Intelligent directional blasting parameter optimization method combined with machine learning

Abstract

The invention relates to a machine learning-combined intelligent directional blasting parameter optimization method, which comprises the following steps of: obtaining drilling data of a to-be-detected tunnel curve section, and performing mechanical analysis on the drilling data to obtain a rock mass mechanical coefficient; blasting design parameters are obtained, the rock mass mechanical coefficient and the blasting design parameters are input into a blasting space-time distribution model, and space-time distribution of blasting energy in the rock mass is obtained; the blasting space-time distribution model is obtained by training an RBF neural network by using a training set; and determining a blasting target range of the to-be-detected tunnel curve section, taking the blasting target range of the blasting energy in the to-be-detected tunnel curve section as an optimization target, carrying out iterative search to determine an optimal blasting design parameter, and carrying out directional blasting based on the optimal blasting design parameter. Accurate directional blasting is achieved, the blasting effect is improved, damage of blasting to surrounding rock mass is reduced, and the safety risk is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering blasting, and in particular to an intelligent directional blasting parameter optimization method combined with machine learning. Background Art

[0002] Blasting technology is widely used in engineering blasting, especially in complex projects such as tunnel excavation, mining, and large-scale infrastructure construction. However, traditional blasting parameter optimization methods face many challenges and limitations, making it difficult to meet the high requirements of blasting accuracy and efficiency in modern engineering.

[0003] First, the mechanical properties of rock mass are complex and diverse. Parameters such as strength, elastic modulus, and Poisson's ratio vary significantly under different geological conditions. Traditional methods for obtaining rock mechanical parameters, such as field and laboratory tests, are often time-consuming and labor-intensive, and fail to accurately reflect the rock's true characteristics in actual blasting environments. Furthermore, the heterogeneity and anisotropy of rock mass complicate parameter acquisition, leading to significant deviations from expected blasting results.

[0004] Secondly, blasting design involves numerous parameters, including row spacing, hole spacing, drill height, hole diameter, plugging length, single-hole charge, and explosive coefficient. These parameters interact with each other, forming a complex parameter system. Traditional optimization methods, such as empirical formulas and trial-and-error methods, struggle to find the optimal combination among these numerous parameters for precise directional blasting. Furthermore, these methods rely heavily on the operator's experience, and different operators may produce different optimization results, lacking consistency and reliability.

[0005] Furthermore, the propagation and distribution of explosive energy within the rock mass during blasting is complex. While traditional simulation methods, such as finite element analysis and empirical formulas, can predict the propagation of explosive energy to a certain extent, they suffer from high computational complexity and oversimplified model assumptions, making it difficult to accurately predict the energy distribution during actual blasting. This not only affects the evaluation and optimization of blasting effects but can also lead to over-excavation, under-excavation, or excessive blasting vibrations, increasing project costs and safety risks.

[0006] Furthermore, the increasing scale and complexity of engineering projects place greater demands on the real-time and dynamic optimization of blasting parameters. Traditional methods struggle to adjust blasting parameters based on real-time monitoring data during project implementation, failing to meet the demands of dynamic optimization.

[0007] In summary, traditional blasting parameter optimization methods have shortcomings in terms of accuracy, efficiency, adaptability, and dynamics, making them difficult to meet the high-precision, high-efficiency, and high-safety requirements of modern engineering blasting. Therefore, how to combine advanced technologies such as machine learning and artificial intelligence to develop an intelligent directional blasting parameter optimization method to improve the accuracy and efficiency of blasting parameter optimization and thus achieve directional blasting is an urgent problem in the field of engineering blasting. Summary of the Invention

[0008] In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide an intelligent directional blasting parameter optimization method combined with machine learning, which realizes precise directional blasting, improves blasting effect, reduces damage to surrounding rock mass caused by blasting, and reduces safety risks.

[0009] To achieve the above object, the present invention provides the following solutions:

[0010] An intelligent directional blasting parameter optimization method combined with machine learning, comprising:

[0011] Obtaining drilling data of the curved section of the tunnel to be measured, performing mechanical analysis on the drilling data, and obtaining a rock mass mechanical coefficient;

[0012] Acquire blasting design parameters, input the rock mass mechanical coefficients and the blasting design parameters into a blasting spatiotemporal distribution model to obtain the spatiotemporal distribution of the explosive energy in the rock mass; the blasting spatiotemporal distribution model is obtained by training an RBF neural network using a training set;

[0013] Determine the blasting target range of the curve section of the tunnel to be measured, take the blasting target range of the explosion energy in the curve section of the tunnel to be measured as the optimization target, iteratively search to determine the optimal blasting design parameters, and perform directional blasting based on the optimal blasting design parameters.

[0014] Optionally, obtaining the rock mass mechanical coefficient includes:

[0015] Acquire drilling data of the tunnel curve section to be measured, wherein the drilling data includes: drilling speed and drilling depth corresponding to each data in the drilling time series;

[0016] Inputting the drilling speed and the drilling depth into a rock strength prediction model to obtain rock strength; the rock strength prediction model is obtained by training a recurrent neural network model using a training set; the training set includes: original drilling data;

[0017] The rock strength is used to determine the uniaxial compressive strength, and the rock mass mechanical coefficient is obtained based on the uniaxial compressive strength.

[0018] Optionally, in the process of training the recurrent neural network model using the training set, the original drilling data is input into the recurrent neural network model for extraction to obtain drilling information:

[0019] c i =f(Uy i +Wc i-1 +a),i∈{1,2,…,n}

[0020] Where c i is the drilling information output by the hidden node, n is the number of elements in the original drilling data; i is the i-th original drilling data; c i-1 is the drilling information contained in the i-1th original drilling data; y i is the input i-th original drilling data; f() is the nonlinear activation function; U is the weight connecting the input layer and the hidden layer; W is the weight connecting the i-th hidden node and the i-1-th hidden node, and a is the bias;

[0021] The drilling information is converted to obtain the original rock strength:

[0022] d j =Softmax(xc i +b),j∈{1,2,…,n}

[0023] Where, d i is the rock strength reference value of the jth borehole output, h i is the drilling information output by the i-th hidden layer, Softmax() is the nonlinear activation function, x is the weight value, and b is the bias.

[0024] Optionally, obtaining the rock mass mechanical coefficient includes:

[0025]

[0026] Among them, F is the rock mass mechanical coefficient and UCS is the uniaxial compressive strength.

[0027] Optionally, the blasting design parameters include: row spacing, hole spacing, drilling height, hole diameter, blocking length, single hole charge and explosive coefficient.

[0028] Optionally, training the RBF neural network using the training set includes:

[0029] The blasting design parameters and rock coefficients of the training data set are used as the input layer of the RBF neural network, and the original spatiotemporal distribution of the training data set is used as the output layer. The RBF neural network is trained to determine the hidden layer structure of the RBF neural network and the weight coefficient between the hidden layer and the output layer, thus obtaining the trained RBF neural network.

[0030] The original spatiotemporal distribution of the explosive energy in the rock mass of the training data set is obtained by simulating the propagation of explosive energy in different rock media by using a numerical simulation method based on blasting design parameters for each blasting point.

[0031] Optionally, determining the optimal blasting design parameters includes:

[0032] Based on the spatiotemporal distribution of the explosion energy in the rock mass, a three-dimensional spatial model is established, and the three-dimensional spatial model is divided into a plurality of grid units;

[0033] For each blasting point, a particle swarm optimization algorithm is used. The blasting target range of the explosion energy in the curve section of the tunnel to be tested is taken as the optimization goal, and the blasting design parameters are used as optimization variables. An optimization model is established, and the optimal blasting parameter combination for each blasting point is determined through iterative search.

[0034] Optionally, during the iteration process, the current blasting design parameters are input into the blasting spatiotemporal distribution model to obtain the spatiotemporal distribution of the explosion energy in each grid unit; the energy distribution of the grid units within the tunnel curve section to be measured is used as an evaluation index of the blasting target range, and is mapped to an optimization target value through a fitness function; based on the optimization target value, the particle swarm algorithm is used to update the parameters of the blasting points to obtain a new parameter combination and enter the next round of iteration; when the number of iterations reaches a preset threshold or the optimization target value converges, the blasting design parameters of each blasting point are output as the optimal blasting parameter combination.

[0035] Optionally, establishing the optimization model includes:

[0036] R=f(b,s,h,d,l,q,k)

[0037]

[0038] Where R is the blasting target range, b is the row spacing, s is the hole spacing, h is the drilling height, d is the hole diameter, l is the blocking length, q is the charge per hole, k is the explosive coefficient, C is a constant related to the explosive type and drilling conditions, and max and min are the maximum and minimum values.

[0039] The beneficial effects of the present invention are:

[0040] By utilizing the recurrent neural network model and the RBF neural network model in machine learning, the present invention can more accurately obtain the rock mechanical coefficients and simulate the spatiotemporal distribution of explosion energy in the rock mass, providing a more precise basis for the optimization of blasting parameters, thereby improving the accuracy of blasting parameter optimization.

[0041] The present invention takes the blasting target range of the explosive energy in the curved section of the tunnel to be tested as the optimization goal and adopts the particle swarm optimization algorithm for iterative search. It can effectively determine the optimal combination of blasting design parameters, achieve precise directional blasting, improve the blasting effect, reduce the damage to the surrounding rock mass caused by blasting, and reduce safety risks.

[0042] The present invention utilizes a machine learning model to automatically train and optimize the process, reducing manual intervention and reliance on experience, improving the efficiency of blasting parameter optimization, and shortening the project cycle.

[0043] The present invention is applicable to a variety of geological conditions and blasting environments. By adjusting the training data set and optimizing the model parameters, it can adapt to different engineering needs and has strong versatility and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. 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 any creative work.

[0045] Figure 1 This is a flow chart of an intelligent directional blasting parameter optimization method combined with machine learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0047] 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.

[0048] like Figure 1As shown, this embodiment discloses an intelligent directional blasting parameter optimization method combined with machine learning, including: obtaining drilling data of a tunnel curve section to be tested, performing mechanical analysis on the drilling data, and obtaining rock mechanical coefficients; obtaining blasting design parameters, inputting the rock mechanical coefficients and the blasting design parameters into a blasting spatiotemporal distribution model, and obtaining the spatiotemporal distribution of explosive energy in the rock mass; the blasting spatiotemporal distribution model is obtained by training an RBF neural network using a training set; determining a blasting target range for the tunnel curve section to be tested, taking the blasting target range of the explosive energy in the tunnel curve section to be tested as the optimization target, iteratively searching to determine the optimal blasting design parameters, and performing directional blasting based on the optimal blasting design parameters.

[0049] Furthermore, obtaining the rock mechanical coefficient includes: obtaining drilling data of the tunnel curve section to be tested, the drilling data including: drilling speed and drilling depth corresponding to each data in the drilling time series; inputting the drilling speed and drilling depth into a rock strength prediction model to obtain rock strength; the rock strength prediction model is obtained by training a recurrent neural network model using a training set; the training set includes: original drilling data; using the rock strength, determining the uniaxial compressive strength, and obtaining the rock mechanical coefficient based on the uniaxial compressive strength.

[0050] Specifically, to obtain the rock mechanical coefficients for the tunnel curve section under test, drilling data for that curve section must first be collected. This drilling data includes the drilling speed and the drilling depth corresponding to each data point in the drilling time series. The drilling speed and drilling depth are then used as input parameters into a rock strength prediction model to determine the rock strength. This rock strength prediction model is developed by training a recurrent neural network model using a training set consisting of the original drilling data. After obtaining the rock strength, the uniaxial compressive strength is further determined based on the rock strength. Finally, based on the determined uniaxial compressive strength, the rock mechanical coefficients are obtained. This process not only involves the precise collection of on-site drilling data but also requires the use of advanced model prediction techniques and in-depth analysis of rock mechanical properties. This provides key rock mechanical parameter support for the design and construction of tunnel projects, ensuring the project's safety and stability.

[0051] Furthermore, in the process of training the recurrent neural network model using the training set, the original drilling data is input into the recurrent neural network model for extraction to obtain drilling information:

[0052] c i =f(Uy i +Wc i-1 +a),i∈{1,2,…,n}

[0053] Where c iis the drilling information output by the hidden node, n is the number of elements in the original drilling data; i is the i-th original drilling data; c i-1 is the drilling information contained in the i-1th original drilling data; y i is the input i-th original drilling data; f() is the nonlinear activation function; U is the weight connecting the input layer and the hidden layer; W is the weight connecting the i-th hidden node and the i-1-th hidden node, and a is the bias;

[0054] Specifically, a Gaussian function is used as a radial basis function, and the orthogonal least squares method is used to select and determine the neurons of the RBF neural network to determine the hidden layer of the RBF neural network: (1) All training set data are used as RBF neurons to obtain the regression vectors processed by the RBF neurons; (2) The influence of each regression vector on the training error is calculated, and the original data corresponding to the regression vector v1 with the greatest influence is selected as the first RBF neuron; (3) The remaining regression vectors v2~vk are orthogonalized with the previously selected regression vector respectively, and from the orthogonalized regression vectors, the original data corresponding to the regression vector with the greatest influence on reducing the training error is selected as a new RBF neuron; (4) Step (3) is repeated k-1 times to obtain k RBF neurons, and the k RBF neurons constitute the hidden layer of the current RBF neural network.

[0055] The LM algorithm is used to determine the weight coefficients between the hidden layer and the output layer: (1) Use random numbers to initialize the parameters between the hidden layer and the output layer of the RBF neural network, and calculate the output value of the average blockiness of the RBF neural network under the current parameters; (2) Use the BP algorithm to obtain the gradient of the output value of the average blockiness with respect to each weight coefficient, and then calculate the Jacobian matrix J; (3) Calculate the error gradient of the current parameter: g = J'residual; where J' is the transpose of the Jacobian matrix, residual is the true value of the average blockiness in the training data set minus the output value of the average blockiness under the current neural network in step (1), and g is the speed of error decrease, that is, the error gradient; (4) Calculate the step size: Δ = (J*J·μI) -1 *g; where Δ is the step size of each iteration parameter update, I is a unit matrix, and μ is a manually set coefficient; (5) Update the weight parameter between the hidden layer and the output layer of the RBF neural network: weight′=weight-Δ; where weight is the weight parameter after the current iteration, and weight is the weight parameter before the current iteration; (6) Repeat steps (2) to (5) until the preset number of iterations is reached.

[0056] Convert the drilling information to obtain the original rock strength:

[0057] d j=Softmax(xc i +b),j∈{1,2,…,n}

[0058] Where, d i is the rock strength reference value of the jth borehole output, h i is the drilling information output by the i-th hidden layer, Softmax() is the nonlinear activation function, x is the weight value, and b is the bias.

[0059] Furthermore, obtaining the rock mass mechanical coefficients includes:

[0060]

[0061] Among them, F is the rock mass mechanical coefficient and UCS is the uniaxial compressive strength.

[0062] Furthermore, the blasting design parameters include: row spacing, hole spacing, drilling height, hole diameter, blocking length, single hole charge and explosive coefficient.

[0063] Specifically, the row spacing is defined as follows: the row spacing refers to the distance between two adjacent rows of blastholes. Function: The size of the row spacing will affect the blasting effect and the degree of rock crushing. Appropriate row spacing can ensure the effective transfer of blasting energy and uniform rock crushing. Selection method: Usually the row spacing is equal to the minimum resistance line W. When parameters such as unit explosive consumption, medium and deep hole density coefficient, charge density and medium and deep hole charge coefficient are constant, the minimum resistance line is proportional to the hole diameter, and can also be calculated using the formula W=K×d, where K is the rock coefficient.

[0064] Definition of Hole Spacing: Hole spacing refers to the vertical distance between two adjacent blastholes in the same row. Function: Hole spacing affects blasting effectiveness and rock fragmentation. A suitable hole spacing ensures effective transfer of blasting energy and uniform rock fragmentation. Selection Method: Hole spacing should be 0.8 to 1.2 times the minimum resistance line, or (0.8 to 1.2)W. It can also be calculated using the formula a = m × W.

[0065] Definition of Drill Height: Drill height refers to the vertical height between the blasthole and the ground. Purpose: Drill height determines the blasting range and the amount of rock crushed. Selection Method: Generally, the height difference between slope steps is determined by design or by unified layers in the construction organization design. If the design does not specify steps or the height difference between steps is significant, the H value is determined based on the layered height difference determined in the construction organization design.

[0066] Aperture Definition: Aperture refers to the diameter of the blasthole. Effect: Aperture size affects the explosive charge and blasting effectiveness. Selection Method: Select the lowest-cost D value based on the performance of the operational drilling rig, the technical terms of the contract, and the permissible range specified in relevant regulations. Ensure that the blasting vibration generated by the single-hole charge is within the permissible range, and ensure blasting effectiveness.

[0067] Definition of Plug Length: The plug length is the length of the blasthole remaining after charging. Purpose: The plug length prevents blasting gas leakage and improves blasting effectiveness. Selection Method: Generally, it should be 20-40 times the hole diameter, and can be adjusted appropriately based on lithology.

[0068] Definition of Single-Hole Charge: The single-hole charge refers to the amount of explosives loaded into each blasthole. Purpose: The single-hole charge determines the blasting energy of each blasthole. Selection method: Q = Kuw / L (first row) or Q = Kabh (second row and later), where K is the unit explosive consumption, u is the hole area, w is the bottom resistance line, L is the hole depth, a is the hole spacing, b is the row spacing, and h is the step height.

[0069] Explosives coefficient definition: The explosives coefficient refers to the specific amount of explosives consumed. Purpose: The explosives coefficient determines the amount of explosives required for the entire blasting project. Selection method: Determine based on the rock structure and previous blasting experience in the area.

[0070] Furthermore, using the training set to train the RBF neural network includes: using the blasting design parameters and rock coefficients of the training data set as the input layer of the RBF neural network, and using the original spatiotemporal distribution of the training data set as the output layer, training the RBF neural network to determine the hidden layer structure of the RBF neural network, and the weight coefficient between the hidden layer and the output layer, thereby obtaining a trained RBF neural network; wherein, the original spatiotemporal distribution of the explosive energy of the training data set in the rock mass is represented by the rock characteristics represented by the rock coefficient, and for each blasting point, a numerical simulation method is performed through the blasting design parameters to simulate the propagation of the explosive energy in different rock media.

[0071] Specifically, the radial basis function (RBF) neural network training process is as follows: the blasting design parameters and rock coefficients in the training dataset serve as the input layer of the RBF neural network, while the original spatiotemporal distribution of the training dataset serves as the output layer. This training process determines the hidden layer structure of the RBF neural network and the weight coefficients between the hidden and output layers, resulting in a trained RBF neural network. It is important to note that the original spatiotemporal distribution of the explosive energy in the rock mass in the training dataset is determined by the rock properties reflected by the rock coefficients. For each blasting point, numerical simulation methods are used to simulate the propagation of explosive energy in different rock media using the blasting design parameters, thereby obtaining the corresponding spatiotemporal distribution data. This training process not only requires the precise input of the blasting design parameters and rock coefficients, but also requires complex numerical simulations to obtain the spatiotemporal distribution of explosive energy. This provides an accurate data foundation for RBF neural network training, ensuring that the neural network can effectively learn and predict the propagation of explosive energy in the rock mass, providing strong technical support for subsequent blasting engineering design and rock stability analysis.

[0072] Furthermore, determining the optimal blasting design parameters includes: establishing a three-dimensional spatial model based on the spatiotemporal distribution of explosive energy in the rock mass, and dividing the three-dimensional spatial model into a plurality of grid cells; using a particle swarm optimization algorithm for each blasting point, taking the blasting target range of the explosive energy in the tested tunnel curve section as the optimization target, and using the blasting design parameters as optimization variables, establishing an optimization model, and determining the optimal blasting parameter combination for each blasting point through iterative search. During the iterative process, the current blasting design parameters are input into the blasting spatiotemporal distribution model to obtain the spatiotemporal distribution of the explosive energy in each grid cell; the energy distribution of the grid cells in the tested tunnel curve section is used as an evaluation index of the blasting target range, and is mapped to an optimization target value through a fitness function; based on the optimization target value, the particle swarm algorithm is used to update the parameters of the blasting point to obtain a new parameter combination, and enter the next round of iteration; when the number of iterations reaches a preset threshold or the optimization target value converges, the blasting design parameters of each blasting point are output as the optimal blasting parameter combination.

[0073] Specifically, a three-dimensional spatial model is established: Based on the spatiotemporal distribution of blast energy in the rock mass, a three-dimensional spatial model is constructed and divided into multiple grid cells. A particle swarm optimization algorithm is employed: For each blasting point, the blasting target range of the tunnel curve section to be tested is used as the optimization objective, and the blasting design parameters are set as optimization variables to establish the optimization model. The optimal blasting parameter combination is then iteratively searched: During the iterative process, the current blasting design parameters are first input into the blasting spatiotemporal distribution model to obtain the spatiotemporal distribution of blast energy in each grid cell. The energy distribution of the grid cells within the tunnel curve section to be tested is then used as an evaluation metric for the blasting target range and converted into an optimization target value using a fitness function. Based on the optimization target value, the particle swarm optimization algorithm is then used to update the parameters of the blasting points, resulting in a new set of parameter combinations for the next iteration. When the number of iterations reaches a pre-set threshold, or the optimization target value converges, the blasting design parameters for each blasting point are output as the final optimal blasting parameter combination. This process fully combines blasting design parameters, rock properties, and the spatiotemporal distribution of blast energy. Through iterative search using the particle swarm optimization algorithm, it can effectively determine the optimal blasting design parameters, providing a scientific and reasonable parameter basis for blasting projects in curved tunnel sections. This helps improve blasting effectiveness, reduce project costs, and ensure construction safety and stability.

[0074] Furthermore, establishing an optimization model includes:

[0075] R=f(b,s,h,d,l,q,k)

[0076]

[0077] Where R is the blasting target range, b is the row spacing, s is the hole spacing, h is the drilling height, d is the hole diameter, l is the blocking length, q is the charge per hole, k is the explosive coefficient, C is a constant related to the explosive type and drilling conditions, and max and min are the maximum and minimum values.

[0078] In actual blasting projects, the minimum and maximum values of various parameters are usually determined by the specific conditions of the project, safety requirements, economic factors, and the guidance of blasting theory: Minimum value bmin: The minimum value of the row spacing is usually determined by the minimum size of the blasting area and the blasting safety distance. Too small a row spacing may lead to unsatisfactory blasting results or safety issues. Maximum value bmax: The maximum value of the row spacing is determined by the maximum size of the blasting area and the blasting effect requirements. Excessive row spacing may result in uneven or insufficient blasting results. Minimum value smin: The minimum value of the hole spacing is usually determined by the minimum drill hole spacing and the blasting safety distance of the drilling equipment. Maximum value smax: The maximum value of the hole spacing is determined by the size of the blasting area and the blasting effect requirements. Minimum value hmin: The minimum value of the drill hole height is usually determined by the minimum depth of the blasting area and the blasting effect requirements. Maximum value hmax: The maximum value of the drill hole height is determined by the maximum depth of the blasting area and the limitations of the drilling equipment. Minimum value dmin: The minimum hole diameter is typically determined by the minimum drilling diameter of the drilling equipment and the required blasting effect. Maximum value dmax: The maximum hole diameter is determined by the maximum drilling diameter of the drilling equipment and the size of the blasting area. Minimum value lmin: The minimum blockage length is typically determined by blasting safety and effectiveness requirements. Excessively small blockage lengths may result in explosive energy loss and safety issues. Maximum value lmax: The maximum blockage length is determined by the size of the blasting area and the required blasting effect. Minimum value qmin: The minimum single-hole charge is typically determined by blasting effectiveness and safety requirements. Excessively small charge amounts may result in unsatisfactory blasting results. Maximum value qmax: The maximum single-hole charge is determined by safety requirements and explosive availability. Excessive charge amounts may lead to safety issues and waste of resources. Minimum value kmin: The minimum explosive coefficient is typically determined by the type of explosive and the required blasting effect. Maximum value kmax: The maximum explosive coefficient is determined by the type of explosive and safety requirements. When determining the minimum and maximum values for these parameters, it's important to consider the specific project conditions, safety requirements, economic factors, and the guidance of blasting theory. Typically, these values are determined based on engineering experience, historical data, blasting theory, and relevant specifications. In practice, these values may need to be adjusted and optimized based on specific circumstances.

[0079] Furthermore, the relationship between the blasting target range and the blasting design parameters includes:

[0080]

[0081] Where A is a constant related to geological conditions and blasting environment.

[0082] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. An intelligent directional blasting parameter optimization method combined with machine learning, characterized in that: include: Obtaining drilling data of the curved section of the tunnel to be measured, performing mechanical analysis on the drilling data, and obtaining a rock mass mechanical coefficient; Acquire blasting design parameters, input the rock mass mechanical coefficient and the blasting design parameters into a blasting spatiotemporal distribution model, and acquire the spatiotemporal distribution of the explosion energy in the rock mass; The blasting spatiotemporal distribution model is obtained by training the RBF neural network using the training set; Determine the blasting target range of the curve section of the tunnel to be measured, take the blasting target range of the explosion energy in the curve section of the tunnel to be measured as the optimization target, iteratively search to determine the optimal blasting design parameters, and perform directional blasting based on the optimal blasting design parameters.

2. The intelligent directional blasting parameter optimization method combined with machine learning according to claim 1 is characterized in that: Obtaining the rock mass mechanical coefficient includes: Acquire drilling data of the tunnel curve section to be measured, wherein the drilling data includes: drilling speed and drilling depth corresponding to each data in the drilling time series; Inputting the drilling speed and the drilling depth into a rock strength prediction model to obtain rock strength; the rock strength prediction model is obtained by training a recurrent neural network model using a training set; the training set includes: original drilling data; The rock strength is used to determine the uniaxial compressive strength, and the rock mass mechanical coefficient is obtained based on the uniaxial compressive strength.

3. The intelligent directional blasting parameter optimization method combined with machine learning according to claim 2 is characterized in that: In the process of training the recurrent neural network model using the training set, the original drilling data is input into the recurrent neural network model for extraction to obtain drilling information: c i =f(Uy i +Wc i-1 +a),i∈{1,2,…,n} Where c i is the drilling information output by the hidden node, n is the number of elements in the original drilling data; i is the i-th original drilling data; c i-1 is the drilling information contained in the i-1th original drilling data; y i is the input i-th original drilling data; f() is the nonlinear activation function; U is the weight connecting the input layer and the hidden layer; W is the weight connecting the i-th hidden node and the i-1-th hidden node, and a is the bias; The drilling information is converted to obtain the original rock strength: <h2 style=";text-align:left;direction:ltr">d<h2 style=";text-align:left;direction:ltr"> j <h2 style=";text-align:left;direction:ltr"> =Softmax(xc<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> +b),j∈{1,2,…,n} Where, d i is the rock strength reference value of the jth borehole output, h i is the drilling information output by the i-th hidden layer, Softmax() is the nonlinear activation function, x is the weight value, and b is the bias.

4. The intelligent directional blasting parameter optimization method combined with machine learning according to claim 2 is characterized in that: Obtaining the rock mass mechanical coefficient includes: Among them, F is the rock mass mechanical coefficient and UCS is the uniaxial compressive strength.

5. The intelligent directional blasting parameter optimization method combined with machine learning according to claim 1 is characterized in that: The blasting design parameters include: row spacing, hole spacing, drilling height, hole diameter, blocking length, single hole charge and explosive coefficient.

6. The intelligent directional blasting parameter optimization method combined with machine learning according to claim 1 is characterized in that: Training the RBF neural network using the training set includes: The blasting design parameters and rock coefficients of the training data set are used as the input layer of the RBF neural network, and the original spatiotemporal distribution of the training data set is used as the output layer. The RBF neural network is trained to determine the hidden layer structure of the RBF neural network and the weight coefficient between the hidden layer and the output layer, thus obtaining the trained RBF neural network. The original spatiotemporal distribution of the explosive energy in the rock mass of the training data set is obtained by simulating the propagation of explosive energy in different rock media by using a numerical simulation method based on blasting design parameters for each blasting point.

7. The intelligent directional blasting parameter optimization method combined with machine learning according to claim 1 is characterized in that: Determining the optimal blasting design parameters includes: Based on the spatiotemporal distribution of the explosion energy in the rock mass, a three-dimensional spatial model is established, and the three-dimensional spatial model is divided into a plurality of grid units; For each blasting point, a particle swarm optimization algorithm is used. The blasting target range of the explosion energy in the curve section of the tunnel to be tested is taken as the optimization goal, and the blasting design parameters are used as optimization variables. An optimization model is established, and the optimal blasting parameter combination for each blasting point is determined through iterative search.

8. The intelligent directional blasting parameter optimization method combined with machine learning according to claim 7 is characterized in that: During the iteration process, the current blasting design parameters are input into the blasting spatiotemporal distribution model to obtain the spatiotemporal distribution of the explosion energy in each grid cell. The energy distribution of the grid cells within the measured tunnel curve section is used as an evaluation index of the blasting target range and mapped to an optimization target value through a fitness function. Based on the optimization target value, the particle swarm algorithm is used to update the parameters of the blasting points to obtain a new parameter combination and enter the next round of iteration. When the number of iterations reaches a preset threshold or the optimization target value converges, the blasting design parameters of each blasting point are output as the optimal blasting parameter combination.

9. The intelligent directional blasting parameter optimization method combined with machine learning according to claim 7, characterized in that: Establishing the optimization model includes: R=f(b,s,h,d,l,q,k) Where R is the blasting target range, b is the row spacing, s is the hole spacing, h is the drilling height, d is the hole diameter, l is the blocking length, q is the charge per hole, k is the explosive coefficient, C is a constant related to the explosive type and drilling conditions, and max and min are the maximum and minimum values.

Images