction result is greater than or equal to the preset control overbreak rate, decrease the degree of perimeter holes, input the increased number of blast holes, the decreased charge amount, the decreased degree of perimeter holes, the current degree of auxiliary holes, the current degree of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into the preset overbreak rate prediction model, and obtain the third prediction result output by the preset overbreak rate prediction model;

[0070] When the third prediction result is greater than or equal to the preset control overbreak rate, increase the degree of auxiliary holes, input the increased number of blast holes, the decreased charge amount, the decreased degree of perimeter holes, the increased degree of auxiliary holes, the current degree of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into the preset overbreak rate prediction model, and obtain the fourth prediction result output by the preset overbreak rate prediction model;

[0071] When the fourth prediction result is greater than or equal to the preset control overbreak rate, increase the degree of cut holes, input the increased number of blast holes, the decreased charge amount, the decreased degree of perimeter holes, the increased degree of auxiliary holes, the increased degree of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into the preset overbreak rate prediction model, and obtain the fifth prediction result output by the preset overbreak rate prediction model.

[0072] Optionally, in this embodiment, the prediction and regulation parameters are as shown in the table. First, the over-excavation rate of the original data is predicted, and the result is 0.06687. It is expected to control the over-excavation rate to be less than 0.02. The control process and results are as shown in the following table. By controlling the blasting parameters, the predicted over-excavation rate is 0.01819.

[0073]

[0074]

[0075] Table of Over-excavation Rate Regulation Process

[0076] Optionally, the process from the original data to Adjustment 1 is the process of obtaining the first prediction result output by the preset over-excavation rate prediction model; the process from Adjustment 1 to Adjustment 2 is the process of obtaining the second prediction result output by the preset over-excavation rate prediction model; the process from Adjustment 2 to Adjustment 3 is the process of obtaining the third prediction result output by the preset over-excavation rate prediction model; the process from Adjustment 3 to Adjustment 4 is the process of obtaining the fourth prediction result output by the preset over-excavation rate prediction model; the process from Adjustment 4 to Adjustment 5 is the process of obtaining the fifth prediction result output by the preset over-excavation rate prediction model.

[0077] Optionally, the process from Adjustment 5 to Adjustment 6 is as follows: until the over-excavation rate prediction result is less than the preset controlled over-excavation rate, determining the target number of blast holes, target charge amount, target perimeter hole degree, target auxiliary hole degree, and target cut hole degree, including:

[0078] In the case where the fifth prediction result is greater than or equal to the preset controlled over-excavation rate, the cut hole degree is increased again. Input the increased number of blast holes, decreased charge amount, decreased perimeter hole degree, increased auxiliary hole degree, increased cut hole degree again, current excavation footage, current surrounding rock grade, current blast hole depth, and current face design area into the preset over-excavation rate prediction model to obtain the sixth prediction result output by the preset over-excavation rate prediction model;

[0079] Until the sixth prediction result is less than the preset controlled over-excavation rate, determine that the target number of blast holes is the increased number of blast holes, determine the target charge amount is the decreased charge amount, determine the target perimeter hole degree is the decreased perimeter hole degree, determine the target auxiliary hole degree is the increased auxiliary hole degree, and determine the target cut hole degree is the increased cut hole degree again.

[0080] Optionally, in the embodiment of the present invention, through a clear adjustment sequence (number of blast holes → charge amount → degree of perimeter holes → degree of auxiliary holes → degree of cut holes), it is ensured that each adjustment is based on the results of the previous steps and is carried out in the order from the largest to the smallest degree of influence on the overbreak rate. This systematic process helps to quickly and effectively find the parameter combination that meets the overbreak rate requirements. After each adjustment, a preset overbreak rate prediction model is used for prediction to ensure that the adjusted parameter combination can make the overbreak rate lower than the preset control overbreak rate. This real-time feedback mechanism helps to improve the accuracy of overbreak rate control. By gradually adjusting the parameters and observing the prediction results, it is possible to quickly identify which parameters have a greater impact on the overbreak rate and make targeted adjustments accordingly. This method can find the optimal parameter combination faster than the traditional trial-and-error method, thereby improving the construction efficiency. By precisely controlling the overbreak rate, material waste and construction period delays caused by excessive overbreak can be reduced. At the same time, reasonable blasting parameters also help to reduce the degree of damage to the surrounding rock and reduce potential safety hazards.

[0081] Optionally, before inputting the current number of blast holes, current charge amount, current degree of perimeter holes, current degree of auxiliary holes, current degree of cut holes, current excavation footage, current surrounding rock grade, current blast hole depth, and current face design area into the preset overbreak rate prediction model, the method further includes:

[0082] Determine the current number of blast holes, current charge amount, current degree of perimeter holes, current degree of auxiliary holes, current degree of cut holes, current excavation footage, current surrounding rock grade, current blast hole depth, and current face design area according to the excavation site data in the tunnel blasting construction record;

[0083] Normalize all excavation site data to obtain test set data and training set data, calculate the range of the number of hidden layer neurons, define the value range of the regularization coefficient, divide the training set data into K folds for cross-validation, and use different activation functions to perform grid search optimization on the regularization coefficient and the number of hidden layer neurons respectively to obtain the regularization coefficient and the number of hidden layer neurons with the optimal determination coefficient;

[0084] Import the test set data, perform cross-validation on the test set data, determine the determination coefficient and absolute mean square error corresponding to each activation function, and determine the target activation function according to the determination coefficient and absolute mean square error corresponding to different activation functions;

[0085] Construct the preset overbreak rate prediction model using the target activation function.

[0086] Optionally, collect the data of the input layer features of each section before the tunnel blasting construction and the overbreak rate of the corresponding section after the construction, and divide them into a training set and a test set according to a ratio of about 4:1. Perform normalization preprocessing on the features X of the training set data to make the training set data between 0 and 1. The normalization expression is:

[0087]

[0088] Calculate the range of the number of neurons in the hidden layer, from the number of features n in the input layer input and the number of target numbers n in the output layer output Calculate the range of the number of neurons in the hidden layer (rounded up). The calculation formula is:

[0089]

[0090] Define the range of the regularization coefficient α. The defined range of the regularization coefficient α is [10 -5 , 10 5 , with 11 logarithmically equally spaced values. Divide the training set into k folds for cross-validation. Shuffle the training set data and divide it into k subsets of similar size and non-overlapping. Then take each subset as the validation set in turn, and the remaining k - 1 subsets as the training set for k times of training and validation. Use four activation functions for grid hyperparameter optimization. Optionally, the different activation functions at least include the identity function, the logistic function, the hyperbolic tangent function, and the rectified linear unit;

[0091] Determine the target activation function according to the coefficient of determination and the absolute mean square error corresponding to different activation functions, including:

[0092] Determine the activation function with the largest coefficient of determination and the smallest absolute mean square error as the target activation function.

[0093] Optionally, use the identity function (Identity), the logistic function (Logistic), the hyperbolic tangent function (Tanh), and the rectified linear unit (Relu) four activation functions respectively for hyperparameter grid optimization, and use the coefficient of determination R 2 to evaluate the fitting degree. For each activation function, the present invention uses the grid search algorithm and the defined hyperparameter range, combined with k-fold cross-validation, to perform a traversal search on the number of neurons in the hidden layer and the regularization coefficient α of the multi-layer perceptron regression model. Figure 2 is the heat map of the hyperparameters and the coefficient of determination of the Identity activation function provided by the present invention, Figure 3 is the heat map of the hyperparameters and the coefficient of determination of the Logistic activation function provided by the present invention, Figure 4 is the heat map of the hyperparameters and the coefficient of determination of the Tanh activation function provided by the present invention, Figure 5During the search process of the heatmap of the hyperparameters and the coefficient of determination of the Relu activation function provided by the present invention, the model will perform multiple trainings on the training data set, record the running time and performance score R of each training 2 and generate a heatmap corresponding to the performance score of the activation function and the hyperparameter values.

[0094] Furthermore, import the test set data, perform the same normalization process as the training set, input the optimal parameters, perform cross-validation on the test set, input the regularization coefficient α and the number of neurons in the hidden layer of the obtained optimal coefficient of determination, and draw a comparison graph of the predicted value and the true value Figure 6 is the comparison graph of the predicted value and the true value under different activation functions provided by the present invention, calculate the absolute error between the two, and output the coefficient of determination R 2 , evaluate the fitting degree of the training model according to the coefficient of determination and the error size, and select the optimal training model.

[0095] Optionally, the present invention uses the random forest algorithm to generate a comparison graph of feature importance Figure 7 is the feature importance parameter graph provided by the present invention. Use the random forest algorithm to evaluate the feature importance of the training set to obtain a comparison graph of feature importance, adjust the blasting construction parameters, and control the size of the overbreak rate. When the overbreak rate of the predicted section is large, the construction parameters of the main influencing factors (i.e., features with larger importance parameters) can be adjusted one by one to control the size of the overbreak rate.

[0096] In an optional embodiment, on-site data of 9 input layer features including the excavation footage, surrounding rock grade, face area, number of blast holes, depth of blast holes, charge amount, and blast hole inclination angle (cutting holes, auxiliary holes, perimeter holes) are recorded during tunnel blasting construction, and the overbreak rate after excavation of the corresponding section is recorded, obtaining a total of 269 pieces of all data. The specific implementation steps are as follows:

[0097] Input the training set data and normalize it. Randomly sample 59 pieces of the existing data as the test set, and the remaining 210 pieces of data as the training set. Normalize the training set data to obtain dimensionless data, calculate the range of the number of neurons in the hidden layer, and calculate the range of the number of neurons in the hidden layer from the number of features n input =9 of the input layer and the number of output layer targets n output =1. Calculate the range of the number of neurons in the hidden layer, and calculate to obtain n min =3.169, n max =20. After rounding up, the number of neurons in the hidden layer is 4-20. Define the range of the regularization coefficient α. The range of the regularization coefficient α is defined as [10 -5 , 10 5 , with a logarithmic equally spaced distribution of 11 values.

[0098] Divide the training set into K folds for cross validation, perform 5-fold cross validation, shuffle the training set data into 5 subsets of similar size and non-overlapping, then use each subset as the validation set in turn, and the remaining 4 subsets as the training set, and perform 5 training and validation. (The number of folds should be trained multiple times according to the determination coefficient R 2 Take the best fold.)

[0099] Optionally, the method of using different activation functions to perform grid optimization on the regularization coefficient and the number of hidden layer neurons to obtain the regularization coefficient of the optimal determination coefficient and the number of hidden layer neurons includes:

[0100] For any activation function, a three-layer BP neural network model including an input layer, a hidden layer and an output layer is constructed based on the test set data, and the configuration of the neural network model includes the number of input variables, the number of output variables, the number of hidden layers, the number of sample points in the training sample set, the number of K-fold cross validation folds, the optimizer, the initial learning rate, the regularization method and the maximum number of training steps;

[0101] Using the preset grid search algorithm, the range of the number of hidden layer neurons and the range of the regularization coefficient, combined with K-fold cross validation, a traversal search is performed on the number of hidden layer neurons and the regularization coefficient of the multilayer perceptron regression model. During the search process, multiple trainings are performed on the data set, and the running time and performance score of each training are recorded to obtain the regularization coefficient of the optimal determination coefficient and the number of hidden layer neurons.

[0102] Optionally, four activation functions are used for grid hyperparameter optimization. The four activation functions of Identity, Logistic, Tanh and Relu are used for training respectively. The two hyperparameters of regularization coefficient α and number of neurons in hidden layer are grid optimized based on the previously obtained range. Based on the normalized test set data, a three-layer BP neural network model with input layer, hidden layer and output layer is constructed. Figure 8 The BP neural network diagram for over-digging rate prediction provided by the present invention has network structure parameters and configuration as shown in the following table:

[0103] Neural Network Configuration Instructions

[0104] Name Configuration Value Number of Input Variables 9 Number of Output Variables 1 Number of Hidden Layers 1 Number of Samples in Training Sample Set 210 Number of Folds in K-Fold Cross-Validation 5 Optimizer lbfgs Initial Learning Rate 0.001 Regularization Method L2 Maximum Number of Training Steps 5000

[0105] According to the above neural network, the optimal hyperparameters of the determination coefficients under the four activation functions are shown in the following table:

[0106] Optimal hyperparameters for four activation functions

[0107]

[0108] Optionally, import the test set data and perform the same normalization process as the training set. Import 59 test set data and perform the same normalization process as the training set. Input the optimal parameters, perform cross-validation on the testing machine, input the regularization coefficient α and the number of neurons in the hidden layer of the obtained optimal coefficient of determination, plot the comparison graph of the predicted values and the true values, calculate the absolute error between the two, and output the coefficient of determination R 2 The fitting degree of the training model is evaluated according to the error size as shown in the following table. The predicted values and the actual values are as Figure 6 shown

[0109] Coefficient of determination and error of the test set

[0110]

[0111]

[0112] Optionally, before adjusting the current number of blast holes, the current charge amount, the current perimeter hole degree, the current relief hole degree, and the current cut hole degree according to the adjustment order of the control parameters, the method further includes:

[0113] Construct a preset random forest model with the input layer feature data of each section before tunnel blasting construction and the overbreak rate of the corresponding section after construction as the construction training set;

[0114] During the construction process, randomly extract samples from the construction training set multiple times to construct decision trees, and randomly select features to split at each node to construct multiple decision trees;

[0115] Calculate the contribution degree of each input layer feature data during the splitting process of the decision tree, and calculate the average value of the contribution degrees of the input layer feature data in all decision trees to obtain the feature importance parameter result of the input layer feature data;

[0116] Sort the number of blast holes, the charge amount, the perimeter hole degree, the relief hole degree, and the cut hole degree from large to small according to the feature importance parameter result to determine the adjustment order of the control parameters.

[0117] Using the random forest algorithm, a feature importance comparison graph is generated. When the excavation site data is the current number of blast holes, the current charge amount, the current perimeter hole degree, the current relief hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area, the target activation function is determined to be the rectified linear unit Relu. The present invention selects a training model with the activation function Relu for overbreak rate prediction. For cross-sections with a relatively large predicted overbreak rate, the blasting parameters are controlled to adjust the magnitude of the overbreak rate. The control method is as follows: The random forest model is trained using the training set data, where the number of trees in the random forest is set to 15,000, and 30% of the original training set data is set as the new training set to generate the importance parameters of each input feature. The results are as Figure 7 shown.

[0118] Optionally, the surrounding rock grade and the face design area cannot be adjusted. Considering the construction period, the excavation footage is not controlled either. According to the Figure 7 feature importance evaluation results, the five blasting parameters of the number of blast holes, the charge amount, the perimeter hole degree, the relief hole degree, and the cut hole degree are adjusted multiple times (within the specified range) from large to small and input into the trained surrogate model to make the magnitude of the overbreak rate meet the given range.

[0119] Compared with the prior art, the present invention trains a surrogate model through historical data in tunnel blasting construction to achieve accurate prediction of the overbreak rate of the excavation cross-section during tunnel blasting construction, reduces the dependence on construction experience, automatically tunes the number of hidden layer neurons and the regularization strength α of the surrogate model, traverses different parameter combinations, and finds the optimal parameter configuration for a specific data set, thereby significantly improving the prediction performance of the model; supports training models with 4 activation functions, including the identity function, the logistic function, the hyperbolic tangent function, and the rectified linear unit, to adapt to the characteristics of different data sets, and uses the feature importance parameters of the random forest to assist in controlling the overly large predicted cross-section overbreak rate, providing scientific and efficient guidance for tunnel blasting construction, thereby improving the efficiency and quality of tunnel construction.

[0120] Figure 9 FIG. 14 is a schematic structural diagram of a tunnel overbreak rate control device provided by the present invention. The tunnel overbreak rate control device includes:

[0121] An input unit 1, which is used to input the current number of blast holes, the current charge amount, the current perimeter hole degree, the current relief hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area into a preset overbreak rate prediction model to obtain the overbreak rate prediction result output by the preset overbreak rate prediction model;

[0122] A determination unit 2, which is configured to, when the overbreak rate prediction result is greater than or equal to a preset controlled overbreak rate, adjust the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, and the current cut hole degree according to the adjustment sequence of the regulation parameters until the overbreak rate prediction result is less than the preset controlled overbreak rate, and determine the target number of blast holes, the target charge amount, the target perimeter hole degree, the target auxiliary hole degree, and the target cut hole degree;

[0123] A processing unit 3, which is configured to use the target number of blast holes, the target charge amount, the target perimeter hole degree, the target auxiliary hole degree, the target cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area as target regulation parameters to perform tunnel blasting construction according to the target regulation parameters;

[0124] The adjustment sequence of the regulation parameters is determined after sorting the number of blast holes, the charge amount, the perimeter hole degree, the auxiliary hole degree, and the cut hole degree from large to small according to the feature importance parameter result, and the feature importance parameter result is determined by inputting the number of blast holes, the charge amount, the perimeter hole degree, the auxiliary hole degree, the cut hole degree, the excavation footage, the surrounding rock grade, the blast hole depth, and the face design area into a preset random forest model and output by the preset random forest model;

[0125] The preset overbreak rate prediction model is determined after being trained according to the sample number of blast holes, the sample charge amount, the sample perimeter hole degree, the sample auxiliary hole degree, the sample cut hole degree, the sample excavation footage, the sample surrounding rock grade, the sample blast hole depth, the sample face design area, and the sample overbreak rate prediction result.

[0126] Through the precise prediction model and parameter regulation, the present invention can significantly reduce the overbreak rate of the tunnel, improve the construction accuracy and efficiency. Through reasonable parameter adjustment, unnecessary material waste and labor costs can be reduced, the construction resource allocation can be optimized. The precise blasting construction parameters help to reduce the safety risks during the construction process and improve the construction safety. Through the automated prediction and regulation process, the optimal construction parameters can be quickly determined, and the decision-making efficiency can be improved. The prediction model obtained by training a large amount of sample data has good generalization ability and can adapt to tunnel blasting construction under different geological and construction conditions. The adjustment sequence determined according to the feature importance parameter result can orderly and efficiently adjust the construction parameters and avoid the time and resource waste caused by blind adjustment.

[0127] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative work.

[0128] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for regulating the over-excavation rate of a tunnel, characterized in that, Including: Input the current number of blast holes, the current charge amount, the current degree of perimeter holes, the current degree of auxiliary holes, the current degree of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current designed face area into a preset overbreak rate prediction model to obtain the overbreak rate prediction result output by the preset overbreak rate prediction model; When the overbreak rate prediction result is greater than or equal to a preset controlled overbreak rate, adjust the current number of blast holes, the current charge amount, the current degree of perimeter holes, the current degree of auxiliary holes, and the current degree of cut holes according to the adjustment sequence of control parameters until the overbreak rate prediction result is less than the preset controlled overbreak rate, and determine the target number of blast holes, the target charge amount, the target degree of perimeter holes, the target degree of auxiliary holes, and the target degree of cut holes; Use the target number of blast holes, the target charge amount, the target degree of perimeter holes, the target degree of auxiliary holes, the target degree of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current designed face area as target control parameters to perform tunnel blasting construction according to the target control parameters; The adjustment sequence of the control parameters is determined by sorting the number of blast holes, the charge amount, the degree of perimeter holes, the degree of auxiliary holes, and the degree of cut holes from large to small according to the characteristic importance parameter result. The characteristic importance parameter result is obtained by inputting the number of blast holes, the charge amount, the degree of perimeter holes, the degree of auxiliary holes, the degree of cut holes, the excavation footage, the surrounding rock grade, the blast hole depth, and the designed face area into a preset random forest model and output by the preset random forest model; The preset overbreak rate prediction model is determined after training based on the sample number of blast holes, the sample charge amount, the sample degree of perimeter holes, the sample degree of auxiliary holes, the sample degree of cut holes, the sample excavation footage, the sample surrounding rock grade, the sample blast hole depth, the sample designed face area, and the sample overbreak rate prediction result.

2. The tunneling overbreak rate regulation method according to claim 1, wherein When the adjustment sequence of the control parameters is successively the number of blast holes, the charge amount, the degree of perimeter holes, the degree of auxiliary holes, and the degree of cut holes, the adjusting the current number of blast holes, the current charge amount, the current degree of perimeter holes, the current degree of auxiliary holes, and the current degree of cut holes according to the adjustment sequence of control parameters includes: Increase the number of blast holes, input the increased number of blast holes, the current charge amount, the current degree of perimeter holes, the current degree of auxiliary holes, the current degree of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current designed face area into the preset overbreak rate prediction model to obtain the first prediction result output by the preset overbreak rate prediction model; When the first prediction result is greater than or equal to the preset controlled overbreak rate, decrease the charge amount, input the increased number of blast holes, the decreased charge amount, the current degree of perimeter holes, the current degree of auxiliary holes, the current degree of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current designed face area into the preset overbreak rate prediction model to obtain the second prediction result output by the preset overbreak rate prediction model; When the second predicted result is greater than or equal to the preset control overbreak rate, reduce the number of perimeter holes, and input the increased number of blast holes, the reduced charge amount, the reduced number of perimeter holes, the current number of relief holes, the current number of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current designed face area into the preset overbreak rate prediction model to obtain the third predicted result output by the preset overbreak rate prediction model; When the third predicted result is greater than or equal to the preset control overbreak rate, increase the number of relief holes, and input the increased number of blast holes, the reduced charge amount, the reduced number of perimeter holes, the increased number of relief holes, the current number of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current designed face area into the preset overbreak rate prediction model to obtain the fourth predicted result output by the preset overbreak rate prediction model; When the fourth predicted result is greater than or equal to the preset control overbreak rate, increase the number of cut holes, and input the increased number of blast holes, the reduced charge amount, the reduced number of perimeter holes, the increased number of relief holes, the increased number of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current designed face area into the preset overbreak rate prediction model to obtain the fifth predicted result output by the preset overbreak rate prediction model.

3. The tunneling overbreak rate regulation method according to claim 2, wherein, Until the overbreak rate prediction result is less than the preset control overbreak rate, determining the target number of blast holes, the target charge amount, the target number of perimeter holes, the target number of relief holes, and the target number of cut holes includes: When the fifth predicted result is greater than or equal to the preset control overbreak rate, increase the number of cut holes again, and input the increased number of blast holes, the reduced charge amount, the reduced number of perimeter holes, the increased number of relief holes, the increased number of cut holes again, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current designed face area into the preset overbreak rate prediction model to obtain the sixth predicted result output by the preset overbreak rate prediction model; Until the sixth predicted result is less than the preset control overbreak rate, determine that the target number of blast holes is the increased number of blast holes, determine that the target charge amount is the reduced charge amount, determine that the target number of perimeter holes is the reduced number of perimeter holes, determine that the target number of relief holes is the increased number of relief holes, and determine that the target number of cut holes is the increased number of cut holes again.

4. The tunnel overexcavation rate regulation method according to claim 1, wherein Before inputting the current number of blast holes, the current charge amount, the current number of perimeter holes, the current number of relief holes, the current number of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current designed face area into the preset overbreak rate prediction model, the method further includes: Determine the current number of blast holes, the current charge amount, the current number of perimeter holes, the current number of relief holes, the current number of cut holes, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current designed face area according to the excavation site data in the tunnel blasting construction record; Normalize all the data at the excavation site to obtain the test set data and the training set data, calculate the range of the number of hidden layer neurons, define the range of values of the regularization coefficient, divide the training set data into K folds for cross-validation, and use different activation functions to perform grid search on the regularization coefficient and the number of hidden layer neurons respectively to obtain the regularization coefficient and the number of hidden layer neurons with the optimal coefficient of determination; Import the test set data, perform cross-validation on the test set data, determine the coefficient of determination and the mean absolute square error corresponding to each activation function, and determine the target activation function according to the coefficient of determination and the mean absolute square error corresponding to different activation functions; Use the target activation function to construct the preset overbreak rate prediction model.

5. The tunnel over-excavation rate regulation method according to claim 4, characterized in that, The different activation functions at least include the identity function, the logistic function, the hyperbolic tangent function, and the rectified linear unit; The determining the target activation function according to the coefficient of determination and the mean absolute square error corresponding to different activation functions includes: Determine the activation function with the largest coefficient of determination and the smallest mean absolute square error as the target activation function.

6. The tunneling overbreak rate regulation method according to claim 5, wherein, When the data at the excavation site is the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, the current cut hole degree, the current excavation footage, the current surrounding rock grade, the current blast hole depth, and the current face design area, determine that the target activation function is the rectified linear unit Relu.

7. The tunnel overexcavation rate regulation method according to claim 4, characterized in that, The using different activation functions to perform grid search on the regularization coefficient and the number of hidden layer neurons respectively to obtain the regularization coefficient and the number of hidden layer neurons with the optimal coefficient of determination includes: For any activation function, construct a three-layer BP neural network model including an input layer, a hidden layer, and an output layer based on the test set data. The configuration of the neural network model includes the number of input variables, the number of output variables, the number of hidden layers, the number of samples in the training sample set, the number of folds in K-fold cross-validation, the optimizer, the initial learning rate, the regularization method, and the maximum number of training steps; Use the preset grid search algorithm, the range of the number of hidden layer neurons, and the range of values of the regularization coefficient, combined with K-fold cross-validation, to perform a traversal search on the number of hidden layer neurons and the regularization coefficient of the multi-layer perceptron regression model. During the search process, perform multiple trainings on the data set, record the running time and performance score of each training, and obtain the regularization coefficient and the number of hidden layer neurons with the optimal coefficient of determination.

8. The method for regulating the overexcavation rate of a tunnel according to claim 1, characterized in that Before adjusting the current number of blast holes, the current charge amount, the current perimeter hole degree, the current auxiliary hole degree, and the current cut hole degree according to the adjustment order of the control parameters, the method further includes: Construct a preset random forest model using the input layer feature data of each section before tunnel blasting construction and the overbreak rate of the corresponding section after construction as the construction training set; During the construction process, randomly select samples from the construction training set multiple times to construct decision trees, and randomly select features for splitting at each node to construct multiple decision trees; Calculate the contribution degree of each input layer feature data in the decision tree splitting process, and calculate the average value of the contribution degrees of the input layer feature data in all decision trees to obtain the feature importance parameter result of the input layer feature data; Sort the number of blast holes, charge amount, number of perimeter holes, number of auxiliary holes, and number of cut holes from large to small according to the feature importance parameter result to determine the adjustment order of the control parameters.

9. A device for regulating the overexcavation rate of a tunnel, characterized in that, Including: An input unit, which is used to input the current number of blast holes, current charge amount, current number of perimeter holes, current number of auxiliary holes, current number of cut holes, current excavation footage, current surrounding rock grade, current blast hole depth, and current heading face design area into a preset overbreak rate prediction model to obtain the overbreak rate prediction result output by the preset overbreak rate prediction model; A determination unit, which is used to, when the overbreak rate prediction result is greater than or equal to the preset control overbreak rate, adjust the current number of blast holes, current charge amount, current number of perimeter holes, current number of auxiliary holes, and current number of cut holes according to the adjustment order of the control parameters until the overbreak rate prediction result is less than the preset control overbreak rate, and determine the target number of blast holes, target charge amount, target number of perimeter holes, target number of auxiliary holes, and target number of cut holes; A processing unit, which is used to use the target number of blast holes, target charge amount, target number of perimeter holes, target number of auxiliary holes, target number of cut holes, current excavation footage, current surrounding rock grade, current blast hole depth, and current heading face design area as target control parameters to perform tunnel blasting construction according to the target control parameters; The adjustment order of the control parameters is determined by sorting the number of blast holes, charge amount, number of perimeter holes, number of auxiliary holes, and number of cut holes from large to small according to the feature importance parameter result, and the feature importance parameter result is determined by inputting the number of blast holes, charge amount, number of perimeter holes, number of auxiliary holes, number of cut holes, excavation footage, surrounding rock grade, blast hole depth, and heading face design area into a preset random forest model and output by the preset random forest model; The preset overbreak rate prediction model is determined after being trained according to the sample number of blast holes, sample charge amount, sample number of perimeter holes, sample number of auxiliary holes, sample number of cut holes, sample excavation footage, sample surrounding rock grade, sample blast hole depth, sample heading face design area, and sample overbreak rate prediction result.

Images